Академический Документы
Профессиональный Документы
Культура Документы
line. In the two decades after the Great Depression the leaders
held their relative positions as the 45
line.
20
20
The volatility in these series derives not from aggregate stock-market volatility, but from the volatility of
individual rm valuations. The large dip and subsequent recovery in both series in 19992001, for example,
1210 B. Jovanovic and P.L. Rousseau
Both gures show, however, that the lines are at or falling during the Electricity and
IT periods, so that replacement at these times was high. This is best seen in Figure 22.
3.5.2. The age of rms at their IPO
According to the third innovation-spawning characteristic, when a GPT arrives it
gives rise to new projects that are unusually protable. When such projects arrive, rms
will be more impatient to implement them. When it is new rms that come upon such
projects (rather than incumbents), they will feel the pressure to list sooner. This argu-
ment is developed and tested in Jovanovic and Rousseau (2001). We argue there that the
Electricity- and IT-era rms entered the stock market sooner because the technologies
that they brought in were too productive to be kept out of the market for very long.
Figure 23 shows HP-ltered average waiting times from founding, rst product or
process innovation, and incorporation to exchange listing based upon individual com-
pany histories and our backward extension of the CRSP database.
21
The vertical dis-
tance between the solid and dotted lines shows that rms often have their rst innovation
soon after founding, but that it then takes years, even decades, to list on a stock ex-
change.
22
We interpret this delay as a period during which the rm and possibly its
lenders learn about what the rms optimal investment should be. But when the tech-
nology is highly innovative, the incentive to wait is reduced and the rm lists earlier,
which is what the evidence shows.
Table 4 lists the rst product or process innovation for some of the better-known
companies, along with their dates of founding, incorporation, and stock exchange list-
ing. It also includes the share of total market capitalization that can be attributed to each
rms common stock at the end of 2003. The rms appearing in the table separate into
roughly 3 groups: those based upon electricity and internal combustion, those based
upon chemicals and pharmaceuticals, and those based upon the computer and Internet.
Let us consider a few of the entries more closely:
comes from Microsofts enormous price appreciation in 1999, when it was worth more than 5 percent of GDP
on its own, and its rapid decline in 2000, which transferred the full 5 percent share to GE. The two rms split
the 5 percent share in 2001.
21
Listing years after 1925 are those for which rms enter CRSP. For 18901924, they are years in which
prices rst appear in the NYSE listings of The Annalist, Bradstreets, The Commercial and Financial Chron-
icle or The New York Times. The 6,632 incorporation dates used to construct Figure 23 are from Moodys
Industrial Manual [Moodys Investors Service (1920, 1928, 1955, 1980)], Standard and Poors Stock Market
Encyclopedia [Standard and Poors Corporation (1981, 1988, 2000)] and various editions of Standard and
Poors Stock Reports [Standard and Poors Corporation (19712003)]. The 4,221 foundings are from Dun and
Bradstreets Million Dollar Directory [Dun and Bradstreet, Inc. (2003)], Moodys, Kelley (1954), and indi-
vidual company web sites. The 482 rst innovations were obtained by reading company histories in Hoovers
Online [Hoovers, Inc. (2000)] and company web sites. We linearly interpolate the series between missing
points before applying the HP-lter to get the time series in Figure 23.
22
Figure 23 includes several years in the 1970s and early 1980s for which it appears that the average time
from rst innovation to listing exceeds that from founding to listing. This is a result of differences in the
sample sizes used to construct each line.
Ch. 18: General Purpose Technologies 1211
Figure 23. Waiting times to exchange listing, 18902003.
Electricity/Internal Combustion Engine Two of largest companies in the United
States today are General Electric (GE) and AT&T. Founded in 1878, GE accounted
for 2.1 percent of total stock market value at the end of 2003, and had already es-
tablished a share of over 2 percent by 1910. AT&T, founded in 1885, contributed
4.6 percent to total market value by 1928, and more than 8.5 percent at the time
of its forced breakup in 1984. Both were early entrants of the Electricity era. GEs
founding was based upon the invention of the incandescent light bulb in 1879,
while AT&T established a long-distance telephone line from New York to Chicago
in 1892 to make use of Bells 1876 invention of the telephone. Both technologies
represented quantum leaps in the modernization of industry and communications,
and both rms brought these technologies to the NYSE about 15 years after found-
ing. General Motors (GM) was an early entrant to the automobile industry, listing
on the NYSE in 1917 nine years after its founding. By 1931 it accounted for
more than 4 percent of stock market value, and its share would hover between
4 and 6.5 percent until 1965, when it began to decline gradually to its share in 2003
of only 0.2 percent. These examples suggest that many of the leading entrants at
the turn of the 20th century created lasting market value. Further, the ideas that
sparked their emergence were brought to market relatively quickly.
Chemicals/Pharmaceuticals Procter and Gamble (P&G), BristolMyers Squibb
and Pzer are both leaders in their respective industries, but took much longer to
list on the NYSE than the Electrication-era rms. In fact, P&G and Pzer were
established before 1850, and thus predate all of them. Despite P&Gs early start
and the creation of the Ivory soap brand in 1879, it was not until 1932 that the
company took its place among the largest U.S. rms by exploiting advances in
radio transmission to sponsor the rst soap opera. Pzers dening moment came
when it developed a process for mass-producing the breakthrough drug penicillin
1212 B. Jovanovic and P.L. Rousseau
Table 4
Key dates in selected company histories
Company name Founding
date
1st major product or
process innovation
Incorporation
date
Listing
date
% of stock market
in 2003
General Electric 1878 1880 1892 1892 2.09
AT&T 1885 1892 1885 1901 0.11
Detroit Edison 1886 1904 1903 1909 0.04
General Motors 1908 1912 1908 1917 0.20
Coca Cola 1886 1893 1919 1919 0.83
Pacic Gas & Electic 1879 1879 1905 1919 0.08
Burroughs/Unisys 1886 1886 1886 1924 0.03
Caterpillar 1869 1904 1925 1929 0.19
KimberlyClark 1872 1914 1880 1929 0.20
Procter & Gamble 1837 1879 1890 1929 0.87
BristolMyers Squibb 1887 1903 1887 1933 0.37
Boeing 1916 1917 1916 1934 0.23
Pzer 1849 1944 1900 1944 1.81
Merck 1891 1944 1934 1946 0.69
Disney 1923 1929 1940 1957 0.32
HewlettPackard 1938 1938 1947 1961 0.47
McDonalds 1948 1955 1965 1966 0.21
Intel 1968 1971 1969 1972 1.40
Microsoft 1975 1980 1981 1986 1.99
America Online 1985 1988 1985 1992 0.52
Amazon 1994 1995 1994 1997 0.14
E-Bay 1995 1995 1996 1998 0.28
Source: Data from Hoovers Online, Kelley (1954), and company web sites.
Note. The rst major products or innovations for the rms listed in the table are: GE 1880, Edison patents
incandescent light bulb; AT&T 1892, completes phone line from New York to Chicago; DTE 1904, in-
creases Detroits electric capacity six-fold with new facilities; GM 1912, electric self-starter; Coca Cola 1893,
patents soft-drink formula; PG&E 1879, rst electric utility; Burroughs/Unisys 1886, rst adding machine;
CAT 1904, gas driven tractor; KimberlyClark 1914, celu-cotton, a cotton substitute used in WWI; P&G
1879, Ivory soap; BristolMyers Squibb 1903, Sal Hepatica, a laxative mineral salt; Boeing 1917, designs
Model C seaplane; Pzer 1944, deep tank fermentation to mass produce penicillin; Merck 1944, cortisone
(rst steroid); Disney 1929, cartoon with soundtrack; HP 1938, audio oscillator; McDonalds 1955, fast food
franchising begins; Intel 1971, 4004 microprocessor (8088 microprocessor in 1978); Microsoft 1980, devel-
ops DOS; AOL 1988, PC-Link; Amazon 1995, rst online bookstore; E-Bay 1995, rst online auction
house.
during World War II, and the good reputation that the rm earned at that time
later helped it to become the main producer of the Salk and Sabin polio vaccines.
In Pzers case, like that of P&G, the companys management and culture had
been in place for some time when a new technology (in Pzers case antibiotics)
presented a great opportunity.
Computer/IT Firms at the core of the recent IT revolution, such as Intel, Mi-
crosoft and Amazon, came to market shortly after founding. Intel listed in 1972,
only four years after starting up, and accounted for 1.4 percent of total stock mar-
Ch. 18: General Purpose Technologies 1213
ket value at the end of 2003. Microsoft took eleven years to go public. Conceived
in an Albuquerque hotel room by Bill Gates in 1975, the company, with its new
disk operating system (MS-DOS), was perhaps ahead of its time, but later joined
the ranks of todays corporate giants with the proliferation of the PC. In 1998, Mi-
crosoft accounted for more than 2.5 percent of the stock market, but this share fell
to 1.5 percent over the next two years in the midst of antitrust action. By the end
of 2003 its share had recovered somewhat to nearly 2 percent of the stock market.
Amazon caught the Internet wave from the outset to become the worlds rst on-
line bookstore, going public in 1997 only three years after its founding. As the
complexities of integrating goods distribution with an Internet front-end came into
sharper focus over the ensuing years, however, and as competition among Internet
retailers continued to grow, Amazons market capitalization by 2003 had fallen to
0.14 percent of total stock market value.
These rms, as well as the others listed in Table 4, are ones that brought new tech-
nologies into the stock market and accounted for more than 13 percent of its value at the
close of 2003. The rms themselves also seem to have entered the stock market sooner
during the Electricity and IT eras, at opposite ends of the 20th century, than rms based
on mid-century technologies.
When rms gather less information before investing, the investments that they under-
take will be riskier. One may conjecture that if new entrants waited less before investing
during the GPT eras, then incumbents also undertook projects earlier than they would
have normally. In these cases, the resulting investments would be riskier than if more
time were allowed to plan them. Moreover, the newness of the GPT would add further
risk. On all these grounds, we would expect interest rate differentials on the average
investment to be higher in the GPT eras.
Figure 24, which shows the spread between interest rates on riskier and safe invest-
ments since 1885, shows that this has been for the most part the case.
23
It is important to
note that we formed the series in Figure 24 by joining three different spreads together,
and that the safe asset is a long-term U.S. government bond before 1920 and a short-
term U.S. Treasury bill thereafter, yet the uctuations in this series should still reect
risk perceptions reasonably well, at least to the extent that term premia rather than risk-
iness are the main factors that lead to yield differentials among the various government
securities.
23
In Figure 24, we use the spread between the interest rates on Baa-rated corporate bonds (from Moodys
Investors Service) and three-month T-bills [from the FRED database of Federal Reserve Bank of St. Louis
(2004) for 19342003 and the Board of Governors of the Federal Reserve System (1976) for 19201934]
for the period from 1920 to the present. For 19001920, we join the spread between the interest rate on
prime commercial paper with 6090 days until maturity [Homer and Sylla (1991, table 49, p. 358)] and the
redemption yields on the U.S. government consol 2s of 1930 [Homer and Sylla (1991, table 46, p. 343)] with
the Baa T-bill spread. Finally, for 18851899, we join the spread between the commercial paper rate [Homer
and Sylla (1991, table 44, p. 320)] and the redemption yields on U.S. government refunding 4s of 1907 [Homer
and Sylla (1991, table 43, p. 316)] with the previous result.
1214 B. Jovanovic and P.L. Rousseau
Figure 24. Nominal interest rate spreads between riskier and safer bonds, 18852003.
During the Electrication period, spreads rose between 1894 and 1907, which is when
uncertainty about the usefulness and possibilities for adoption of the new technology
was greatest. Spreads fell after that as the future of Electricity became clearer. In the
IT era, spreads have a generally-upward trend throughout, though they did fall for a
while in the late 1990s. This may well reect the lag in the widespread adoption of IT.
The spreads sharp rise in 1930 and very slow decline over the next 15 years probably
has to do with the macroeconomic instability induced by events prior to and during
the Great Depression, and then the heavy borrowing by the U.S. government to nance
World War II, which raised rates on T-bills.
Another measure of risk perceptions can be obtained from the distribution of rat-
ings for issues of new corporate bonds. Figure 25 uses data from Hickman (1958,
pp. 153154) and Atkinson (1967, p. 97) for the period from 19081965 to show four-
year averages, starting at the dates shown on the horizontal axis, of the percent of the
total par value of rated new corporate bond issues that received a Moodys rating of
single-A or lower and Ba or lower. In other words, the solid line excludes the highest
rated bonds (i.e., classes Aaa and Aa), but includes some investment grade bonds (i.e.,
A and Baa) along with the sub-investment grades (i.e., Ba and lower). The dashed line
includes only the sub-investment grades.
The dashed line in Figure 25 indicates that subinvestment grade bonds made up a
larger part of the value of total rated new issues during the Electrication era than af-
ter the start of the Great Depression, and though these data end in 1965, we note that
subinvestment grade issues began to rise again only on the eve of the IT revolution
in the mid-1960s. The solid line shows that issues of bonds not receiving the highest
Moodys ratings actually rose during the latter part of the Electrication era, peaking
in the 19241927 period, which was when a host of Electricity-related innovations and
appliances were being brought to market. This does not imply an increase in junk-bond
Ch. 18: General Purpose Technologies 1215
Figure 25. Percent of rated corporate bond offerings with Moodys ratings of A or lower and Ba or lower,
four-year averages, 19081965.
issuance at this time, but rather is consistent with the view that investors recognized the
risks involved with large-scale use of the new technology and were a bit more cautious
about overpaying for debt securities associated with it.
3.5.3. The stock market performance of the young vs. old after entry
Young rms are smaller. If creative destruction does indeed mean that old rms give
way to young rms, then we should see signs of it in Figure 26, which depicts the rela-
tive appreciation of the total market value of small versus large rms since 1885.
24
We
dene small rms as those in the lower quintile of CRSP, and large rms as those
in the upper quintile. The regression line in Figure 26 (with t -statistics in parenthe-
ses) shows small rms outperforming large ones in the long run and an annual growth
premium of about 7.5 percent. But the two GPT eras do not show a faster rise in rela-
tive appreciations than other times, and this is puzzling. Surprisingly, recessions do not
seem to hurt the long-term prospects of small rms: The relative index rises in 10 of the
23 NBER recessions.
The two periods that we wish to focus on are 19291931 and the early 1970s. In
both periods, the small-capitalization rms lost out relative to the large-capitalization
ones. The rst period comes at the end of the Electrication era and the relative decline
of smaller rms is what one would have expected. But the early 1970s come at the
beginning of a new GPT, and small rms should have outperformed the large rms at
24
Being a total value index, this differs from the relative stock price index that is plotted in Figure 8 of Hobijn
and Jovanovic (2001). For the post-1925 period, in which they overlap, the qualitative behavior of the two
series is essentially the same.
1216 B. Jovanovic and P.L. Rousseau
Figure 26. The relative capital appreciations of small vs. large rms, 18852001.
that time. Yet the opposite happened. It is only after 1974 that the small-capitalization
rms start to perform better.
Regression evidence on age and stock market performance. If the GPT is brought in
by young rms, then the capital loss imposed by the GPTs arrival should fall more
heavily on old rms. To test this using data on individual rms, let
A
i
= age since listing of rm i in 1970;
S
i
= share (in rm is sector) of IT capital in the capital stock in 2001.
This measures a rms exposure to the impact of the new technology within its sector.
We use the change in a rms stock price over intervals that start in 1971 and end
in 1975, 1980, 1985, 1990 and 1995 as measures of expected performance. These should
reect the markets assessment of how well the rm will handle the consequences of
the GPT. The regressions take the form
ln
P
i,1975
P
i,1970
= c
0
+ c
1
A
i
+ c
2
S
i
c
3
A
i
S
i
.
We summarize the rm-level results in Table 5.
The interaction between the rms age (A) and its exposure to the new technology (S)
is negative and signicant only when the period during which we measure price ap-
preciation extends to 1990 and 1995. We would have expected this coefcient to be
negative always, since older rms in sectors where IT would become important would
be less able to adjust to the new technology than newer rms. The interaction term has
a positive coefcient for the 19711975, 19711980 and 19711985 periods, but it is
Ch. 18: General Purpose Technologies 1217
Table 5
Age and stock market performance
Dependent variable: ln(P
t +i
/P
t
)
19711975 19711980 19711985 19711990 19711995
constant 0.737 0.143 0.152 0.057 0.577
(24.3) (2.96) (2.58) (0.59) (6.06)
A 0.007 0.001 0.001 0.003 0.002
(6.40) (0.46) (0.55) (0.97) (0.51)
S 3.497 2.266 1.035 0.602 2.719
(7.60) (3.37) (1.20) (0.46) (1.88)
A S 0.047 0.043 0.016 0.122 0.106
(2.22) (1.14) (0.39) (2.09) (1.76)
R
2
0.089 0.009 0.003 0.006 0.012
N 2218 1814 1367 981 843
Note. The table presents coefcient estimates for the subperiods included in the column headings with
t -statistics in parentheses. The R
2
and number of observations (N) for each regression appear in the nal
two rows.
statistically signicant only for the 19711975 period. It thus seems that IT rms took a
long time to realize gains in the market after the technologys arrival. There are not very
many rms with continuous price data prior to 1900, but we have enough observations
to attempt the same regression for the Electrication era. In this case, we got
ln
P
i,1899
P
i,1894
= 2.111
(1.09)
0.129
(0.46)
A
i
2.307
(0.88)
S
i
+ 0.213
(0.55)
A
i
S
i
,
with t -statistics in parentheses and R
2
= 0.015, N = 56. In this very small sample, we
do not see a direct effect of age on capital depreciation as Electrication got underway,
and the interaction term is not statistically signicant.
3.6. Consumption, interest rates, and the trade decit
If it is unanticipated, the arrival of a GPT is good news for the consumer because it
brings about an increase in wealth. How quickly wealth is perceived to rise depends on
how quickly the public realizes the GPTs potential for raising output. The rise in wealth
would raise desired consumption. But to implement the GPT rms would also need to
increase their investment. Therefore aggregate demand would rise, and in a small open
economy this would lead to a trade decit. In a closed economy, on the other hand, since
income does not immediately rise, the rise in aggregate demand would cause the rate of
interest to rise so that the rise in aggregate demand would be postponed.
How much consumption rises depends on two factors. The rst is the GPTs perva-
siveness worldwide if the entire world is equally affected then consumption could not
1218 B. Jovanovic and P.L. Rousseau
rise right away and the main effects would be transmitted though the rate of interest.
The second is the openness of the U.S. economy. Even if, say, the United States were
the only country affected by the GPT, the rise in consumption would be related to how
easily capital could ow in.
In these respects, the IT episode differs from the Electrication episode in several
important respects. Capital inows into the United States simply were not in the cards
during a large part of the Electrication episode. World War I exhausted the European
nations and the United States could not borrow from the rest of the world to nance its
electrication-led expansion it was instead a creditor during this period. Moreover,
even if the war had not taken place, it is not clear whether the United States could
have borrowed much from the rest of the world because Britain, Germany, France, and
several other countries were undergoing the same process Electrication was more
synchronized across the developed world than IT has so far been.
In sum, we would expect the United States to have behaved more like a closed econ-
omy during the Electrication era and more like a small open economy during the
IT era. Specically, we would expect to see
(1) a larger rise in the trade decit during the IT era than during the Electrica-
tion era,
(2) a smaller rise in consumption during the Electrication era then during the IT era,
(3) a larger rise in the rate of interest during the Electrication era.
3.6.1. The trade decit
Figure 27, which plots the trade decit as a percentage of GNP since 1790 along with
an HP trend, shows sharply-rising trade decits at the start of the IT revolution, though
not in the early years of Electrication.
25
The trade decit indeed opens up fairly dra-
matically during the IT era, whereas during the Electrication era we see a surplus. As
we mentioned, this surplus was driven by the various Colonial wars that took place at
the turn of the century and, of course, by World War I.
3.6.2. The consumptionincome ratio
We expected to see a smaller rise in consumption during the Electrication era than
during the IT era, and after we adjust for the downward long-run trend, this is indeed
what has happened. Private consumption rises gradually during each GPT era, and this
is set against a long-run secular trend for private consumption that is negative. Figure 28
shows the ratio of consumption to GDP since 1790.
26
As our GPT hypotheses would
25
GDP and total imports and exports of goods and services are from the U.S. Bureau of Economic Analysis
(2004) for 19292003. For 17901920, imports and exports are from U.S. Bureau of the Census (1975, series
U-8 and U-1, p. 864, respectively), and the GDP series are from Kendrick (1961) and Berry (1988).
26
The series for consumption and GDP are from the U.S. Bureau of Economic Analysis (2004) for
19292003, Kendrick (1961, table A-IIb, cols. 4 and 11, pp. 296297) for 18891929, and Berry (1988,
Ch. 18: General Purpose Technologies 1219
Figure 27. The trade decit as a percent of GDP, 17902003.
Figure 28. The ratio of consumption to income, 17902003.
suggest, the arrival of Electricity in 1890 seems to mark the end of a long-term decline
in the ratio that been underway for a century. And though the level of the series falls
during the Great Depression and World War II, never to return to its pre-1930 levels,
consumption takes another sharp upward turn near the start of the IT revolution and
continues to rise.
table 9, pp. 2526) for 17901889. The BEA gures are for personal consumption, but the Kendrick and
Berry gures include the government sector as well. Since consumption in the government sector was much
smaller prior to World War I, we suspect that the downward trend in the 19th century is a result of changing
private consumption patterns rather than a reduction in the government sectors consumption.
1220 B. Jovanovic and P.L. Rousseau
3.6.3. Interest rates
We expected a larger rise in the rate of interest during the Electrication era than during
the IT era. Relative to HP trends, the evidence is not favorable. Figure 29 shows that
ex-post real interest rates were about the same during the two GPT eras, and much
lower in the middle 40 unshaded years of the 20th century.
27
The dashed line is the HP
detrended series. The averages are presented in Table 6. We note that the ex-post rate
is quite high in the rst era, before 1894. If the arrival of electricity and its impact was
foreseen prior to 1894, interest rates would have risen earlier, but this probably does
not explain why they were so high then. More likely, the pre-1894 era reects a lack of
nancial development: The stock market was small then, and the nancial market not
as deep. This may have given rise to an overall negative trend in interest rates over the
134-year period as a whole.
Figure 29. The ex-post real interest rate on commercial paper, 18702003.
Table 6
Era Ex-post real interest rate
18701893 7.78
18941930 2.61
19311970 0.16
19712003 2.75
27
Commercial paper rates are annual averages from the FRED database for 19342003 and from Homer
and Sylla (1991) for earlier years. We compute the ex-post return by subtracting ination as computed by
the growth of the implicit price deator for GNP from the U.S. Bureau of Economic Analysis (2004) for
19292003 and Berry (1988) for earlier years.
Ch. 18: General Purpose Technologies 1221
4. Conclusion
Technological invention is uneven, and comes in bursts; that much has for a long time
been clear to students of growth. Electricity and IT are, to most observers, the two most
important GPTs to date, or at least they seem so according to the three criteria that
Bresnahan and Trajtenberg proposed. In this chapter we have analyzed how the U.S.
economy reacted to the creation of these two GPTs. Having discussed in detail GPTs
with reference to the Electrication and IT eras, we believe that we have shown that the
concept is a good way to organize how we think of technological change and its effects.
The Electricity and IT eras differ in some important ways. Electrication was more
broadly adopted, whereas IT seems to be technologically more revolutionary. The pro-
ductivity slowdown is stronger in the IT era but the ongoing spread of IT and its
continuing precipitous price decline are reasons for optimism about growth in the com-
ing decades relative to what happened in the middle of the 20th century following the
spread of Electricity. But it is the similarities between the two epochs that are the most
instructive and that will guide our expectations about how the next GPT will affect
economic life when it comes along.
Acknowledgements
We thank Jason Cummins, Bart Hobijn, Josh Lerner and Gianluca Violante for provid-
ing us with some of the data used here. This research was supported in part by National
Science Foundation Grant No. 30-3207-00-0079-286.
References
Atkeson, A., Kehoe, P.J. (1993). Industry evolution and transition: The role of information capital. Staff
Report No. 162. Federal Reserve Bank of Minneapolis, Minneapolis, MN.
Atkinson, T.R. (1967). Trends in Corporate Bond Quality. With the assistance of E.T. Simpson. Columbia
University Press, New York, for National Bureau of Economic Research.
Bahk, B.H., Gort, M. (1993). Decomposing learning by doing in plants. Journal of Political Economy 101,
561583.
Bartel, A., Lichtenberg, F. (1987). The comparative advantage of educated workers in implementing new
technology. Review of Economics and Statistics 69, 111.
Berndt, E.R., Dulberger, E.R., Rappaport, N.J. (2000). Price and quality of desktop and mobile personal
computers: A quarter century of history. Working paper. MIT Sloan School, Cambridge, MA.
Berry, T.S. (1988). Production and population since 1789: Revised GNP series in constant dollars. Bostwick
Paper No. 6. The Bostwick Press, Richmond, VA.
Bessen, J. (2002). Technology adoption costs and productivity growth: The 70s as a technology transition.
Review of Economic Dynamics 5, 443469.
Board of Governors of the Federal Reserve System (1976). Banking and Monetary Statistics 19141941.
Board of Governors of the Federal Reserve System, Washington, DC.
Boldrin, M., Levine, D.K. (2001). Growth cycles and market crashes. Journal of Economic Theory 96,
1339.
1222 B. Jovanovic and P.L. Rousseau
Bradstreet Co. (18851925). Bradstreets. Bradstreet Co., New York.
Brady, D.S. (1966). Price deators for nal product estimates. In: Brady, D.S. (Ed.), Output, Employment,
and Productivity in the United States After 1800. Columbia University Press, New York, pp. 91116.
Bresnahan, T.F., Trajtenberg, M. (1996). General purpose technologies: engines of growth?. Journal of
Econometrics, Annals of Econometrics 65, 83108.
Caballero, R.J., Hammour, M.L. (1994). The cleansing effect of recessions. American Economic Re-
view 84, 13501368.
Comin, D. (2002). Comments on James Bessens Technology adoption costs and productivity growth: The
70s as a technology transition. Review of Economic Dynamics 5, 470476.
The Commercial and Financial Chronicle (18851925).
Cowles, A., Associates (1939). Common Stock Price Indexes. Cowles Commission for Research in Eco-
nomics, Monograph No. 3, second ed. Principia Press, Bloomington, IN.
Cummins, J.G., Violante, G.L. (2002). Investment specic technical change in the United States (1947
2000): Measurement and macroeconomic consequences. Review of Economic Dynamics 5, 243284.
David, P. (1991). Computer and dynamo: The modern productivity paradox in a not-too-distant mirror. In:
Technology and Productivity: The Challenge for Economic Policy. OECD, Paris, pp. 315347.
Devine, W.D. (1983). From shafts to wires: Historical perspectives on electrication. Journal of Economic
History 43, 347372.
Dow Jones and Company, Inc. (19972004). The Wall Street Journal. Dow Jones, New York.
DuBoff, R.B. (1964). Electric power in American manufacturing, 18891958. Ph.D. Dissertation. Univer-
sity of Pennsylvania.
Dun and Bradstreet, Inc. (2003). D&B Million Dollar Directory. Dun and Bradstreet, Inc., Bethlehem, PA.
Feder, B. (1988). Advances in drugs, courtesy of computers. New York Times (August 3), 5.
Federal Reserve Bank of St. Louis (2004). FRED Database. Federal Reserve Bank of St. Louis, St. Louis,
MO.
Gates, B. (1999). Business @ the Speed of Thought. Warner Books, New York.
Goldin, C., Katz, L.F. (1999a). The shaping of higher education: The formative years in the United States,
1890 to 1940. Journal of Economic Perspectives 13, 3762.
Goldin, C., Katz, L.F. (1999b). The returns to skill in the United States across the twentieth century. Work-
ing Paper No. 7126. National Bureau of Economic Research, Cambridge, MA.
Gordon, R.J. (1990). The Measurement of Durable Goods Prices. University of Chicago Press, Chicago, IL.
Gordon, R.J. (2000). Does the new economy measure up to the great inventions of the past?. Journal of
Economic Perspectives 14, 4974.
Gort, M. (1969). An economic disturbance theory of mergers. Quarterly Journal of Economics 94, 624642.
Greenwood, J., Jovanovic, B. (1999). The information-technology revolution and the stock market. Ameri-
can Economic Review Papers and Proceedings 89, 116122.
Greenwood, J., Jovanovic, B. (2001). Accounting for growth. In: Hulten, C.R., Dean, E.R., Harper, M.J.
(Eds.), New Developments in Productivity Analysis. University of Chicago Press, Chicago, IL, for Na-
tional Bureau of Economic Research, pp. 179222.
Greenwood, J., Seshadri, A., Yorukoglu, M. (2002). Engines of liberation. Economie dAvant Guarde 2.
Greenwood, J., Yorukoglu, M. (1997). 1974. CarnegieRochester Conference Series on Public Policy 46,
4995.
Griliches, Z. (1957). Hybrid corn: An exploration in the economics of technological change. Economet-
rica 25, 501522.
Griliches, Z. (1969). Capital-skill complementarity. Review of Economics and Statistics 51, 465468.
Helpman, E., Trajtenberg, M. (1998a). A time to sow and a time to reap: Growth based on general purpose
technologies. In: Helpman, E. (Ed.), General Purpose Technologies and Economic Growth. MIT Press,
Cambridge, MA, pp. 5583.
Helpman, E., Trajtenberg, M. (1998b). The diffusion of general purpose technologies. In: Helpman, E.
(Ed.), General Purpose Technologies and Economic Growth. MIT Press, Cambridge, MA, pp. 85119.
Hickman, W.B. (1958). Corporate Bond Quality and Investor Experience. Princeton University Press, Prince-
ton, NJ, for National Bureau of Economic Research.
Ch. 18: General Purpose Technologies 1223
Hobijn, B., Jovanovic, B. (2001). The IT revolution and the stock market: Evidence. American Economic
Review 91, 12031220.
Homer, S., Sylla, R. (1991). A History of Interest Rates, third ed. Rutgers University Press, New Brunswick,
NJ.
Hoovers, Inc. (2000). Hoovers Online: The Business Network. Hoovers, Inc., Austin, TX.
Hopenhayn, H.A. (1992). Entry, exit, and rm dynamics in long run equilibrium. Econometrica 60, 1127
1150.
Hornstein, A., Krusell, P. (1996). Can technology improvements cause productivity slowdowns?. NBER
Macroeconomics Annual 1996, 209259.
Jovanovic, B., Nyarko, Y. (1996). Learning by doing and the choice of technology. Econometrica 64, 1299
1310.
Jovanovic, B., Rousseau, P.L. (2001). Why wait? A century of life before IPO. American Economic Review
Papers and Proceedings 91, 336341.
Jovanovic, B., Rousseau, P.L. (2002a). Moores law and learning-by-doing. Review of Economic Dynam-
ics 4, 346375.
Jovanovic, B., Rousseau, P.L. (2002b). The Q-theory of mergers. American Economic Review Papers and
Proceedings 92, 198204.
Jovanovic, B., Rousseau, P.L. (2002c). Mergers as reallocation. Working Paper No. 9277. National Bureau
of Economic Research, Cambridge, MA.
Kelley, E.M. (1954). The Business Founding Date Directory. Morgan and Morgan, Scarsdale, NY.
Kendrick, J. (1961). Productivity Trends in the United States. Princeton University Press, Princeton, NJ.
Kortum, S., Lerner, J. (1998). Stronger protection or technological revolution: What is behind the recent
surge in patenting?. CarnegieRochester Conference Series on Public Policy 48, 247304.
Krusell, P., Ohanian, L.E., Rios-Rull, J.V., Violante, G.L. (2000). Capital-skill complementarity and inequal-
ity: A macroeconomic analysis. Econometrica 68, 10291053.
Kuznets, S. (1961a). Capital in the American Economy: Its Formation and Financing. Princeton University
Press, Princeton, NJ.
Kuznets, S. (1961b). Annual estimates, 18691955. Manuscript. Johns Hopkins University, Baltimore, MD.
Laitner, J., Stolyarov, D. (2003). Technological change and the stock market. American Economic Re-
view 93, 12401267.
Lerner, J. (2000). 150 years of patent protection. Working Paper. National Bureau of Economic Research,
Cambridge, MA.
Lichtenberg, F., Siegel, D. (1987). Productivity and changes in ownership of manufacturing plants. Brook-
ings Papers on Economic Activity 3, 643673. Special Issue on Microeconomics.
McGuckin, R., Ngyen, S. (1995). On productivity and plant ownership change: New evidence form the
longitudinal research database. RAND Journal of Economics 26, 257276.
Moodys Investors Service (1920, 1928, 1955, 1980). Moodys Industrial Manual. Moodys Investors Service,
New York.
Nelson, R.L. (1959). Merger Movements in American Industry, 18951956. Princeton University Press,
Princeton, NJ, for National Bureau of Economic Research.
Nelson, R., Phelps, E. (1966). Investment in humans, technological diffusion, and economic growth. Amer-
ican Economic Review 56, 6979.
The New York Times Co. (18971928). The New York Times. The New York Times Co., New York.
The New York Times Co. (19131925). The Annalist: A Magazine of Finance, Commerce, and Economics.
The New York Times Co., New York.
Rousseau, P.L. (1999). Share liquidity and industrial growth in an emerging market: The case of New
England, 18541897. Historical Working Paper No. 103. National Bureau of Economic Research, Cam-
bridge, MA.
Schoar, A. (2002). Effects of corporate diversication on productivity. Journal of Finance 57, 23792403.
Schwert, G.W. (2000). Hostility in takeovers: In the eyes of the beholder?. Journal of Finance 55, 2599
2640.
1224 B. Jovanovic and P.L. Rousseau
Standard and Poors Corporation (1981, 1988, 2000). Stock Market Encyclopedia. Standard and Poors Cor-
poration, New York.
Standard and Poors Corporation (19712003). Stock Reports. Standard and Poors Corporation, New York.
United States Bureau of the Census, Department of Commerce (1975). Historical Statistics of the United
States, Colonial Times to 1970. Government Printing Ofce, Washington, DC.
United States Bureau of the Census, Department of Commerce (1980, 1992, 2003). Statistical Abstract of the
United States. Government Printing Ofce, Washington, DC.
United States Bureau of the Census, Department of Commerce (20002004). Current Population Survey.
Government Printing Ofce, Washington, DC.
United States Bureau of Economic Analysis (2002). Survey of Current Business. Government Printing Ofce,
Washington, DC.
United States Bureau of Economic Analysis (2004). Survey of Current Business. Government Printing Ofce,
Washington, DC.
University of Chicago Center for Research on Securities Prices (2004). CRSP Database. University of
Chicago Center for Research on Securities Prices, Chicago, IL.
Wiggins, S.N., Raboy, D.G. (1996). Price premia to name brands: An empirical analysis. Journal of Indus-
trial Economics 44, 377388.
Williamson, J.G., Lindert, P.H. (1980). American Inequality: A Macroeconomic History. Academic Press,
New York.
Wilson, J.W., Jones, C.P. (2002). An analysis of the S&P 500 index and Cowless extensions: Price indexes
and stock returns, 18701999. Journal of Business 75, 505533.
Abstract Attention has increasingly shifted towards the long-run perspective on
technological innovation, which suggests that progress comes in waves, each one
originating with a major breakthrough or general purpose technology (GPT). This
paper seeks to assess whether nanotechnology is likely to be (or become) a GPT, a
characteristic that other researchers have sometimes assumed though not necessarily
documented. Based on a survey of existing literature, this paper will explore the
extent to which nanotechnology addresses three primary characteristics of a GPT:
pervasiveness, innovation spawning, and scope for improvement. The paper draws
on patent and patent citation databases to highlight the types of quantitative and
qualitative information that would be necessary, and in some instances is still
lacking, to characterize fully the nature of nanotechnology.
Keywords Nanotechnology General purpose technology Patent analysis
JEL Classications 0330 0300 0340
1 Introduction
Whenever a new class of technologies emerges, conjectures are advanced on how
likely it is that they will change rms productivity, household production, con-
sumption patterns, and socio-economic relationships. If a core technology has a
J. Youtie (&)
Georgia Tech Enterprise Innovation Institute, Atlanta, GA 30332-0640, USA
e-mail: jan.youtie@innovate.gatech.edu
M. Iacopetta
School of Economics, Georgia Institute of Technology, Atlanta, GA 30332-0615, USA
e-mail: maurizio.iacopetta@econ.gatech.edu
S. Graham
College of Management, Georgia Institute of Technology, Atlanta, GA 30332-0520, USA
e-mail: stuart.graham@mgt.gatech.edu
123
J Technol Transfer
DOI 10.1007/s10961-007-9030-6
Assessing the nature of nanotechnology: can we uncover
an emerging general purpose technology?
Jan Youtie Maurizio Iacopetta Stuart Graham
Springer Science+Business Media, LLC 2007
substantial and pervasive effect across the whole of society, it is often termed a
General Purpose Technology (GPT). The dissemination of electricity at the turn
of the 19th century is often said to have the character of a GPT, with reference
made to the long wave of downstream innovations spawned by the electric dynamo
that reshaped the functioning of the economy. Similarly, the dissemination of
microelectronics in the last quarter of the 20th century has in it the hallmarks of a
GPT in that it led to new forms of organizations, new products, and has increased
the level of competition in service goods that were traditionally produced and
consumed locally.
The question we explore in this paper is whether the family of nanotechnologies
has the potential of inducing changes in the economy that are comparable in scope to
electricity, information and communications technology (ICT), and others that have
been previously documented as major breakthroughs.
We discuss the question by drawing on techniques and ideas from two interrelated
streams of research. One line of research has hypothesized that the long-run
behavior of the nancial market and of macro aggregates are best understood by
investigating the conditions that have favored the arrival and the process of
dissemination of major technologies. The main idea from this literature is that
technological change follows a sequence of events in which a major technological
innovation is preceded by a number of smaller inventions that expand the range of
applicability of the core technology, the so-called General Purpose Technology.
In this paper we will briey summarize the main features an innovation should have
to be part of the club of GPTs and discuss the prediction of theories that explain the
rise and fall of productivity and of rms value as the outcome of the dissemination
of a GPT.
We also draw from a stream of research that describes and characterizes tech-
nological developments by means of quantitative data taken from patent datasets. In
particular, we propose a comparison of the level of generality of nanotechnologies
relative of that of ICT (usually considered a GPT) and innovations in the drug
industry (not considered a GPT). We suggest that the kind of tests proposed in the
literature are not easily applicable to emerging technologies because these tests have
been devised for situations in which a considerable amount of historical data has
been recorded. Nevertheless, the estimations that we perform seem to suggest that
nanotechnology satises at least one major feature of a GPT, namely that of gen-
erality.
The paper begins with an introduction of the main attributes of a GPT and briey
explains the extent to which the existing literature may be used to test whether
nanotechnology has one or more of these attributes. Section 3 shows how macro
aggregates are predicted to respond to the arrival and the dissemination of a GPT.
Section 4 illustrates strategies that have been used to identify a GPT. Section 5
introduces the Generality Index and estimates it for nanotechnology and two other
classes of technologies. A conclusion follows.
2 The GPT concept and nanotechnology
Previous research has suggested that a GPT must have at least three attributes:
pervasiveness, an innovation spawning effect, and scope for improvement (Helpman
& Trajtenberg, 1994). Pervasiveness is intended to reect the performance of some
J. Youtie et al.
123
function that is vital to the functioning of a large segment of existing or potential
products and production systems. Bresnahan and Trajtenberg (1995, p. 4) argue that
continuous rotary motion and binary logic are the pervasive elements of steam
power and ICT, respectively, each of which is considered a GPT.
A pervasive technology would have relatively little visibility in the functioning of
the economy unless it fostered new inventions that directly or indirectly result from
the early major invention. For instance, the dynamo led to the invention of both the
light bulb and electric motor, and stimulated major innovation in plant and urban
design (David, 1990). Similarly, the microchip led to an explosion of imaging
technologies, memory devices, and digital technologies.
Helpman and Trajtenberg (1994) suggest that such widespread adoption of a core
technology is a consequence of a variety of actors coordinating their beliefs about
the promise of the technology. Complementary technologies are developed as long
as the various actors involved share beliefs that the GPT is spawning innovations in
multiple technological areas. Indeed, widespread market adoption may be a con-
sequence of the settling of beliefs among scientists, entrepreneurs, established
businesses, government, and consumers.
It remains a theoretical and empirical question whether the core technology of
these breakthroughs could be improved substantially. Evidence for the scope of
improvement for ICT was cleverly summarized by Moores Law which predicted
that the force of competition would stimulate the semiconductor industry to double
the number of transistors per chip every 1824 months. While the regularity of
Moores Law has been observed, it is not clear whether its regularity results from
technological factors or from industry coordination around a smooth and predicable
trajectory with clear transaction-cost benets.
Theory suggests that all three aspects will be present in the true breakthrough
technologies, those most widely used by rms and households. The mere fact that
an innovation can be applied in several areas of production (pervasiveness) does
not mean that it will be used. In order for society to employ the technology
pervasively, its adoption must be convenient from a cost-consideration point of
view, that is, it must reach a certain level of efciency (scope for improvement),
and it must lead to the development of new secondary or complementary
technologies (innovation spawning). Some authors add a fourth element to the
denition of a GPT, that of wide dissemination (Lipsey, Bekar, & Carlaw, 1998),
although this element is often considered a logical consequence of the other three
attributes.
Are there indications that these three basic GPT attributes might be present in
nanotechnology? Can we say that nanotechnology performs (or will perform) a
generic function, whose efciency will be greatly improved over time, perhaps as
much as that of the microprocessor, and that it stimulates the appearance of new
kinds equipment comparable to the modem, or memory storage devices?
With a new technology it is hard to conjecture what aspect of it, if any, will
perform a generic function. Although most scholars who are engaged in nano-
science agree that nanotechnology is very small in scale, (in the range of
1100 nanometers (nm), with one nanometer equaling one billionth of a meter), only
a few seem to embrace the notion that if a technology is small in scale it should be
considered nanotechnology. From the perspective of this analysis such a change in
scale could be paralleled to a generic function, notwithstanding the disagreements
among nano-scientists. This would be the case, for instance, if a new scientic
Assessing the nature of nanotechnology
123
principle or a new methodology allows a signicant drop in scale that leads to a
radical transformation in the range of inputs used in production.
For instance, many industries have voiced concerns over the limits to Moores
Law, recognizing that any stop to the exponential growth of computing capability
would have economic consequences (Harriott, 2000). Nanotechnology, however, has
the potential to sustain circuit density increases through small scale lithography
alternatives, such as nanoimprint lithography, or eventually self-assembly (Arnold,
1995). Moreover, matter at the nanoscale has been shown to exhibit novel properties
that cannot be projected from larger or smaller scales (Kostoff et al., 2006;
Tannenbaum, 2005). These novel behaviors, and the human skills to manipulate and
engineer them, could form the basis for future pervasive applications.
The GPT characteristic of innovation spawning may be embodied in evidence
of a nanotechnology-oriented value chain of initial, intermediate, and downstream
innovations. Lux Research (2006) identies one such nano-value chain, consisting of
an initial set of nanomaterials (such as carbon nanotubes), employed by market
intermediaries to create coatings that enhance properties of nishes. The nal
products then integrate these coatings into a diverse set of products (which may
include automobiles, airplanes, electronics displays, nano-treated clothes, refriger-
ator surfaces with microorganism growth inhibitors, and self-cleaning windows that
oxidize organic matter, among others). Lux Research suggests that this nano-value
chain is supported by a set of tools including scanning probe microscopes,
nanofabrication tools, and computer modeling systems (See also Meyer, 2006 for an
alternative perspective on the nano-value chain).
Figure 1 presents a schematic, sequencing both science and commercial tech-
nologies. The sequence begins with scientic and technological discoveries in
instruments (such as scanning tunneling microscopes or STM and atomic force
microscopes or AFM) and nanomaterials (such as buckminsterfullerenes and carbon
nanotubes), forming the basis for a value chain. Intermediate and complementary-
goods producers are offered, including nanocoatings and composites manufacturers,
nano-core processing and memory. The end of the value chain shows a broad range
of end-use goods. The gure shows that the boundaries between positions in the
value chain overlap.
It is fair to assume that the development of a complete value chain will more
likely follow a coordination of beliefs. In nanotechnology, this coordination of
Fig. 1 Potential sequence of science, intermediate goods, and nal products in nanotechnology
J. Youtie et al.
123
beliefs appears to be taking hold. Evidence of coordination in science includes
Richard Feynmans legendary talk at the American Physical Societys annual
meeting in December 1959 (Theres Plenty of Room at the Bottom), Eric
Drexlers Engines of Creation (1986) and subsequent formation of The Foresight
Institute, the launch of Nanotechnology by the Institute of Physics as a multidisci-
plinary science and engineering journal in 1990, the creation of the Feynman Prize
rst awarded in 1993 to recognize eminent research in nanotechnology, and the
notion that advances at the nanoscale are situated in a convergence of disciplines
(Roco & Bainbridge, 2003).
1
Coordination in the public and private sector may be inferred from the intro-
duction of specic nanotechnology patent classes and cross-referencing categori-
zations: International Patent Class B82, the Japanese Patent and Trade Ofce
(PTO) Class ZNM, the US Patent and Trademark Ofce (USPTO) Class 977, and
the European Patent Ofce Class Y01N. In addition, there are multiple profes-
sional associations (e.g., NanoBusiness Alliance) and trade magazines (e.g., Small
Times) dedicated to nanotechnology. Signicant coordination of consumer beliefs
may be decades away from occurring, although several references to nanotech-
nology are evident in popular press and public policy documents. For instance,
popular media such as Michael Crichtons Prey (2002) has portrayed risks from
nanotechnology, while public policy has responded with legislation. In 2003, the US
Congress enacted The 21st Century Nanotechnology Research and Development
Act (Public Law 108-153) which includes specic mention of societal concerns. This
statutory mandate has been followed by the NSFs allocation of resources to create
a network of science museums and research centers to help educate the public
about nanotechnology and social change (NSF, 2005).
Scope for improvement in the family of nanotechnologies may lie in reductions
in size, lower costs, and greater complexity. Although nanotechnology is still at a
relatively early stage, advances have occurred in semiconductor manufacturing
technology (advancing from 90 nm to 45 nm during 20052007) (Kanellos, 2005,
2006), and in instrument costs (Atomic Force Microscopes can be obtained at
lower prices and/or with greater availability of features at existing prices) (Lux
Research, 2006). It is also expected that the eld of nanotechnology will like ICT
evolve in terms of complexity, with Roco (2004, 2005) suggesting that nanotech-
nology will undergo four generations of development over the next 20 years.
Armed with such information, several researchers have proposed that nano-
technology is a GPT. Huang et al. (2003) demonstrate through patent analysis
that nanotechnology covers a wide range of classes, although Porter, Shapira, and
Youtie (2006) criticize the use of an overly broad denition. Moreover, Shea
(2005) suggests that nanotechnology is a GPT because it is likely a disruptive and
radical technology, but the authors approach does not specically address the
concepts of pervasiveness, innovation spawning, and scope for improvement.
While Palmberg and Nikulainen (2006) do examine whether nanotechnology
exhibits these three characteristics of a GPT, they do not apply methods
commonly used to test for them.
Counting the number of patents by year (showing increases over time) or patents
by patent classication (showing increasing diversity) are commonly used to claim
1
The disciplinary convergence hypothesis is not without its detractors (See for example, Khushf,
2004).
Assessing the nature of nanotechnology
123
that nanotechnology is a GPT. However, such information may not in and of itself be
adequate to make any such determination. Hall and Trajtenberg (2004), for instance,
nd that GPTs (as measured by patent citations) do not necessarily have dispro-
portionately higher growth rates in terms of newly issued patents. Moreover, they
argue that some patent classications tend to be more broad-based than others,
particularly chemical-related classications. They suggest that patent classes by
themselves do not provide sufcient substantiation of breadth and pervasiveness of a
candidate GPT.
3 Why GPTs are important
In the rest of this paper, we explore the gap between studies like Palmberg and
Nikulainen (2006), which have claimed that nanotechnology is a GPT, and Hall and
Trajtenberg (2004), which nd that there are problems with using counting
methodologies. In so doing, we rst nd it necessary to justify the assessment of
whether a technology is a GPT. We see three chief reasons for engaging in this
exercise, contending that there is value to understanding (1) the returns to effective
government innovation policy; (2) the technological drivers of economic growth; and
(3) the manner in which society could most effectively prepare for these broad
technological changes.
3.1 R&D policy
An effective R&D policy would have in it some element of spurring innovation.
It is commonly believed that measures aimed at making intellectual property rights
stronger, or inventors appropriability of the social surplus generated by inventions
greater, tend to alter the supply of innovation, without regard to missed oppor-
tunities for diffusion. In the GPT theory, the improvement of the core technology
goes hand in hand with the range of applications it stimulates in various sectors of
the economy. The lower the price of capital embodying complementary technol-
ogieswhich facilitate the dissemination of the core technologythe wider the
range of adoption, and the brighter the prospects of returns on R&D investments
directed at improving and expanding the scope of the GPT (Brenhan &
Trajtenberg, 1995). However, this argument does not hold in a context in which
innovations are unrelated to each otheras it is assumed by most growth theories
(Aghion & Howitt, 1992; Romer, 1990). In such circumstances, any positive
spilloversgoing from the inventors to the usersare detrimental to the
innovation rate.
It is not difcult to nd cases suggesting that improvements in a GPT are asso-
ciated, to some extent, with the level of appropriability (i.e., how easily the prots
from innovations can be captured). In the semiconductor industry, the level of
appropriability is considered low in comparison to other industries, and yet the
industry has exhibited spectacular improvements in both invention and economic
growth (Irwin & Klenow, 1994). Clearly, the diffusion of information technologies in
many economic sectors has been facilitated by the steep decline in the production
cost of the microchip. Therefore, if a technology is a GPT it may be more efcient to
resolve the classical tension between creating monetary incentives for innovators
J. Youtie et al.
123
and fostering the diffusion of innovation by opting for a relatively high level of
externalities.
3.2 Economic change
The second reason for investigating the nature of a technology is that any such
analysis provides insight into the source of economic expansions or slowdowns. In
the United States, the economy experienced sustained high-output growth during
the 1960s, while from the early 1970s to the early 1980s output growth was low
relative to the post-WWII average. Since the mid-1990s there has been, for the most
part, a return to strong growth. Helpman and Trajtenberg (1994) argue that these
patterns are associated with the diffusion of GPTs.
When considering economic growth, one view suggests that the appearance of a
GPT is followed by two distinct phases. In the rst phase, resources are diverted
from existing production activity to the creation of new technologies complementary
to the GPT. This redeployment of resources from one sector to the other would
cause wage and labor productivity rates to stagnate or even decline. This phase is
often called the time to sow, for the economy is developing technologies that are
unproductive in the short-run either because they are not yet efcient or because
adopters do not possess the necessary skills and knowledge to use them efciently.
This period of economic slowdown may persist.
It has been argued that the productivity slowdown that lasted for almost 25 years
(starting sometime in the rst half of the 1970s in the US and other advanced
countries) was partly due to the spread of computers. The introduction of computers
would have rendered obsolete existing skills and would have caused the abandon-
ment of existing routines. This view sees the slowdown of productivity as the cost the
economy suffered to modernize production and upgrade the types of skills needed in
the New Economy. Paul David (1990) draws a parallel between the diffusion of
electricity and computers, arguing that in both cases there was a long delay between
the introduction of the GPT and the corresponding productivity surge. It is natural
to pose a similar parallel between ICT and nanotechnology, and we explore such a
relationship below.
3.3 Societal synchronization
To the extent that a GPT can produce economic benets, and it requires a high level
of synchronization in society, the identifying of a GPT may be benecial in allowing
society to plan for a higher level of needed synchronization. Under the theory of
GPTs, coordination between inventors and users expectations about the usefulness
of the emerging technology is considered vital. Improvements in the GPT requires
R&D investments, which will likely be made only if the investors expect the
development of new complementary technologies or the renement of existing
technology in downstream sectors. In turn, complementary technologies will emerge
only if inventors are optimistic about the prospect that the GPT will be widely
adopted.
Helpman and Trajtenberg (1994) developed a formal framework showing the
dynamic links between the core and complementary technologies. One important
aspect that emerges from their analysis is that at any given point in time, researchers
Assessing the nature of nanotechnology
123
must decide whether to devote resources to developing complementary technologies
associated with an existing technology, or to a new GPT. The choice depends not
only on what happened in the past but also on expectations about the future. The
decision is also affected by the past, because a new complementary technology is
more productive in an economy in which a wide range of other complementary
technologies have been developed.
For these reasons, an incumbent GPT has advantages over a new one. The
conjecture is that although a new GPT is more productive when it is combined with
the same number of technologies that complement an existing GPT, it is never-
theless not as productive when the range of new complementary technologies is
small. If investors believe that at a certain point a large number of complementary
technologies associated with the new GPT are forthcoming, then it is more likely
that these technologies will be developed. Otherwise, the economy may become
trapped in an equilibrium where the new GPT is never adopted and the few rms
who ventured to invest in it will fail.
4 Strategies to identify a GPT
In the foregoing, we have (1) suggested that identifying whether a technology is a
GPT is a valuable exercise, and (2) cataloged other studies nding that nanotech-
nology exhibits the characteristics of a GPT. In this section, we identify what we
believe is a more systematic approach, one that has been undertaken in recent years
by empirical scholars and economic historians in assessing whether a technology is a
GPT, and apply it to nanotechnology. The objective of this discussion is to both
examine whether nanotechnology holds up as a GPT, and also to identify tools
that may be used to test other emerging technologies.
Jovanovic and Rousseau (2005) propose new ways to use historical evidence to
test for the existence of scope for improvement, wide range of use, and the
likelihood of spawning complementary innovations. They capture a technologys
scope for improvement by analyzing the decline of equipment prices (nding that
this decline was greater for ICT than for electricity). The authors suggest that wide
range of use (pervasiveness) can be analyzed, in their case by comparing the
amount of electric power as a percentage the total horsepower in several manu-
facturing sectors with an estimate of the percentage of ICT investments across the
same sectors. Electricity exhibited dissemination across a broader range of industries
than did ICT over a comparable time period.
The authors also examine the likelihood that a technology will spawn comple-
mentary innovations by using patent analysis. They rely on a notion that patents
presage investment in new technologies, representing the rise in initial public
offerings (IPOs) and the subsequent change in the structure of capital in favor of
new technologies. The authors nd innovation spawning in the surge of
IPO activity that followed the introduction of both electricity and information
technologies.
Hall and Trajtenberg (2004) and Moser and Nicholas (2004) venture beyond a
simple tally of the number of patents to contend that the data can reveal much more
on the question of whether a technological development is a GPT. Instead of
measuring the pervasiveness of a GPT by looking at how a new technology affects
the composition of capital across sectors, these authors look at the extent to which
J. Youtie et al.
123
patents associated with a GPT are cited outside the technology area or industry in
which the GPT originated. To address the diffusion delay, which has been associated
with a prolonged productivity slowdown followed by an acceleration of productivity,
this literature measures the citation lags (the amount of time between the issue of a
focal patent and the issue of a future patent that cites back to it), which they contend
should be longer for GPTs than for incremental technologies. These authors
measure the scope for improvement not as a reduction in production cost, but
instead with the number of citations within the technological area to which the GPT
belongs. An interesting question that the patent literature addresses is the date of
arrival of a GPT. Historians often pick a specic event, usually identied with a
major investment (e.g., for electricity it has been identied as the construction of a
hydro-electric facility at Niagara Falls, New York). Macroeconomists tend to
consider an arbitrary threshold in the data, for instance investments in the new
technologies rising above a certain percentage of overall investment. In contrast, the
patent literature utilizes an Originality measure, which allows for a determination
of the date of arrival of inuential innovations based on forward citations.
Does the patent approach lead to the same conclusions as the macro approach?
Moser and Nicholas contend that in the case of electricity it does not. These authors
nd that patented inventions associated with electricity led in the 1920s were not as
pervasive as were chemical and mechanical inventions, because these latter
inventions were cited by later patents outside their technological areas more often
than were electricity inventions. Moreover, chemical and mechanical inventions
were cited more often in general, indicating to the authors a stronger propensity to
spur innovation.
These differing conclusions may stem from chemicals and mechanical industries
being more science-based than was electricity. This observation is an important
warning for all nanotechnology investigations that rely exclusively on patents data.
There is a risk that the traditional test developed by Hall and Trajtenberg (2004)
may be biased in the sense that a science-based technology is more likely to be
designated as a GPT, even when such a determination is a spurious result, and not
borne out by a judgment made after considering the economic effects that it actually
generates.
5 Evidence from patent data
By and large, employing patent data to uncover evidence of a general purpose
technology is a problematical exercise when studying emerging technologies.
Patent data, by its nature, offers information about the current state of a technology,
and more commonly about the past development of that technology. Because new
technologies are in the process of emerging, the patent characteristics that have
traditionally been collected are either not available, or numbers are rather small and
thus prone to statistical error. So, empirical research studying whether technologies
are or were GPTs has been undertaken only after the technology has matured
sufciently so as to allow researchers to collect an adequate body of information
from the patent record (Hall & Trajtenberg, 2004 (ICT); Moser & Nicholas, 2004
(electricity)).
Because nanotechnology is an emerging technology, we are faced with the same
limitations. To demonstrate the embryonic nature of this technology area, we
Assessing the nature of nanotechnology
123
present data on the patenting of nanotechnology inventions 19832005 (Fig. 2). Four
trends of granted patents are plotted in Fig. 2, based on two different denitions of
nanotechnology patents.
2
One denition is derived from the classication system
used at the USPTO, while another is based upon the keyword system built by the
CNS-ASU team (Porter et al., 2006). Note that the latter denition produces sub-
stantially more US nanotechnology patents during the 19832005 period (12,553
NSF-CNS patents versus 2,639 USTPO-dened patents). Figure 2 presents alter-
native counts within each denition, measuring each patent both in its year of
application and grant.
The time trends demonstrate that nanotechnology has been a slowly growing
technology space. Signicant application activity did not begin until the late 1980s
(under either denition), with relatively small numbers of patents being issued until
the 1990s. This relative paucity in the patent record is particularly problematic for
the uncovering of GPTs since the primary measurement of this characteristic of
technologies has been made with large numbers of patents, using the patents
forward citation stream.
All patent applicants are required by law to disclose the prior art (patents and
non-patents) upon which the instant invention builds. Following some give-and-take
with the patent ofce (Graham, 2006), and after the patent examiner has had an
0
400
800
1200
1600
1
9
8
3
1
9
8
5
1
9
8
7
1
9
8
9
1
9
9
1
1
9
9
3
1
9
9
5
1
9
9
7
1
9
9
9
2
0
0
1
2
0
0
3
2
0
0
5
Year
C
o
u
n
t
NSF CNS app NSF CNS iss USPTO app USPTO iss
Fig. 2 US nanotechnology patents, 19832005. Note: NSF CNS nanotechnology patents selected
according to a keyword list generated at Georgia Tech in cooperation with the CNS at Arizona State
University. USPTO nanotechnology patents selected according to US Patent and Trademark Ofce
classications. iss denotes patents graphed according to issue date. app denotes patents graphed
according to application year, and thus the seeming decline in applications is not a true decline but
instead a consequence of truncation because the data source is issued patents
2
These plots are subject to substantial right-side truncation due to the data being collected from the
granted patentsthese gures do not include information available in the US since 2001 on
published applications.
J. Youtie et al.
123
opportunity to add some references (Alcacer and Gittleman, 2006), the patent issues
with a list of patent backward citations to prior art. Researchers have used these
backward citations by looking to patents that issue in the future, and which cite
back to the focal patent, to measure the pervasiveness of the technology.
Hall and Trajtenberg (2004) suggest using these so-called forward citations in
uncovering GPTs by employing the HerndahlHirschman Index (HHI) of the
patent classes assigned to the focal patents forward citations. The resulting
measure of pervasiveness is termed a patents Generality Index and is dened
by the formula
Generality G
i
1
X
n
i
j
S
2
ij
where S
ij
= share of patent is forward citations in class j. The theory behind using
this measure is that it captures information on the extent to which the focal patent is
being applied in a wide range of technologiesthe so called pervasiveness of a
patent. As a patents Generality score approaches 1.0, we know that the patent is
being cited to by patents in a broader set of classications, and thus we can infer
that the patent is being applied more broadly in distant applications.
If we examine the Generality scores across all patents in a particular tech-
nologyin nanotechnology for instanceand compare these against the scores for
patents in other technologies, we may infer something about the pervasiveness of
the technologys application throughout the economy. Obviously, any measure built
in this way will be very sensitive to right truncation. In the study of new and
embryonic technologies in which the patent record is slowly developing, the absence
of a sufcient forward time window will pose great difculties in calculating a
useful Generality Index for individual patents, and thus entire patented
technologies.
The trends depicted in Fig. 2 demonstrate that, in the emerging nanotechnologies,
substantial numbers of patents began to issue from the USPTO in the 1990s, thus
giving us a sufciently long forward window to develop credible Generality
scores on the earliest patents issued in this new technology space. We present these
data below with one important caveat: These generality scores are only represen-
tative to the extent that these early patents, and their characteristics, are represen-
tative of later patents issued in this emerging technology area.
Table 1 reports generality scores for nanotechnology patents as dened by the
NSF-CNS keyword method, and compares these with scores for drug and
computer patents dened by international patent classications (Graham, 2006).
Scores are reported for the index built from three different measures of patent
forward citation breadth: US patent classications, International patent classica-
tions, and NBER patent-database aggregated technology classes (Hall, Jaffe, &
Trajtenberg, 2001). Generality scores for 19901993 irrespective of classication
scheme demonstrate that the nanotechnology patents are more general than drugs
patents issued in the relevant year, and compare favorably with, and indeed are
higher at every reading than, computer patents. Computer patents are representative
of ICT, a technology that we previously mentioned has been found to be a GPT
(Hall & Trajtenberg, 2004).
Assessing the nature of nanotechnology
123
These scores on early nanotechnology patents provide us with limited evidence
that nanotechnology exhibits the pervasiveness characteristic of a GPT. As
mentioned previously, conventional measurement techniques for assessing the
existence of a GPT require sufcient time to elapse to allow for forward citations
to develop in the patent record. So, are researchers at an impasse as regards
nanotechnologyis the evidence of GPT pervasiveness hidden in the latent
nature of patent information?
Not necessarily. If ICT is a guide, then it is possible to determine whether this
technology, in its early development, also showed associated patents with signi-
cantly higher generality scores. And indeed, in 19751979, patents designated in
primary international class G06F, a class broadly representative of computer soft-
ware technologies (Graham & Mowery, 2003) showed relatively high generality
scores as compared to all other patents.
3
Thus, for this GPT, the patents that
appeared early in the technological trajectory demonstrated this measure of per-
vasiveness and thus we can take some condence in our nanotechnology patents
showing the same characteristic.
But, as outlined above, a GPT is not characterized by pervasiveness alone. Can
the patent record help us in determining whether nanotechnologyand indeed, any
candidate GPTexhibits evidence of coordination of beliefs among actors or a
sequential development of complementary technologies? We theorize here that
the patent record may contain information that could help in uncovering these two
characteristics of GPTs.
As regards coordination of belief, the key insight behind this characteristic of a
GPT is that different actors in society are conforming to a set of beliefs con-
cerning the wide applicability of the technology. The set of actors needed to coor-
dinate are necessarily broad, but include at least marketplace actors (such as
Table 1 Comparison of Generality Index scores across three technologies, 19901993
Nanotechnology Drugs Computers
Variable Count Mean Count Mean Count Mean
1990 Gen US 287 0.620 2188 0.386 1961 0.612
Gen IC 287 0.642 2187 0.385 1961 0.443
Gen TC 287 0.540 2187 0.273 1961 0.424
1991 Gen US 293 0.623 2405 0.394 2306 0.610
Gen IC 293 0.617 2405 0.389 2306 0.445
Gen TC 293 0.507 2405 0.278 2306 0.431
1992 Gen US 411 0.596 2349 0.387 1956 0.612
Gen IC 411 0.582 2349 0.388 1956 0.405
Gen TC 411 0.487 2349 0.268 1956 0.417
1993 Gen US 364 0.608 2499 0.380 2999 0.609
Gen IC 364 0.605 2498 0.376 2999 0.398
Gen TC 364 0.511 2498 0.264 2999 0.423
Variable denition: Gen US = Generality based on USPTO-classes; Gen IC = Generality based on
International Patent Classes; Gen TC = Generality based upon NBER patent database technology
classes (Hall et al., 2001).
3
These scores (built from USPTO classes) for computer patents 19751979 were 0.63, 0.65, 0.66,
0.68, and 0.67, respectively, and are signicantly higher than the scores for all patenting in those
years (0.49 in each year).
J. Youtie et al.
123
entrepreneurs) and government actors. Since 2001, the patent record has contained
information concerning the identity of the actor who included the citation in the
patent references: the inventor (and the inventors attorney agent), or the patent
ofce examiner (Alcacer & Gittleman, 2006). While the use of these data would
entail care due to possible endogeneity (the idea that information is owing between
the parties that leads to the citation placement),
4
nevertheless to the extent that the
generality measure is built from the citation record, and that citation record can be
tested to determine whether both marketplace actors (inventors) and government
actors (examiners) concur about the pervasiveness of a technology, it may prove
fruitful to use these records as an indicator of coordination of beliefs.
Finally, GPTs are characterized by the sequential development of complementary
technologies. This sequential development may be seen in the patent record: Cita-
tion lags (the time between the patent grant and the arrival of its forward citations)
tend to be longer as complementary technologies require more time to develop,
citation counts themselves are higher, as GPTs are characterized by a burst of
invention (Hall & Trajtenberg, 2004). In a GPT, we are also likely to see in the
patent record the technologies being adopted in a wide range of industrial sectors, as
market actors throughout the economy begin to deploy the technology generally, to
many different industrial and technology settings. Therefore, patent records may
again be used to offer a window into the development ex post of technologies that
build upon, and are complementary to, the focal technology.
6 Conclusion
This study has considered the notion that nanotechnology may be a breakthrough
innovation with long-run economic and societal affectswhether it is a General
Purpose Technology (GPT), and has explored appropriate methodologies to ad-
dress this question. GPTs are a phenomenon of much interest because these tech-
nologies have a profound impact on growth and the productivity levels in the
economy and of individual rms, despite initial periods of slowdown as resources are
devoted to the development of complementary technologies. Hence, GPTs peak the
attention of public and private sector decision makers.
Our literature review demonstrates that there is a growing body of work that
considers nanotechnology a GPT. As we show, however, few if any of these studies
claiming a nanotechnology-GPT link have developed a systematic approach for
determining if this designation is tting.
The core premise inherent in the characterization of a GPT is that it must meet
three criteria: pervasiveness, innovation spawning, and scope for improvement. Our
study has highlighted qualitative evidence suggesting that nanotechnology may
exhibit characteristics of a GPT. More importantly, we put forth the application of
quantitative methods to uncover the nature of nanotechnology. Through the use of
patent data, citations analysis, and a Generality index derived from patent clas-
sications, we demonstrate that nanotechnology exhibits similar pervasiveness
levels to that of ICT, an existing GPT. We further propose that patent data may be
effectively used to examine innovation spawning attributes such as: (1) identifying
4
Another problem for any such analysis would arise if the placement of citations by examiners was
driven by some internal USPTO job incentive.
Assessing the nature of nanotechnology
123
a critical number (burst) of technologies in intermediate sectors, and (2) the
types of convergence of believes needed for diffusion to occur. The question is
still open regarding how to use quantitative data to solicit information about the
scope for future improvements in nanotechnology.
We believe that an effort to evaluate the currently available data through the lens
of existing theory and empirical methodology is important for understanding the
possible socio-economic implications of nanotechnology during the early phases of
dissemination. We conjecture that if nanotechnology is more of an incremental
technology, there will be little change in short or long run investment decisions. But
if it is a major breakthrough innovation, as a good part of the evidence we explored
suggests, disruptions of business routines are to be expected, existing skills may
become obsolete more rapidly, and capital shifts from established rms to younger
rms of investors that are more likely to embrace the new technology may occur.
Moreover we observed the governments optimal strategy to spur innovation is
drastically different when an emerging technology has the character of a GPT than
when it is an incremental technology. The level of appropriability of technologies
complementary to the core innovation should be lower in the former case than in the
latter one. The fast dissemination of complementary technologies associated with an
emerging technology is a necessary condition to reach a critical level of acceptability
which induces even relatively moderate risk-takers to invest in the emerging
technology.
Acknowledgements Authors contributed equally to this work. This research was undertaken at
Georgia Tech through the Center for Nanotechnology in Society (CNS-ASU), supported by the
National Science Foundation under NSF NSEC CNS SES #0531194, in collaboration with research
sponsored under NSF NSEC CNS SES #0531146. We gratefully acknowledge helpful comments we
received from Philip Shapira and participants at the Technology Transfer Society Annual Confer-
ence, September 27, 2006, Atlanta, Georgia, USA.
References
Aghion, P., & Howitt, P. (1992). A model of growth through creative destruction. Econometrica,
60(2), 323351.
Alcacer, J., & Gittleman, M. (2006). Patent citations as a measure of knowledge ows: The inuence
of examiner citations. Review of Economics and Statistics 88(4), 774779.
Arnold, W. (1995). The SIA lithography roadmap. Microlithography World, Winter, 711.
Bresnahan, T. J., & Trajtenberg, M. (1995). General purpose technologies: Engines of Growth?
NBER Working Paper No. 4148, Cambridge, MA: National Bureau of Economic Research.
Crichton, M. (2002). Prey. New York: HarperCollins.
David, P. (1990). The dynamo and the computer: An historical perspective on the modern pro-
ductivity paradox. The American Economic Review, 80(2), 355361.
Drexler, E. (1986). Engines of creation: The coming era of nanotechnology. New York: Anchor
Books.
Graham, S. J. H. (2006). The determinants of patentees use of continuation applications in the
United States patent and trademark ofce, 198099. In Andersen (Ed.), Intellectual property
rights: Innovation, governance and the institutional environment. Edward Elgar Publishers.
Graham, S. J. H., & Mowery, D. C. (2003). Intellectual property protection in the software industry,
with David C. Mowery. In Wesley Cohen & Steven Merrill (Eds.), Patents in the knowledge-
based economy: Proceedings of the science, technology and economic policy board. National
Academies Press.
J. Youtie et al.
123
Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER patent citation data le: Lessons,
insights and methodological tools. NBER Working Paper No. 8498. Cambridge, MA: National
Bureau of Economic Research.
Hall, B., & Trajtenberg, M. (2004). Uncovering GPTS with patent data. NBER Working Paper No.
10901. Cambridge, MA: National Bureau of Economic Research.
Harriott, L. (2000). Limits of lithography. Working Paper, Bell Labs-Lucent Technologies.
Helpman, E., & Trajtenberg, M. (1994). A time to sow and a time to reap: Growth based on general
purpose technologies. NBER Working Paper No. 4854. Cambridge, MA: National Bureau of
Economic Research.
Huang, Z., Hsinchun, C., Yip, A., Ng, G., Guo, F., Chen, Z., & Roco, M. (2003). Longitudinal patent
analysis for nanoscale science and engineering: Country, institution, and technology eld.
Journal of Nanoparticle Research, 5, 333363.
Irwin, D. A., & Klenow, P. J. (1994). High tech R&D subsidies: Estimating the effects of sematech.
NBER Working Paper No. 4974. Cambridge, MA: National Bureau of Economic Research.
Jovanovic, J., & Rousseau, P. (2005). General purpose technologies. Working Paper 11093.
Cambridge, MA: National Bureau of Economic Research.
Kanellos, M. (2005). AMDs 65 nano question solved: Its a second half thing. CNET News.com,
November 2, http://news.com.com/2061-10791_3-5929097.html?tag=nl.
Kanellos, M. (2006). Intel shows test chips made on future processes. CNET News.com,
January 25, http://news.com.com/Intel+shows+test+chips+made+on+future+processes/2100-
1006_3-6030489.html.
Khushf, G. (2004). A hierarchical architecture for nano-scale science and technology: Taking stock
of the claims about science made by advocates of NBIC convergence. In Baird, D., Nordmann,
A., & Schummer, J. (Eds.), Discovering the nanoscale (pp. 2133). Amsterdam: IOS Press.
Kostoff, R., Stump, J., Johnson, D., Murday, J., Lau, C., & Tolles, W. (2006). The structure and
infrastructure of global nanotechnology literature. Journal of Nanoparticle Research, 8(34),
301321.
Lipsey, R. G., Bekar, C., & Carlaw, K. (1998). What requires explanation? In Helpman, E. (Ed.),
General purpose technologies and economic growth (pp. 1454). Cambridge: MIT Press.
Lux Research (2006). The nanotech report: 4th edition, volume 1. New York: Lux Research.
Meyer, M. (2006). What do we know about innovation in nanotechnology? Some propositions about
an emerging eld between hype and path-dependency. Paper presented at the 2006 Technology
Transfer Society Conference, September 2729, Atlanta, Georgia.
Moser, P., & Nicholas, T. (2004). Was electricity a general purpose technology? The American
Economic Review, Papers and Proceedings, 94(2), 388394.
Palmberg, C., & Nikulainen, T. (2006). Industrial renewal and growth through nanotechnology?
Helsinki: ETLA, The Research Institute of the Finnish Economy.
Porter, A., Shapira, P., & Youtie, J. (2006). Rening search terms for nanotechnology. Brieng
paper, prepared for presentation at national science foundation, rst year review for the center
for nanotechnology in society (CNS-ASU), Arizona State University, Tempe, AZ.
Roco, M. (2004). Nanoscale science and engineering: Unifying and transforming tools. AIChE
Journal, 50(5), 890897.
Roco, M. (2005). International perspective on government nanotechnology funding in 2005. Journal
of Nanoparticle Research, 7(6), 707712.
Roco, M., & Bainbridge, W. (Eds.) (2003). Converging technologies for improving human perfor-
mance: Nanotechnology, biotechnology, information technology and cognitive science. Boston:
Kluwer Academic Publishers.
Romer, P. (1990). Endogenous technical change. Journal of Political Economy, 98, 71103.
Shea, C. (2005). Future management research directions in nanotechnology: A case study. Journal of
Engineering and Technology Management, 22, 185200.
Tannenbaum, R. (2005). Advances in materials and self-assembly. In Advanced Technology and the
Future of Manufacturing, Proceedings. Atlanta, Georgia: Georgia Institute of Technology.
US National Science Foundation (2005). New grants are awarded to inform the public and
explore the implications of nanotechnology, Press Release 05-179, http://www.nsf.gov/news/
news_summ.jsp?cntn_id=104505&org=olpa&from=news.
Assessing the nature of nanotechnology
123
c h a p t e r 3
...................................................................................................................................................
NETWORKS OF
I NNOVATORS
...................................................................................................................................................
walter w. powell
sti ne grodal
3. 1 Introduction
........................................................................................................................................................................................
In February of 2001, two rival consortia published rough draft (roughly 90 per cent
complete) sequences of the human genome in Nature and Science. The public
Human Genome Project consisted of Wve key institutions and eleven collaborators,1
supported by the US National Institutes of Health, Department of Energy, and the
Wellcome Trust in the United Kingdom. The rival private consortia, led by the
biotechWrmCelera, included both commercial Wrms andacademic researchers from
the Universityof California, PennState, Case Western, Johns Hopkins, Cal Tech, Yale,
Rockefeller, as well as scientists in Spain, Israel, and Australia. These projects have
been acclaimed for their remarkable scientiWc achievement; they were also the
product of considerable organizational innovation. In contrast to the Manhattan
Project or Project Apollo, both of which were hierarchically organized, national
projects, the Human Genome Project (HGP) and the Celera team were pluralist,
multiorganizational, multinational confederations. These twogroups were intensely
rivalrous, but collaborated intensively within their own groups (Lambright 2002).
HGP involved management by two government agencies and a private British
We are grateful to David Mowery and Jan Fagerberg for their careful readings of earlier drafts.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 56
foundation that coordinated activities in government labs, universities, and
nonproWt institutes in the US and England. As the lead Wrm, Celeras organization
was more focused, but its research team included scientists and state-of-the-art
equipment at private Wrms, public and private universities, and nonproWt institutes
in four countries.
Both projects were organized as large-scale networks, and their rivalry spurred
each side to engage in a high-stakes learning race. While the cost, scale, and
distributed nature of these projects may have been unusual, the form of organiza-
tioncollaboration across multiple organizational boundaries and institutional
formsis no longer rare. Indeed, many analysts have noted that the model of
networks of innovators has become commonplace over the past twodecades (Powell
1990; Rosenbloom and Spencer 1996; Roberts and Liu 2001; Chesbrough 2003).
Collaboration among ostensible rivals was once regarded as a provisional or
transitional step taken to enter new markets, spread risks, or to share early stage
R&D costs (Mowery 1988). Such forays were often followed by mergers as the
transitory activities became incorporated inside the boundaries of the Wrm. Recent
studies suggest, however, that various forms of interorganizational partnerships are
now core components of corporate strategy. Even where these linkages endure for
relatively lengthy periods of time, they do not entail vertical integration (Gomes-
Casseres 1996; Hagedoorn 1996; Noteboom 1999; Ahuja 2000a). Contemporary
studies of industrial performance are replete with reports of a signiWcant upsurge
invarious types of interorganizational collaboration. While these collaborations can
take a number of forms (including research consortia, joint ventures, strategic
alliances, and subcontracting) and span a wide range of key functions, a National
Research Council analysis of trends in industrial research and development (R&D)
suggests that the innovation process has undergone the most signiWcant transform-
ation over the past decade (Merrill and Cooper 1999). In a survey of the period
196098, Hagedoorn (2002) Wnds a sharp growth in R&D collaborations, beginning
in the late 1970s and continuing through the mid-1990s.
A National Research Council assessment of eleven US-based industries, pur-
posefully diverse in character and technology but all resurgent in the 1990s, observes
in every sector an increased reliance on external sources of R&D, notably univer-
sities, consortia and government labs, and greater collaboration with domestic and
foreign competitors, as well as customers in the development of new products
and processes (Mowery 1999: 7). Other surveys also point to the enhanced centrality
of interorganizational collaboration, especially in R&D. For example, National
Science Foundation data show a marked increase in the number of international
alliances between US and Western European countries between 1980 and 1994;
but by the mid-1990s, the formation rates for intranational alliances linking US
Wrms with their domestic competitors outpace international linkages (National
Science Board 1998). The former collaborations were motivated largely by concerns
with market access, while the latter focus more on the development of new
technologies.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 57
networks of innovators 57
Similarly, there is now ample research illustrating the growing links between US
Wrms and universities (Powell and Owen-Smith 1998), and greater involvement by
Wrms and government labs in research joint ventures (Link 1996, 1999). In the realm
of science, Hicks and Katz (1996) Wnd that research papers are much more likely to
be co-authored and involve authors with multiple institutional aYliations that
span universities, government, and industry. Distributed networks of practice
are the organizing basis for many technical communities, suggesting both that
sources of knowledge are now more widely dispersed and that governance mechan-
isms are emerging to orchestrate distributed knowledge. The open source software
movement is but one highly visible example of this trend (OMahony 2002; Weber
2003), which illustrates how advances in information technology have greatly
facilitated virtual networks. In short, as Mowery (1999: 9) observes, the diversity
of institutional actors and relationships in the industrial innovation process has
increased considerably. Complex networks of Wrms, universities, and government
labs are critical features of many industries, especially so in Welds with rapid
technological progress, such as computers, semiconductors, pharmaceuticals, and
biotechnology.
Our goal in this chapter is to assess the state of scholarly research on the role of
networks in the innovation process. We begin with a review of the factors that have
triggered the increased salience of networks. We discuss diVerent types of networks,
distinguishing between networks that are based more on contractual or market
considerations, and those that are based on less formal, and more primordial
relationships, such as common membership in a technological community or a
regional economy. We then turn to a discussion of the analytical leverage provided
by the tools of network analysis. This stream of research, which spans sociology,
social psychology, organizational behavior, and business strategy, highlights key
distinctions between highly clustered, dense networks, steeped in overlapping ties
and high in trust, and weak-tie networks, that provide access to novel, non-redun-
dant information. We next reviewa number of empirical studies of the contribution
of networks to the innovative output of Wrms. We take up the issue of knowledge
transfer, examining how the codiWcation of knowledge can shape what is transmit-
ted through networks. We brieXy discuss the governance of networks, and then
conclude with an assessment of what types of organizations and settings derive the
greatest impact on innovation from participation in networks.
Research on the relationship between networks and innovation is a relatively
recent area of inquiry. While there is a good deal of work underway, direct analyses
measuring the impact of interWrm networks on performance are limited. Much of
the extant research focuses on the eVects of networks on patenting, access to infor-
mation, and the generation of novel ideas. Moreover, the studies often examine
high-tech industries, where investment in R&D is pronounced. Attention to
the consequences of network ties for the Wnancial performance of Wrms is relatively
rare.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 58
58 walter w. powell and stine grodal
3. 2 Why have Networks Grown
in Importance?
........................................................................................................................................................................................
The advantages of a heterogeneous group of contacts are well established in both
social theory and network analysis. A strong tradition of theory and research,
running from Simmel (1954) to Merton (1957) to Granovetter (1973) to Burt
(1992), makes abundantly clear that there are informational, status, and resource
advantages to having broad and diverse social circles. Below we review an array of
recent empirical studies that demonstrate how interorganizational relationships
lead to various beneWts with respect to information diVusion, resource sharing,
access to specialized assets, and interorganizational learning. In science and tech-
nology-based Welds, the advantages that accrue fromdiverse sources of information
and resources are considerable. Not surprisingly, then, as the commercialization of
knowledge has assumed greater importance in economic growth, collaboration
across organizational boundaries has become more commonplace. Interorganiza-
tional networks are a means by which organizations can pool or exchange resources,
and jointly develop new ideas and skills. In Welds where scientiWc or technological
progress is developing rapidly, and the sources of knowledge are widely distributed,
no single Wrm has all the necessary skills to stay on top of all areas of progress and
bring signiWcant innovations to market (Powell and Brantley 1992; Powell, Koput,
and Smith-Doerr 1996; Hagedoorn and Duysters 2002). In such settings, networks
can become the locus of innovation, as the creation of knowledge is crucial to
improving competitive position.
Collaborative networks have long been central to the production process in craft-
based industries (Eccles 1981), in industrial districts (Brusco 1982; Piore and Sabel
1984), and in Welds such as aerospace where assembly depended upon key inputs
from diverse participants. The growth of knowledge-intensive industries has
heightened the importance of networks in R&D as well as product development
and distribution. A persistent Wnding from a diverse set of empirical studies is that
internal R&D intensity and technological sophistication are positively correlated
with both the number and intensity of strategic alliances (Freeman 1991; Hagedoorn
1995).
For organizations in rapidly developing Welds, heterogeneity in the portfolio of
collaborators allows Wrms to learn from a wide stock of knowledge. Organizations
with broader networks are exposed to more experiences, diVerent competencies,
and added opportunities (Beckman and Haunschild 2002). Such access creates an
environment in which creative abrasion, the synthesis that is developed from
multiple points of view, is more likely to occur. In this view, innovation occurs at
the boundaries between mind sets, not within the provincial territory of one
knowledge and skill base (Leonard-Barton 1995: 62). By having access to a more
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 59
networks of innovators 59
varied set of activities, experiences, and collaborators, companies broaden the
resource and knowledge base that they can draw on. By developing more multiplex
ties with individual partners, either through pursuing multiple collaborations or
expanding an existing R&Dpartnership into downstreamdevelopment, companies
increase the points of contact between them. When relationships are deepened,
greater commitment and more thorough knowledge sharing ensue. Organizations
with multiple and/or multifaceted ties to others are likely to have developed better
protocols for the exchange of information and the resolution of disputes (Powell
1998). Parties that develop a broader bandwidth for communication are, in turn,
more capable of transferring complex knowledge. In science-driven Welds such as
biotechnology, organizations that develop ties to diVerent kinds of organizations
and carry out multiple types of activities with these organizations are central players
in industry networks (Powell et al. 2004). These centrally positioned organizations
are both capable of pulling promising new entrants into the network and collabor-
ating with a wide assortment of incumbents. Moreover, research shows that in
biotechnology, organizations lacking such connections fail to keep pace and fall by
the wayside (Powell et al. 2005).
3. 3 Varieties of Networks
........................................................................................................................................................................................
The literature onnetworks emphasizes that they are most pronouncedinthe domain
between the Xexibility and autonomy of markets and the force and control of
organizational authority (Powell 1990). Networks thus combine some of the incen-
tive structures of markets with the monitoring capabilities and administrative
oversight associated with hierarchies (Mowery, Oxley, and Silverman 1996). For
our purposes, we include networks based on formal contractual relations, such as
subcontracting relationships, strategic alliances or participation in anindustry-wide
research consortium, and informal ties, based on common membership in a pro-
fessional or trade association, or even a looser aYliation with a technological
community.
One can diVerentiate networks with respect to their duration and stability, as well
as whether they are forged to accomplish a speciWc task or evolve out of pre-existing
bonds of association. Networks vary from short-term projects to long-term rela-
tionships, and the diVerent temporal dimensions have important implications for
governance. Some networks are hierarchical, monitoredby a central authority; while
others are more heterarchical, with distributed authority and strong self-organizing
features. Grabher and Powell (2004) focus on temporal stability and forms of
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 60
60 walter w. powell and stine grodal
governance to diVerentiate four key types: informal networks (based on shared
experience); project networks (short-term combinations to accomplish speciWc
tasks); regional networks (where spatial propinquity helps sustain a common
community); and business networks (purposive, strategic alliances between two
parties). These types do not represent essentialist categories; rather they may overlap
and interweave with one another. Consider these forms as useful coordinates to
locate networks with respect to diVerent combinatory elements.
Several key concepts provide potent analytical tools that apply across diVerent
types of networks and permit assessment of their eVects. First, consider the diVer-
ences between strong and weak ties (Granovetter 1973). In interpersonal terms, a
strong tie is a personwithwhomyou interact on a regular basis, while a weak tie is an
acquaintance, or a friend of a friend. Strong ties are important for social support, but
much of the novel information that a person receives comes fromweak ties. Strong
ties are based on common interests, consequently most information that is passed
reinforces existing views. Weak ties introduce novelty in the form of diVerent ideas
or tastes, and by introducing new information they are, for example, invaluable in
job searches and other circumstances where a small amount of new information is
highly useful. Weak ties have a longer reach, but a much narrower bandwidth than
strong ties. The latter are more cohesive, and often prove to be more eVective at the
exchange of complex information. Figure 3.1 illustrates the diVerence between
strong and weak ties.
Much of the research on interWrm networks extrapolates from interpersonal
relations.2 In general, this is a plausible analytical move; however, it elides the
question of whether relationships at the Wrm level are dependent on ongoing
interpersonal ties, and whether the business relationship would be harmed or
severed if the key participants were to depart. The extent to which interorganiza-
tional ties are contingent upon relations among individuals is a key question for
scholarly research, as well as a critical challenge for business strategy (Gulati 1995;
Powell 1998).
A second notable contrast is the distinction between networks as bridges and as
structural holes (Burt 1992). Bridges are points of connection between parties that
Reference
person
Strong tie
Weak tie
Fig. 3.1 Strong and weak ties
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 61
networks of innovators 61
lack ties, such as when Aknows B, and Cknows B but not A. B is the bridge between
A and C, thus the gateway to a linkage between A and C. Granovetter (1973) argued
that bridges are the links that make weak ties possible. Burt (1992) deepened the
argument by moving fromthe who question(i.e., which position ina network is best
situated) to the question of how certain structural arrangements generate beneWts
and opportunities. He coined the term structural holes as a potential connection
between clusters of units that are not connected. The possibility of making such a
connection provides leverage, or opportunities for arbitrage. Those positioned to
take advantage of structural holes can broker gaps in the social structure. See Figure
3.2 for illustration.
There is debate as to whether strong or weak ties, or bridges or structural holes,
oVer greater opportunities for innovation (Ahuja 2000a; Ruef 2002). Clearly, vari-
ation in network structures is associated with diVerent content in relationships.
Strong ties between two parties may restrict information gathering in terms of the
breadth of search, but the information that is exchanged is thick, or detailed and
rich. Weak ties are thinner and less durable, but provide better access to non-
redundant information. There is also disagreement as to whether networks can be
designed or pruned to produce optimal shapes, without triggering repercus-
sions. Whether location in a network is highly malleable or not, position in a
network both empowers and constrains opportunities.
A third point of contrast is between networks formed intentionally across a
market interface to accomplish a task and emergent networks that grow out of
ongoing relationships. The former may be considered an instrumental or strategic
relation, while the latter stems from more primordial relations, such as common
ethnicity, friendship, or location. These diVerent starting points matter, but in the
Xuid world of networks, the point of origin does not Wx the evolution of a relation-
ship. Consider two cases. There is a global trend toward vertical disaggregation in
manufacturing, as Wrms are relying on suppliers for design and component inputs
Redundant tie
Structural hole
Reference
person
Fig. 3.2 Structural holes and redundant ties
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 62
62 walter w. powell and stine grodal
in a variety of industries (see Womack, Jones, and Roos 1990, for autos; McKendrick,
Doner, and Haggard 2000, for disk drives). Often these outsourcing decisions are
driven by the need to reduce costs, save time, and enhance Xexibility, while the large
Wrm concentrates on those activities in which it has some form of competitive
advantage. But as many analysts have noted, there is no natural stopping point in
this relation (Sabel 1994; Helper, McDuYe, and Sabel 2000). The subcontractor can
end up involved in design issues, doing critical R&D, or become central to eVorts to
improve quality. What began as a choice to outsource can, in some circumstances,
become either a deep, mutually dependent collaboration or a highly ambiguous and
opportunistic partnership. Helper et al. (2000) and Dyer and Nobeoka (2000)
illustrate the marked trend for automobile subcontracting to evolve into interde-
pendent, bilateral relationships.
Or consider the contemporary life sciences, where many R&D partnerships
emerge out of ongoing intellectual relationshipsco-authorships, mentormentee
relationships, and common training (Murray 2002). These informal personal rela-
tionships may, however, come to involve signiWcant intellectual property in the form
of patents, and thus become highly formalized contractual agreements between
organizations. We oVer these examples as illustrations that networks forged out of
strategic purposes can take on strong relational elements, while more personal ties
can become contractual and highly speciWed. While it is possible to assign networks
to either a transcational (i.e., based on a consideration of business opportunities
without regard to prior social relations) or relational (i.e., embedded in ongoing
social relationships) category, it is inappropriate to assume that relations remain
Wxed. As networks evolve, there is considerable give and take. Intense competition
can render calculative strategic alliances more embedded, while the prospect of great
Wnancial reward can turn a handshake relationship between individuals into a
formal legal linkage between Wrms.
Figure 3.3 provides a typology of diVerent forms of networks, with the horizontal
axis representing degree of purposiveness, ranging from informal to contractual.
The vertical axis represents the extent of embeddedness, varying from open, epi-
sodic, or Xuid to recurrent, dense connections among a fairly closed group (Gran-
ovetter, 1985). We illustrate each of the four cells with examples of types of
innovation networks. In the lower left cell we place informal networks, such as a
scientiWc invisible college, that emerge out of sharedexperience or common interest.
Although these relations tend to be temporary and short-lived, the gray arrows are
intended to show that these informal linkages can evolve into formal business
alliances or more enduring primordial relations, where participation is more con-
stant andless Xuid. The primordial network inthe upper left cell is characterized by a
common social identity, continuous participation, and close ties. All these features
are often found in professional networks, craft-based occupations, ethnic commu-
nities, and industrial districts. The upper right cell is typiWed by involvement in a
common project. Membership in such a network is typically restricted and often
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 63
networks of innovators 63
governed by a lead Wrm. A supply-chain network or a large construction project
are apt examples. The gray arrows are meant to illustrate that supply-chain relations
can evolve into either occupational communities and industrial districts or into
formal business partnerships. The most purposive andinstrumental type of network
is the strategic alliance, represented by the lower right cell. We turn now to a
discussion of the relationship between these forms and the innovation process.
Primordial
M
o
r
e
c
l
o
s
e
d
m
e
m
b
e
r
s
h
i
p
H
i
g
h
l
y
f
l
u
i
d
Embeddedness
Informal
Calculativeness
Contractual
Supply Chain
Invisible College
Strategic
Biotech Venture
capital
Pharma
University
Research
institution
Firm
Government
Focal
firm
Supply chain
Type:
Nodes similar
Common social identity
Function:
Multifunctional, craft based,
project-based
Structure:
Dense, since the network exists
prior to activity, work relationships
are nested in ongoing personal
relations
Examples:
Film, construction, ethnic
community, diamond trade
Type:
Nodes dissimilar
Common work identity
Function:
Horizontal or vertical specialization
Division of labor
Incremental innovation
Structure:
Spider web
The establishment of ties creates the
network. The network can become a
form of social identification over
time, as for example in the Toyota
supply chain network.
Examples:
Type:
Nodes both similar and dissimilar
Common interest
Function:
Fast access to news and novel ideas
Discovery
Structure:
Ties reinforce the structure of the
college
Examples:
research collaboration, information
sharing
Type:
Nodes both similar and dissimilar
Identity different
Function:
Division of labor
Recombination
Structure:
Purposive network, emerges through
active tie creation
Examples:
Common in high technology
industries like semiconductors,
biotechnology, and chemicals
Fig. 3.3 Network typology
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 64
64 walter w. powell and stine grodal
3. 4 Empirical Studies of the Role
of Networks in Innovation
........................................................................................................................................................................................
3.4.1 Formal Ties
Most empirical studies of the relationship between networks and innovation focus
on formal ties established among organizations. This streamof research documents
a strong positive relationship between alliance formation and innovation, across
such diverse industries as chemicals (Ahuja 2000a), biotechnology (Powell et al.
1996, 1999; Walker, Kogut, and Shan 1997; Baum, Calabrese, and Silverman 2000),
telecommunications (Godoe 2000), and semiconductors (Stuart 1998, 2000). The
diversity of the research contexts suggests the eVects of network structure may be
generalizable. Nevertheless, most research has focused on high technology indus-
tries, and uses patents as a proxy for innovation. More direct measures of innovative
outputs are needed. Some of the important themes that emerge from this research
highlight speciWc tie characteristics, technological uncertainty, and network evolu-
tion. In addition, researchers have emphasized the increased beneWts in the form of
resources and knowledge that alliances provide to entrepreneurial Wrms.
Tie characteristics. One line of research focuses on how diVerent types of ties
inXuence the beneWts derived from alliances. Vinding (2002) identiWed 548 Danish
manufacturing Wrms that developed one or more new products over a two-year
period. In interviews with a subset of the companies, he Wnds that the impact of a
collaboration on innovation is signiWcantly related to both the type of partner and
the pattern of previous collaborative relationships. The importance of prior inter-
action with partners points to the signiWcance of relationship building, and how
elements such as trust and cognitive understandings require time to develop.
Domestic partners were found to have a greater positive impact on innovative
performance than foreign partners, possibly due to the higher costs, both psycho-
logical andWnancial, associatedwithmore distant collaborations. Vindings research
emphasizes beneWts derived fromstrong local ties. Similarly, in a ten-year (198090)
case study of the R&D portfolio of a Norwegian telecommunications organization,
Godoe (2000) reports comparable results with respect to strong ties. His analysis
suggests that radical innovations were more likely to emerge from intimate and
prolonged interaction. But in the Norwegian case, the aYliations were not local, but
instead based on membership in international telecommunications associations.
Powell et al. (1999) emphasize that experience with collaboration and centrality in
the network derived from a diverse set of ties are important determinants of
innovation among biotechnology Wrms over the period 198899. Their analyses
suggest that centrality and experience resulted in more patenting. The most conse-
quential connections in terms of patenting were R&Dpartnerships. The diversity of
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 65
networks of innovators 65
network ties also had a positive inXuence on rates of patenting. Powell et al. (1999)
found that while network experience had a positive inXuence on patenting, the rate
of increase diminished with additional experience, suggesting possible declining
returns to network connectivity. The question of whether there are limits to con-
nectivity needs to be investigated further.3 These results suggested a cycles of
learning process in which R&D collaborations generate attention that attracts
other partners, who collaborate in developing novel ideas. This enhanced diversity
of aYliations increases a Wrms experience at managing collaborations and transfer-
ring knowledge, and increases their centrality in the industry network. Greater
centrality is associated with a higher rate of patenting, and both centrality and
higher volumes of patenting trigger subsequent R&D partnerships, restarting the
cycle for centrally placed Wrms.
Most research has looked at the presence or absence of a formal collaboration.
Ahuja (2000a), however, developed a more nuanced analysis including both direct
and indirect ties, and the level of indirectness. Drawing data from a sample of 97
Wrms in the global chemicals industry, he used the number of patents as a measure of
innovative output, while collaboration was measured through formal ties. More
distant connections through aYliates of partners were coded as weak or indirect ties.
The results show that both direct and indirect ties have a positive inXuence on
innovation, though the impact of indirect ties is smaller than the impact of direct
ties. The number of direct ties also negatively moderates the impact of indirect ties.
In contrast to Burts (1992) arguments about the arbitrage opportunities available
through non-redundant contacts, Ahuja shows that a network with many structural
holes can reduce innovative output, as measured by rates of patenting. A key
advantage of close-knit networks may be due to their superior ability to transfer
tacit knowledge (Van Wijk, Van den Bosch, and Volberda 2003). In an analysis of the
exchange of information across project teams in a large multinational computer
company, Hansen (1999) also illustrates that complex knowledge is transferred most
easily through tightly knit networks.
Entrepreneurial Wrms. One active area of research concerns the eVects of networks
on survival chances of newly founded Wrms. Larsons (1992) ethnographic study of
how a startup Wrm grew and prospered by drawing on external resources and
support for key business functions illustrates how relationships are forged and
sustained as startup Wrms grow. While not explicitly looking at innovative output,
Larson added insight into the signal importance of networks in obtaining resources
necessary to fuel a startup Wrms success. Shan, Walker, and Kogut (1994) examined
whether biotechnology startupWrms cooperative relationships withother Wrms had
a positive eVect on patenting. Their results oVer support for the argument that
collaborative relationships increased innovation, because formal cooperative rela-
tionships explained innovative output, while innovative output did not account for
the pattern of alliances. The salience of alliances for young and small Wrms is further
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 66
66 walter w. powell and stine grodal
emphasized in Stuarts (2000) study of innovation in the semiconductor industry.
His dataset includes 150 Wrms, followed by the consultancy Wrm Dataquest over the
period 198591. Drawing on sales Wgures, patterns of strategic alliances, and
patenting activity, Stuart shows that Wrms possessing technologically sophisticated
alliance partners patented at a substantially greater rate than those that lacked such
ties. Firms establishing strategic alliances with large partners also grew at a higher
rate than Wrms without access to such partners. The returns to networks with regard
to patenting were greatest for both young and small Wrms.
Baum, Calabrese, and Silverman(2000) pursue a similar question, asking howthe
composition of a startup Wrms alliance portfolio aVects its performance. Using data
on 142 biotechnology Wrms founded in Canada between 1991 and 1996, they Wnd a
positive eVect of alliance formation on startup innovation. Network eYciency,
deWned as the diversity of information and capabilities per alliance, showed a large
positive eVect on the number of biotech patents. Alliances with direct competitors
had a negative eVect on innovation, however. These results were moderated when
the rival biotechnology Wrm had a larger share of the relevant market or if the rival
biotechnology Wrm was highly innovative. Of the various performance measures
used, the number of patents and the volume of R&D expenditures were most
signiWcantly inXuenced by rates of alliance formation.
Network dynamics. Drawing on Cohen and Levinthals (1990) ideas about absorp-
tive capacity, Powell et al. (1996) argued that Wrms utilize external collaborations to
stay abreast in rapidly developing technological Welds. But organizations cannot be
passive recipients of new knowledge. What can be learned is crucially aVected by
what is already known (Powell et al. 1996: 120). To understand the news generated
externally, organizations have to make news internally. In this fashion, the rate of
acquisition of skills and resources fromthe outside is closely linked to the generation
of expertise internally. In their work on the global biotechnology industry, they Wnd
that Wrms that develop experience at managing collaborative R&D relationships
garner faster access to centrally positioned organizations. As experience at collabor-
ating grows, Wrms widen the network of organizations withwhomthey partner. As a
Wrms experience with collaboration and its diversity of partners increase, the more
central and visible the Wrmbecomes in the industry. This centrality leads, in turn, to
growthinthe size of the Wrm, andtothe ability tocoordinate more alliances, creating
a feedback cycle. This cycle of learning has been shown to be associated with positive
Wnancial performance (Powell et al. 1999), and a greater ability to collaborate with
diverse kinds of organizations, which permits Wrms to retain a leadership position in
the industry (Powell et al. 2005).
The general picture that emerges from research in organizational sociology and
business strategy is one in which networks and innovation constitute a virtuous
cycle. External linkages facilitate innovation, and at the same time innovative
outputs attract further collaborative ties. Both factors stimulate organizational
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 67
networks of innovators 67
growth, and appear to enhance further innovation. Ahuja (2000b) and Stuart
(2000), for example, demonstrate that Wrms with many prior patents are more likely
to form alliances than Wrms lacking patents, suggesting a recursive process of
innovation and growth in which collaborative ties play a central role. Further
attention needs to be given, however, to such issues as the eVects of the duration
of linkages, experience with collaboration, and the consequences of broken ties on
rates of innovation.
Technological uncertainty. Two additional aspects of the innovation process involve
the relationship between strategy and alliance formation and the level of techno-
logical uncertainty in the Weld. Eisenhardt and Schoonhoven (1996) studied the
population of semiconductor Wrms launched in the US between 1978 and 1985, and
found that the more risk-taking a companys strategy, the more alliances a company
formed. One explanation is that as Wrms gain credibility for developing pioneering
technologies, access to Wnancial and other resources for developing innovative
technology is secured through alliances. An alternative view is that alliances are
necessary to share the attendant risks inhigh-velocity environments. Sarkar, Echam-
badi, and Harrisons (2001) analysis of managers in a range of high-tech industries
revealed that an active strategy of alliance formation enhanced performance, as
measured by market share, sales growth, market development, and product innov-
ation. They also report that managers who perceived the environment as more
uncertain were more likely to pursue alliances. In addition, smaller Wrms derived
more value from network linkages than larger Wrms, presumably because smaller
companies viewed the technological landscape as more uncertain.
Rosenkopf and Tushman (1998) examined the role of technical communities in
the Xight simulation industry, where cooperative technical organizations play a
critical role in developing standards and advancing the state of the art. In a study
covering the years 195892, they found long periods marked by incremental change
punctuated by shorter eras of ferment. They show that the rate of founding of
technical networks increases during periods of discontinuity, and stabilizes into
core cliques when ferment declines and a dominant design emerges. Subsequent
technological change disrupts dominant cliques, and triggers the formation of new
networks, restarting the cycle. Thus, both technological strategy and industry
evolution are linked to patterns of network formation, with external networks
assuming greater importance during periods of technical discontinuity and for
Wrms with more risk-taking strategies. The importance of industry technical com-
mittees in standards setting has also been emphasized in the computer industry
(Farrell and Salomer 1988) and videocassette recorders (Cusumano et al. 1992).
The overall conclusion of this group of studies is that networks provide access to
more diverse sources of information and capabilities than are available to Wrms
lacking such ties, and, in turn, these linkages increase the level of innovation inside
Wrms. Younger and smaller Wrms may beneWt more fromcollaborative relationships
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 68
68 walter w. powell and stine grodal
than do larger Wrms. Most notably, Wrms with a central location within networks
generate more innovative output. Both direct and indirect ties provide a positive
contribution to innovation, but the eVect of indirect ties is moderated by the
prevalence of direct ties. The evidence for the beneWts of structural holes is not
uniform; where structural holes might be beneWcial is in the search for new infor-
mation, but the knowledge transfer process appears to be facilitated by closer-knit
networks. From the view of the dynamics of collaboration, successful external
relations appear to beget more ties, which fuel Wrmgrowth and innovation. Clearly,
there are limits to this cycle, but research has not addressed this question in depth
thus far.
The majority of the studies reviewed in this section have been carried out using
patents as the dependent variable and formal relationships as the independent
variable. Patents provide a measure of novelty that is externally validated through
the patent examination process, hence they are a useful indicator of knowledge
creation (Griliches 1990). But patents have some limitations. Some kinds of innov-
ations are not patented, and there is variation in the extent of patenting across
industries. (See Chapter 14 by Malerba on inter-industry variation in innovation
processes.) On the other hand, the focus of many of these studiessemiconductors,
chemicals, biotechnologyis in Welds where patenting is commonplace, and com-
petitors in these sectors are active patentors. The attention to these high-technology
industries raises questions, however, as to the generalizability of the results to other
less knowledge-intensive industries.
One study that speaks to diVerences across industries is Rowley, Behrens, and
Krackhardts (2000) analysis of strong and weak ties in the steel and semiconductor
industries. This study made a notable eVort to distinguish between strong aYlia-
tions, where alliances entailed signiWcant resource commitments and regular inter-
actions, and more arms-length transactions, where there was a rapid exchange,
and the relationship was characterized by less frequency and depth. For example,
equity alliances, joint ventures, and R&D partnerships were categorized as strong
ties, while licensing, patent agreements, and marketing relations were classiWed as
weak ties. Recognizing that weak ties serve as bridges to novel information, while
strong ties are useful for both social control and the exchange of tacit knowledge,
they Wnd divergent results. In the steel industry strong ties are positively associated
with performance; while in semiconductors weak ties are more eYcacious. They
suggest these Wndings reXect the importance of search and product innovation in
semiconductors, and a focus on improvements in the production process for steel.
Much of the research on buyersupplier relations and subcontracting has
focused on more traditional industries, such as automobiles or textiles. To be sure,
these industries make considerable use of technological advances, but they are less
science-driven. As a consequence, the sources of relevant knowledge are not as
widely dispersed. Strong ties thus tend to predominate over weak ties. But the
content of those ties can evolve, changing from contractual to relational. Consider
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 69
networks of innovators 69
subcontracting relations, particularly one of the more notable examplesthe
Toyota auto production network. Researchers stress how the density of overlapping
ties that connect this chain of production facilitates knowledge sharing, mutual
learning, and fast responsiveness (Dyer 1996; Dyer and Nobeoka 2000). But the trust
and reciprocity that characterize this dense network are the outcome of a long
developmental process. In the 1950s and early 1960s, when Japanese Wrms competed
on the basis of lowest cost, relationships with subcontractors were hierarchical and
asymmetric. As Wrms increasingly competed on the basis of quality and innovation,
however, complex multitiered supply relationships underwent signiWcant change.
These relationships can remain hierarchical in two key respects: the larger lead Wrm
often has a signiWcant Wnancial stake in a supplier or aYliate, and it initiates the
production process. But the asymmetry has been sharply reduced. Suppliers, in an
eVort to remain competitive, make signiWcant investments in new equipment,
constantly upgrade workers skills, and take on more critical aspects of the assembly
process (Helper et al. 2000). In turn, the larger Wrms oVer long-termcontracts, share
employees and provide technical assistance, and make Wnancial investments to fund
equipment upgrades.
Too explicit a focus on formal, contractual linkages, however, neglects the myriad
informal ties that connect organizations. All kinds of informal interactions take
place between organizations, including participation in ad hoc industry commit-
tees, or executive education programs, conferences, trade association activities, and
the like. Personnel mobility and common educational backgrounds may also foster
informal linkages across Wrms. Suchinformal connections may be the basis onwhich
more formal, contractual alliances are forged. Indeed, the success of formal aYlia-
tions may hinge on the strength of informal ties. Thus we turn to a discussion of
noncontractual relations.
3.4.2 Informal Ties
Informal patterns of aYliation have long been a central topic in sociology and
anthropology, where studies of friendship networks, advice and referral networks,
and communities are common. There is also a well-established strand of research in
organization theory that points out howinformal relations within organizations are
often not closely aligned with formal authority (Dalton 1959; Blau 1963). Asmall line
of work focuses on the impact of informal networks in large, multinational com-
panies (Ghoshal and Bartlett 1990; Hansen 1999). Relatively few studies, however,
link informal ties to the innovation process, and there is scant research on informal
interorganizational relations.
Scholars have often argued that the sharing of complex information is enhanced
by embedded ties, which suggests that informal ties have the potential to make a
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 70
70 walter w. powell and stine grodal
signiWcant contribution to innovation. There is a strong sense among researchers
that informal relations undergird formal ties. Powell et al. (1996) argue that, in the
life sciences, beneath most formal ties lie a sea of informal ties. Nevertheless, many
organizations are largely unaware of the extent to which formal activities are
buttressed by informal connections (Cross, Borgatti, and Parker 2002).
One of the key studies of informal networks among Wrms was Von Hippels (1987)
work on the sharing of proprietary information among US steel mini-mill produ-
cers. Based on interviews with plant managers and other engineers with direct
knowledge of manufacturing processes, he found that the trading of proprietary
knowledge withboth cooperating andrival Wrms was commonplace. He was initially
surprised that proprietary knowledge was so leaky, but he came to recognize that
information exchange was highly reciprocal and conditioned on expectations that
requests for help would be met. Much of the information that was shared focused on
production problems, matters of pollution control and safety, and issues dealing
with industry-wide concerns. But when relationships among engineers in rival Wrms
were particularly close, more proprietary information was exchanged. Von Hippel
also found that engineers had strong norms of membership in a professional
community that cut across Wrms, and that information trading was a means to
secure reputation and status in that community. He provides numerous examples of
howthe sharing of complex information by engineers contributed to the productiv-
ity of mini-mills.
The cluster of individuals that share a similar set of skills and expertise has been
dubbed a community of practice (Wenger 1998), or a network of practice
(Brown and Duguid 2001). Similar in some respects to a technical community, or
a sophisticated hobby club, these loose groups are engaged in related work practices,
thoughthey do not necessarily work together. SuchXuidgroups are important to the
circulation of ideas. Saxenian (1994) observed ample sharing of proprietary know-
ledge among engineers in Silicon Valley, many of whom have as strong a commit-
ment to their peers within the same occupational group as to their companies.
Saxenian argues that informal knowledge sharing, widely institutionalized as a
professional practice in Silicon Valley, is one of the crucial factors contributing to
its fertile innovative climate. Cohen and Fields (1999) stress that professional ties in
Silicon Valley are forged in complex collaborations between entrepreneurs, scien-
tists, Wrms, and associations, focused on the pursuit of innovation and its commer-
cialization. This collaborative process generates and reWnes the intangible raw
material of technical changeideas.
Kreiner and Schultz (1993) analyze the importance of informal ties through in-
depth interviews with university researchers and industry research directors in the
Danish biotech Weld. They stress that successful collaborative R&D alliances within
the Danish biotech industry are often based on informal ties. A barter economy,
where materials, laboratory tests, chemicals, etc. are exchanged, was pervasive in this
sector. They show that norms of sharing information on the frontier of research aid
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 71
networks of innovators 71
in the formation of more formal networks. As in the mini-mills, information
exchange is not under managerial control, even though such reciprocal Xows can
be channeled by managerial actions.
Many studies of informal relationships stress the signiWcance of trust. Tsai and
Ghoshal (1998) studied the association between intraWrmnetworks and innovation in
Wfteen business units of a multinational electronics company. They found, not
surprisingly, that social ties led to a higher degree of trustworthiness among business
units. Trust increasedresource-exchange andcombinationbetweenthe business units,
which contributedto product innovation. The importance of trust also looms large in
Uzzis (1997) analysis of the diVerence between arms-length ties (a deal in which
costs are everything) and embedded ties (you become friends with these people
business friends. You trust them and their work. Theyre part of the family). Uzzi
conducted interviews and ethnographic observations at twenty-three womens better-
dress Wrms in the NewYork City apparel industry. His study is notable not only for the
quality of his data, but also for his attention to the performance consequences of
diVerent kinds of exchange relations. Uzzi found that organizational performance
increases withthe use of embeddedties tonetworkpartners, as these ties were superior
at conveying complex, context-dependent knowledge. He argued, however, that a
balance between a Wrms embedded ties and a Wrms arms-length ties needed to be
struck, because a network structure comprising only arms-length ties or embedded
ties decreased organizational performance.
The signiWcance of a balanced network structure, mixing formal and informal
aYliations, is also emphasized in Ruef s (2002) analysis of entrepreneurship. He
found that individuals positioned in heterogeneous networks, comprising both
strong and weak ties, are more likely to be regarded as innovative by peers, in
comparison to entrepreneurs in more homogeneous networks. Rosenkopf has
found a similar interweaving of formal and informal relations in her research on
industry-wide expert communities in the areas of Xight simulation and mobile
phones. Rosenkopf, Metiu, and George (2001) analyze joint participation by cellular
service Wrms in technical committees, Wnding that such membership facilitates
subsequent formal interWrm alliance formation. The eVect of participation in
technical committees decreases when Wrms have already established prior alliances,
suggesting that the eVect of informal ties is more catalytic when Wrms do not already
have established alliance partners.
3.4.3 Multi-Party Relationships
Most studies of networks and innovation have examined either dyadic ties, or a focal
Wrm in the context of an overall network. Rosenkopf and Tushmans (1998) work
on expert communities emphasized the importance of studying multi-party
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 72
72 walter w. powell and stine grodal
relationships that connect technical professionals across organizations. Akera (2001)
uses archival data to map the importance of the IBM user group Share in the early
days of computing, oVering a portrait of a large multi-party collaboration. Share was
formed soon after IBMs release of the Wrst mainframe computer in 1953, to enable
IBMs costumers to swap programs, and collaborate on programming so that
duplicative eVort was avoided. Some of the main innovations that came out of
Share provided the basis for both systems programming and operating systems,
which today form the backbone of modern computer use. Share conveyed infor-
mation about hardware changes that substantially improved the design of IBM
computers and peripherals. One of the main contributions of the Share network
was the creation of technical standards, but a second-order eVect was also the
diVusion of knowledge across companies and between users.
Amodern analogue to Share is the Linux community, founded as a group of users
trying to develop an alternative to the operating system supplied by Microsoft. The
network of Linux programmers has proven eVective in developing software in a
highly distributed fashion. In the beginning, most programmers had never met each
other and only kneweach other virtuallyby the username they used when coding.
The Linux community has a very modest organizational structure, relying on a
combination of interpersonal networks and an individuals reputation as a skilled
programmer toserve as the admissionticket tothe network(OMahony 2002; Weber
2003). One diVerence between the Share network and the Linux network is that
corporate interests drove the Share network, while Linux has been primarily driven
by the end users. Nevertheless, in both cases, the decomposability of programming
tasks is an important factor in facilitating distributed networks.
Another large network that has been widely studied is the network of scientists,
often termed invisible colleges (Crane 1972). An invisible college is an informal
network of researchers who form around a common problem or paradigm. By
studying invisible colleges, Crane (1972) hoped to understand how knowledge
grows and how the structure of scientiWc communities aVects the expansion of
knowledge. There are now numerous studies of scientiWc networks, mapping the
structure of co-authorship and citations (Newman 2003 provides a good overview),
though few attend explicitly to the issue of innovation. David (2001), however,
develops a formal model to show that the liberal sharing of knowledge within the
scientiWc community is a major driver of scholarly innovation. One of the historic
characteristics of scientiWc communities is that information and research results
have been distributed openly among members of the relevant community. The shift
toward increasing research commercialization by universities has led some scholars
to question whether the innovative beneWts of invisible colleges will persist, or if
commercial interests will block informal knowledge sharing among scientists
(Powell and Owen-Smith 1998; Owen-Smith 2003). Chapter 8 by Mowery and
Sampat on universityindustry interfaces oVers a more detailed discussion of the
role of universities in the innovation process.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 73
networks of innovators 73
Another line of research that has attended to scientiWc and technological net-
works, dubbed ActorNetwork Theory (ANT), examines howparticular deWnitions
or conWgurations of science and technology triumph over alternative conceptions.
ActorNetwork Theory is unique in its treatment of artifacts and technologies, as
well as people and organizations, as members of a network (Callon 1998; Latour
1987). The primary contribution of ANT to the relationship between networks and
innovation is to showthat not only can networks facilitate innovation, but they also
constrain it by determining the kind of innovations produced, their subsequent
interpretation, and their Wnal use (Callon 2002).
A related line of work that looks at networks as systems of activity is the markets-
as-networks approach, developed by Scandinavian marketing researchers (Hakans-
son and Snehota 1995). This approach examines the multiple relationships among
organizations, and shows how these diVerent aspects of interorganizational rela-
tionships transform and evolve over time. For example, supplier networks may
change frequently, with diVerent elements of production being either outsourced
or insourced (Waluszewski 1995). The resources that are exchanged among the
partners in a production network are constantly changing. What determines
whether an entity is a resource depends on the situation, and its use in combination
with other resources. Resources are, thus, always polyvalent in both use and value.
The participants in a production network, both individually and collectively, de-
velop bonds characterized by trust and commitment. These bonds also have an
organizing eVect on networks, as they shape the identities of actors, and account for
diVerent levels of commitment among participants. This rich vein of qualitative
research has not explicitly focused on innovation, however.
The various studies of multi-party networks tend to emphasize the processual
aspects of collaboration. This attention to content is welcome, but it sometimes
comes at the expense of measuring the output of relationships, particularly how the
sharing and processing of information by members of a network can determine the
generation of novelty. A fuller understanding of the innovation process needs to
examine the topic of information sharing, a subject to which we now turn.
3. 5 Knowledge Transfer
........................................................................................................................................................................................
The role of knowledge transfer is clearly central to the innovation process. Research
has highlightedtwo diVerent aspects of the knowledge-transfer process, each of which
inXuences innovation, albeit in diVerent respects. One explanation for the exchange
of information through networks emphasizes the importance of complementary
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 74
74 walter w. powell and stine grodal
assets in the division of innovative labor (Mowery, Oxley, and Silverman 1996). If
WrmAis good at producing a speciWc component and WrmB is capable of using that
component to produce an engine, they collaborate in a joint production in which
their capabilities reinforce one another. In biotechnology, for example, small Wrms
with close ties to university scientists may excel at drug development, but lack the
skills and resources to manage or fund costly clinical trials. By working closely with a
research hospital and an established Wrm that has a limited pipeline of new medi-
cines, the parties collaborate in a division of labor that is mutually rewarding, and
can result in the participants learning from one another, and accomplishing tasks
they could not do individually.
A second form of knowledge sharing occurs when existing information within a
network is recombined in novel ways. Indeed, novelty is often the unanticipated
result of reconWguring existing knowledge, problems, and solutions (Nelson and
Winter 1982; Fleming and Sorenson 2001). As a consequence of such collisions or
transpositions, Wrms can generate something they were unable to create on their
own. Both forms of knowledge transfer depend on some manner of successful
exchange of ideas, however.
An oft-used distinction is drawn between tacit and explicit knowledge (Cowan,
David, and Foray 2000). Interest in tacit knowledge stems from Polanyis (1956)
argument that we frequently know a good deal more than we can express verbally.
Explicit knowledge is highly codiWed, as in blueprints, recipes, manuals, or in the
form of training. Tacit knowledge lacks such extensive codiWcation (Nonaka and
Takeuchi 1995). Valuable, productive knowledge often demands considerable eVort
to acquire, and suchknowledge is frequently altered inthe process of acquisitionand
application. Perhaps the most vivid example is the continuing eVort of US auto-
makers to acquire, understand, and implement the Japanese system of lean produc-
tion (Womack, Jones, and Roos 1990; Dyer and Nobeoka 2000). Knowledge of
complex production technologies is rarely obtained in a fully digestible form;
understanding inevitably entails learning by using. The distinction between codiWed
and tacit is key because the latter demands considerably more trial-and-error
learning to apply the new knowledge in a diVerent setting.
Many studies point to the relatively easy transferability of explicit knowledge in
contrast to tacit knowledge. Simonin (1999) shows that knowledge transfer within
alliances is negatively aVected by both the nature of knowledge and diVerences in
organizational culture. He observes important diVerences in knowledge exchange
between long and short-lived alliances. Older alliances develop a common language
and shared mental models between partners, suggesting a learning curve within
alliances where the negative eVects of lack of experience and knowledge complexity
subside as the alliance matures. Thus, as an alliance ages and participants develop
relationship-speciWc understanding, there is the opportunity to convey more subtle
forms of information more eVectively. Complex tacit knowledge can become more
explicit as partners develop a wider bandwidth of communication.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 75
networks of innovators 75
If knowledge tacitness is a limiting factor in the transfer of knowledge, then the
cost of transferring knowledge is proportional to the type of knowledge transferred.
Easily transferred knowledge is widely dispersed at a lowcost (Boisot 1998), thus the
likelihood that explicit knowledge contains novel elements that would lead to
innovation is lower. On the other hand, when knowledge is very sticky (von
Hippel 1998) and contains a large tacit component, the degree of diYculty and the
costs of transfer are high. Consequently, the expected gains realized from this
information are uncertain, as the cost of obtaining information may exceed its
value. This suggests that when knowledge involves a moderate level of complexity,
the beneWts derived fromtransfer may be greatest. Figure 3.4 suggests a hypothetical
inverted U-shaped relationship between innovation and codiWcation. Here we
assume that there is variability in the cost of information transfer, and that the
greatest value may be derived when novel ideas are transmitted without too much
diYculty.
Szulanskis (1996) analysis of the transfer of internal benchmarking eVorts in
eight companies suggests key dimensions along which knowledge transfer can be
The costs of
knowledge
transfer
The value of the
innovative output
exceeds the costs of
transfer and application.
Knowledge
codification
There is a medium range of
knowledge codification where the
value of the innovative output
exceeds the costs of knowledge
transfer and recombination. This
domain includes information that
has important elements of novelty
but there are not significant
barriers to transfer.
When knowledge is highly
codified, the value of the
information is low; thus,
while transfer is inexpensive,
the gains are small.
When knowledge is very
tacit, the difficulty of
transfer and implementation
is high, and the payoffs are
uncertain and the process is
costly.
Fig. 3.4 Knowledge codication and innovation
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 76
76 walter w. powell and stine grodal
distinguished. He demonstrates that relationships between sender and receiver are
important, in that both parties need mutual awareness of state-of-the-art practices.
Obviously, communication is critical to information exchange. But even when
relationships function well, some knowledge is causally ambiguous, or sticky, and
thus not easily transferred. Moreover, information exchange is hindered when the
parties have diVerential levels of absorptive capacity, that is, the ability to recognize
the value of newinformation, assimilate it, and apply it to commercial ends (Cohen
and Levinthal 1990). This capacity is essential to innovative capability. For example,
internal success with R&Dand R&Dexpenditures positively aVect a Wrms ability to
exploit the opportunities presented in external relations (Cohen and Levinthal
1990; Powell et al. 1996).
The productive transfer of knowledge is also essential whentwoor more organiza-
tions are able to combine their diVerent capabilities, and create a product or service
that they would not be able to construct on their own. A good illustration is the
Italian motorcycle industry where the locus of innovation is broadly dispersed,
and through cooperation, the participants beneWt from specialization and variety
generation (Lipparini, Lorenzoni, and Zollo 2001). Because all the participants
provide valuable inputs, there is a high commitment to knowledge generation.
The lead Wrms develop relational capacities aimed at pooling the skills of specialized
participants, helping the overall Xow of information and resources in the network.
A parallel analysis of the Italian packaging machinery industry stresses the creation
of a supplier network in which specialized roles are highly complementary
(Lorenzoni and Lipparini 1999). Over time, managers of the core companies de-
veloped a specialized supplier network and each participant focused on a narrow,
but highly competitive set of core competencies. This network structure enabled
every step of the supply chain to specialize in improvements of their speciWc
component, thereby increasing the responsiveness of the participants to market
conditions.
3. 6 Governance and Incentive Issues
........................................................................................................................................................................................
Many studies of interWrm networks draw data from a single point in time and thus
do not examine how collaborations unfold over time. Even studies that do look at
dynamics tend to do so from the perspective of a dyad. Initially, the choice by a
young Wrmof whomto partner with is often driven by resource needs. As both Wrms
and the network mature, various dyadic choices increasingly reXect structural
properties of the network. Thus, the existing network structure shapes search
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 77
networks of innovators 77
behavior. Consequently, networks both enable information to become knowledge
and determine the nature of knowledge (Kogut 2000). But we do not, as yet, have a
parallel understanding of the management and governance of networks to accom-
pany analyses of structure and topology. Concerns with how the parties in a
relationship adapt to changing circumstances, or attend to the incentives to adjust
the relationship to make improvements remain largely unexamined.
Not surprisingly, then, many studies assume that as long as a relationship
persists, the participants are achieving their goals, and view termination as a sign
of failure. Such a view misses the point that the relationship may have completed
its goal or outlived its usefulness. Inkpen and Ross (2001) Wnd evidence that
the termination of alliances might not signify failure, but simply be a sign of the
conclusion of collaborative activity (for example, a new product is launched).
Nevertheless, some collaborations seem to persist even though they have stagnated
or outlived their usefulness. Inkpen and Ross suggest several reasons why interWrm
alliances may persist beyond their optimal duration.4 Alliances can be diYcult
and expensive to form; thus, once established, there may be reluctance to disband
them. Moreover, as more Wrms pursue alliances, a bandwagon eVect is created
and many Wrms jump on it out of fear of being left without a partner. There are
also challenges and costs associated with managing a partnership. If the relationship
is poorly coordinated, the costs can outweigh the beneWts. Alliances can also
become synonymous with a Wrms values, making them diYcult to discontinue.
Finally, there may be costs associated with closing an alliance. All of these factors
may contribute to alliances existing beyond the period when they create value for
a Wrm.
Several studies have pointed to problems of stagnation that can occur in some
long-term associations. Although this work does not deal directly with rates of
innovation, focusing instead on viability and survival, the general point is apt.
When the participants in a network become too tightly knit and information
circulates only among a small group, networks can become restrictive and ossiWed.
Information that cycles back and forth only among the same participants can lead to
lock-in or sclerosis. When networks turn inward and become restricted in terms of
access to new members, the possibility of group think increases. Grabhers (1993)
study of steelmaking in the Ruhr illustrates how a highly cohesive, homogeneous
region became so overembedded that no producers opted for alternative strategies.
This cognitive lock-in eventually led to the decline of steelmaking in Germany.
Powell (1985) shows how the failure of editors to renew and update their networks
leads to a decline in the quality of a book publishers list and reputation. Portes
and Sensenbrenner (1993) illustrate how ethnic community networks can become
restrictive and subject successful entrepreneurs to ostracism when they deviate
too much from community standards. In sum, the ties that bind economic
actors together can become the ties that blind, thwarting recognition of preferable
alternatives.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 78
78 walter w. powell and stine grodal
3. 7 Summary
........................................................................................................................................................................................
Interorganizational networks have grown considerably in importance over recent
decades. Networks contribute signiWcantly to the innovative capabilities of Wrms by
exposing them to novel sources of ideas, enabling fast access to resources, and
enhancing the transfer of knowledge. Formal collaborations may also allow a
division of innovative labor that makes it possible for Wrms to accomplish goals
they could not pursue alone. Research on alliances documents that investments in
mutual learning and a portfolio of diverse collaborations are associated with in-
creased patenting. While patenting is only an input to the innovation process, the
strength of these empirical results highlights the importance of access to hetero-
geneous contacts.
We have argued that the nature of knowledge, conceptualized in terms of tacitness
or explicitness, is animportant factor indetermining whether members of a network
can eVectively share information and skills. Networks that are rooted in a division of
innovative labor logic may Wnd it easier to transfer tacit knowledge in the form of
Wnished inputs, while networks involved in the co-creation of novel ideas may
succeed or fail on the basis of their ability to convey and transfer ideas that are not
easily codiWed.
Another central challenge to networks of innovators is developing the capacity to
simultaneously enhance the Xowof information among current participants and be
open to new entrants. The twin tests of increasing cohesion within the network and
recognizing promising sources of newideas are diYcult to surmount. Some research
suggests that a mixture of strong and weak ties aVords the proper blend of reliability
and novelty.
Much more research is needed, however, to ascertain how mixtures of thick,
reliable aYliations can be combined with novel linkages to with newcomers. More-
over, since some aYliations are essentially person to person ties, a greater under-
standing is needed of howrelations among individuals are aggregated to productive
relations between corporate actors. Or, in cases where this issue is neglected, as in
studies of contractual relations between Wrms where the informal relational under-
pinnings are not analyzed, the opportunity to examine the intertwining of the
careers of individuals and the strategies of Wrms is missed. Equally important, the
network literature has not focused explicitly on diVerent measures of innovative
output, whether it is newproducts or services, newmodes of organizing production,
or more rapid response to competitive demands. The standard measures are based
on either patents, an input to the innovation process, or problem solving, without
suYcient attention to either the timeliness or the optimality of the solution. The
network literature is still relatively young, however, so it may be premature to expect
such sophisticated answers. We look forward to future research that oVers a more
compelling analysis of the speciWc ways inwhichnetworks shape innovative outputs.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 79
networks of innovators 79
Notes
1. The Wve that formed the core of the human sequencing component were the Whitehead
Institute, Washington University, Baylor College of Medicine, the Joint Genome Institute
(a cluster of three national laboratories at the Department of Energy), and the Sanger
Institute in England.
2. Notable exceptions include Ahuja 2000a, Rowley et al. 2000, Powell et al. 2005.
3. Owen-Smith and Powell (2003) found that US universities with strong ties to a limited set
of commercial partners had fertile patent portfolios, with fertility measured by the
impact of patent citations. Those universities with few ties also had les fertile patents and
patented much less. The optimal strategy for research universities, with respect to patent
volume and impact, appears to be one of diverse ties to a wide array of industrial partners.
Diversity mitigates possible capture from too close relations with commercial Wrms.
4. Inkpen and Ross (2001) do not specify how to measure the appropriate duration of
relationships. Instead, they assume that some relationships become stale and dutiful
over time, and no longer generate beneWts that outweigh the costs of sustaining them.
References
*Ahuja, G. (2000a), Collaboration Networks, Structural Holes, and Innovation: A Longi-
tudinal Study, Administrative Science Quarterly 45: 42555.
(2000b), The Duality of Collaboration: Inducements and Opportunities in the For-
mation of InterWrm Linkages, Strategic Management Journal 21: 31743.
Akera, A. (2001), The IBM User Group, Share, Technology and Culture 42: 71036.
Baum, J. A. C., Calabrese, T., and Silverman, B. S. (2000), Dont Go It Alone: Alliance
Network Composition and Startups Performance in Canadian Biotechnology, Strategic
Management Journal 21: 26794.
Beckman, C., and Haunschild, P. (2002), Network Learning: The EVects of Partners
Heterogeneity of Experience onCorporate Acquisitions, Administrative Science Quarterly
47: 921724.
Blau, P. M. (1963), Consultation Among Colleagues, in Dynamics of Bureaucracy, Chicago:
University of Chicago Press, 15769.
Boisot, M. H. (1998), Knowledge Assets: Securing Competitive Advantage in the Information
Economy, Oxford: Oxford University Press.
Brown, J. S., and Duguid, P. (2001), Knowledge and Organization: A Social Practice
Perspective, Organization Science 12(2): 198213.
Brusco, S. (1982), The EmilianModel: Productive DecentralizationandSocial Integration,
Cambridge Journal of Economics 6: 16784.
*Burt, R. S. (1992), Structural Holes, Cambridge, Mass.: Harvard University Press.
Callon, M. (1998), Actor-Network TheoryThe Market Test, in J. Law and J. Hassard
(eds.), Actor Network Theory and After, Cambridge, Mass.: Blackwell, 18696.
* Asterisked items are suggestions for further reading.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 80
80 walter w. powell and stine grodal
(2002), From Science as an Economic Activity to Socioeconomics of ScientiWc Re-
search, in P. Mirowski and E. M. Sent (eds.), Science Bought and Sold, Chicago: University
of Chicago Press.
Chesbrough, H. (2003), The Era of Open Innovation, Boston: Harvard Business School
Press.
Cohen, S., and Fields, G. (1999), Social Capital and Capital Gains in Silicon Valley,
California Management Review 41(2): 10830.
Cohen, W., and Levinthal, D. (1990), Absorptive Capacity: A New Perspective on Learn-
ing and Innovation, Administrative Science Quarterly 35: 12852.
Cowan, R., David, P. A., and Foray, D. (2000), The Explicit Economics of Knowledge
CodiWcation and Tacitness, Industrial and Corporate Change 9: 21153.
Crane, D. (1972), Invisible Colleges: DiVusion of Knowledge and ScientiWc Communities,
Chicago: University of Chicago Press.
Cross, R., Borgatti, S., and Parker, A. (2002), Making Invisible Work Visible, California
Management Review 44(2): 2546.
Cusumano, M., Mylonadis, Y., and Rosenbloom, R. (1992), Strategic Maneuvering
and Mass-Market Dynamics: The Triumph of VHS over Beta, Business History Review
66: 5194.
Dalton, M. (1959), Power Struggles in the Line, in Men Who Manage, New York: Wiley,
71109.
David, P. A. (2001), Cooperation, Creativity, and Closure in ScientiWc Research Networks:
Modeling the Simpler Dynamics of Epistemic Communities, Conference at Centre Saint-
Gobain, Paris.
*Dyer, J. H. (1996), Specialized Supplier Networks as a Source of Competitive Advantage:
Evidence from the Auto Industry, Strategic Management Journal 17(4): 27191.
and Nobeoka, K. (2000), Creating and Managing a High-Performance Knowledge-
Sharing Network: The Toyota Case, Strategic Management Journal 21: 34567.
Eccles, R. (1981), The QuasiWrm in the Construction Industry, Journal of Economic
Behavior and Organization 2: 33557.
Eisenhardt, K. M., and Schoonhoven, C. B. (1996), Resource-based View of Strategic
Alliance Formation: Strategic and Social EVects in Entrepreneurial Firms, Organization
Science 7: 13650.
Farrell, J., and Salomer, G. (1988), Coordination through Committees and Markets,
RAND Journal of Economics 19(2): 23552.
Fleming, L., and Sorenson, O. (2001), Technology as a Complex Adaptive System:
Evidence from Patent Data, Research Policy 30: 101939.
*Freeman, C. (1991), Networks of Innovators: A Synthesis of Research Issues, Research
Policy 20: 499514.
Galunic, D. C., and Rodan, S. (1998), Resource Recombinations in the Firm: Knowledge
Structures and the Potential for Schumpeterian Innovation, Strategic Management Jour-
nal 19: 1193201.
Ghoshal, S., and Bartlett, C. (1990), The Multinational Corporation as an Interorgani-
zational Network, Academy of Management Review 15: 56185.
Godoe, H. (2000), Innovation Regimes, R&D and Radical Innovations in Telecommuni-
cations, Research Policy 29: 103346.
Gomes-Casseres, B. (1996), The Alliance Revolution: The New Shape of Business Rivalry,
Cambridge, Mass.: Harvard University Press.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 81
networks of innovators 81
Grabher, G. (1993), The Weakness of Strong Ties: The Lock-in of Regional Development in
the Ruhr Area, in G. Grabher (ed.), The Embedded Firm, London: Routledge.
and Powell, W. W. (2004), Introduction, in Critical Studies in Economic Institutions:
Networks, London: Edward Elgar.
*Granovetter, M. (1973), The Strength of Weak Ties, American Journal of Sociology 78:
136080.
(1985), Economic Action and Social Structure: The Problem of Embeddedness,
American Journal of Sociology 91: 481510.
Griliches, Z. (1990), Patent Statistics as Economic Indicators: A Survey, Journal of
Economic Literature 28: 1661707.
Gulati, R. (1995), Does Familiarity Breed Trust? The Implications of Repeated Ties for
Contractual Choice in Alliances, Academy of Management Journal 38: 85112.
Hagedoorn, J. (1995), Strategic Technology Partnering During the 1980s: Trends,
Networks, and Corporate Patterns in Non-Core Technologies, Research Policy 24:
20731.
(1996), Trends and Patterns in Strategic Technology Partnering since the Early Seven-
ties, Review of Industrial Organization 11: 60116.
(2002), Inter-Firm R&D Partnerships: An Overview of Major Trends and Patterns
since 1960, Research Policy 31: 47792.
, and Duysters, G. (2002), External Sources of Innovative Capabilities: The
Preference for Strategic Alliances or Mergers and Acquisitions, Journal of Management
Studies 39(2): 16788.
HA
Corresponding author.
E-mail addresses: ernstd@eastwestcenter.org (D. Ernst),
linsukim@united.co.kr (L. Kim).
with the major markets in North America and Europe
(e.g. Dicken, 1992).
A progressive liberalization and deregulation of
international trade and investment, and the rapid de-
velopment and diffusion of information and commu-
nication technology (IT) have fundamentally changed
the global competitive dynamics, in which MNCs op-
erate. While both market access and cost reductions
remain important, it became clear that they have to
be reconciled with a number of equally important re-
quirements that encompass: the exploitation of uncer-
tainty through improved operational exibility (e.g.
Kogut, 1985; Kogut and Kulatilaka, 1994); a com-
pression of speed-to-market through reduced product
development and product life cycles (e.g. Flaherty,
1986); learning and the acquisition of specialized
external capabilities (e.g. Antonelli, 1992; Zander
0048-7333/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0048- 7333( 02) 00072- 0
1418 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
and Kogut, 1995); and a shift of market penetra-
tion strategies from established to new and unknown
markets (e.g. Christensen, 1997).
In response to the increasingly demanding re-
quirements of global competition, three interrelated
transformations have occurred in the organization
of international economic transactions. First, global
production networks (GPN) have proliferated as a
major organizational innovation in global operations
(e.g. Borrus et al., 2000). Second, these networks
have acted as a catalyst for international knowledge
diffusion, providing new opportunities for local ca-
pability formation in lower-cost locations outside
the industrial heartlands of North America, Western
Europe and Japan. Third, a long-term process of dig-
ital convergence (e.g. Chandler and Cortada, 2000),
enabling the same infrastructure to accommodate
manipulation and transmission of voice, video, and
data, has created new opportunities for organizational
learning and knowledge exchange across organiza-
tional and national boundaries, hence magnifying the
rst two transformations (Ernst, 2002c).
The combination of these three transformations has
changed dramatically the international geography of
production and innovation. We focus on the rst two
of these transformations. The rst transformation sig-
nals a new divide in industrial organization: a transi-
tion is under way from multinational corporations,
with their focus on stand-alone overseas investment
projects, to global network agships that integrate
their dispersed supply, knowledge and customer bases
into global (and regional) production networks (Ernst,
1997, 2002a). There is a growing acceptance in the
literature that, to capture the impact of globalization
on industrial organization and knowledge diffusion,
the focus of research needs to move from the indus-
try and the individual rm to the international dimen-
sion of business networks (e.g. Ghoshal and Bartlett,
1990; Rugman and DCruz, 2000). Our understanding
of these networks is limited. Most studies have focused
too narrowly on the perspective of the network ag-
ship (agship bias). We need research that explores
as well implications for network suppliers, especially
lower-tier suppliers from developing countries.
Equally important is the second transformation:
GPN in their operations reportedly disseminate im-
portant knowledge to local suppliers in low-cost loca-
tions, which could catalyze local capability formation.
Knowledge transfer, however, is not automatic. It re-
quires a signicant level of absorptive capacity on the
part of local suppliers and a complex process to in-
ternalize disseminated knowledge. Our understanding
of knowledge transfer and local capability formation
is limited. International knowledge transfer has been
extensively studied, but research has primarily fo-
cused on such formal mechanisms as foreign direct
investment (FDI) and foreign licensing (FL) (Reddy
and Zhao, 1990). These formal mechanisms, how-
ever, are only the tip of the iceberg. A larger amount
of knowledge is transferred through various informal
mechanisms (Westphal et al., 1985; Kim, 1991; Ernst,
2000a). Research on informal knowledge transfer is
scarce. The importance of local capabilities in assim-
ilating, adapting, and improving imported technology
has long been recognized, but few studies exist on
the complex process of local capability formation in
developing countries (e.g. Kim, 1997).
GPN transform the production and use of knowl-
edge, with far-reaching implications for an evolu-
tionary theory of economic change. There is a fun-
damental trend towards an increasing mobility of
knowledge, yet little do we know about drivers and
implications. A major constraint is a lack of com-
munication between research on GPN, research on
international knowledge diffusion, and research on
local capability formation. While all three are highly
relevant strands of research, their lack of interaction
obstructs our understanding of how global networks
affect knowledge diffusion and the formation of local
capabilities. There is a need to bridge this gap through
appreciative theories, as dened in Nelsons (1995)
thought-provoking review of economic growth theory.
This paper develops a conceptual framework that
links together the above three areas of research, as
a rst step towards an appreciative theory. We argue
that globalization has culminated in an important or-
ganizational innovation: the spread of GPN. These
networks combine concentrated dispersion of the
value chain across rm and national boundaries, with
a parallel process of integration of hierarchical layers
of network participants. This has created new op-
portunities for international knowledge diffusion that
lower-tier network suppliers should strive to exploit.
To substantiate this argument, we proceed as follows.
Section 2 analyzes the three dynamic forces that drive
the rapid development of GPN and highlights the
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1419
characteristics of the agship model of GPN. Section 3
explores the categories of knowledge, and the mecha-
nisms of knowledge transfer from agship companies
to local network suppliers. And in Section 4, we dis-
cuss how GPN can act as mediators of local capability
formation.
2. Driving forces and characteristics of GPN
2.1. Driving forces
What has driven the shift in industrial organiza-
tion from multinational corporations to global
network agships that integrate their dispersed sup-
ply, knowledge and customer bases into global (and
regional) production networks? To answer this ques-
tion, we introduce a stylized model of globalization
drivers, focusing on three inter-related explanatory
variables: institutional change through liberalization,
information technology, and competition.
The rst driving force is liberalization, which in-
cludes four main elements: trade liberalization; liberal-
ization of capital ows; liberalization of FDI policies;
and privatization. While each of these has generated
separate debates in the literature, they hang together.
Earlier success in trade liberalization has sparked an
expansion of trade and FDI, increasing the demand
for cross-border capital ows. This has increased the
pressure for a liberalization of capital markets, forc-
ing more and more countries to open their capital ac-
counts. In turn this has led to a liberalization of FDI
policies, and to privatization tournaments.
The overall effect of liberalization has been a
considerable reduction in the cost and risks of in-
ternational transactions and a massive increase in
international liquidity. Global corporations (the net-
work agships) have been the primary beneciaries:
liberalization provides them with a greater range of
choices for market entry between trade, licensing,
subcontracting, franchising, etc. (locational special-
ization) than otherwise; it provides better access to ex-
ternal resources and capabilities that a agship needs
to complement its core competencies (outsourcing);
and it has reduced the constraints for a geographic
dispersion of the value chain (spatial mobility).
The second important driver of GPN is the
rapid development and diffusion of information and
communication technology (IT). These technolo-
gies have had a dual impact: they increase the need
and create new opportunities for globalization. This
argument is based on two propositions. The rst
proposition is that the cost and risk of developing IT
has been a primary cause for market globalization:
international markets are required to amortize fully
the enormous R&D expenses associated with rapidly
evolving process and product information technolo-
gies (Kobrin, 1997, p. 149). Of equal importance
are the huge expenses for IT-based organizational
innovations. (Brynjolfson and Hitt, 2000; Ernst and
OConnor, 1992). As the extent of a companys R&D
effort is determined by the nature of its technology
and competition rather than its size, this rapid growth
of R&D spending requires a corresponding expan-
sion of sales, if protability is to be maintained. No
national market, not even the US market, is large
enough to amortize such huge expenses.
The second proposition explains why international
production rather than exports have become the main
vehicle for international market share expansion. Of
critical importance has been the enabling role played
by IT: it has substantially increased the mobility, i.e.
dispersion of rm-specic resources and capabilities
across national boundaries; it also provides greater
scope for cross-border linkages, i.e. the integration of
dispersed specialized clusters. This has substantially
reduced the friction of time and space, both with re-
gard to markets and production: a rm can now serve
distant markets equally well as local producers; it can
also now disperse its value chain across national bor-
ders in order to select the most cost-effective location.
In addition, IT and related organizational innova-
tions provide effective mechanisms for constructing
exible infrastructures that can link together and co-
ordinate economic transactions at distant locations
(Antonelli, 1992; Hagstrm, 2000). This has im-
portant implications for organizational choices and
locational strategies of rms. In essence, IT fosters
the development of leaner, meaner and more agile
production systems that cut across rm boundaries
and national borders. The underlying vision is that of
a network of rms that enable a global network ag-
ship to respond quickly to changing circumstances,
even if much of its value chain has been dispersed.
The third driving force is competition. Together
with liberalization, IT has drastically changed the
1420 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
dynamics of competition. Again, we reduce the com-
plexity of these changes and concentrate on two im-
pacts: a broader geographic scope of competition; and
a growing complexity of competitive requirements.
Competition now cuts across national borders, a rms
position in one country is no longer independent from
its position in other countries (e.g. Porter, 1990). This
has two implications. The rm must be present in
all major growth markets (dispersion). It must also
integrate its activities on a worldwide scale, in order
to exploit and coordinate linkages between these dif-
ferent locations (integration). Competition also cuts
across sector boundaries and market segments: mutual
raiding of established market segment efdoms has
become the norm, making it more difcult for rms
to identify market niches and to grow with them.
This growing complexity of competition has
changed the determinants of rm organization and
growth, as well as the determinants of location. No
rm, not even a dominant market leader, can gen-
erate all the different capabilities internally that are
necessary to cope with the requirements of global
competition. Competitive success thus critically de-
pends on a capacity to selectively source specialized
capabilities outside the rm that can range from sim-
ple contract assembly to quite sophisticated design
capabilities. This requires a shift from individual to
increasingly collective forms of organization, from the
multidivisional (M-form) functional hierarchy (e.g.
Chandler, 1977) of multinational corporations to
the networked global agship model (Ernst, 2002c).
Until recently, these fundamental changes in the orga-
nization of international production have been largely
neglected in the literature, both in research on knowl-
edge spill-over through FDI, and in research on the
internationalization of corporate R&D.
2.2. Characteristics of GPN
The concept of a GPN covers both intra-rm and
inter-rm transactions and forms of coordination
(Fig. 1): it links together the agships own subsidia-
ries, afliates and joint ventures with its subcontrac-
tors, suppliers, service providers, as well as partners
in strategic alliances (e.g. Ernst, 1997; 2002b). These
arrangements may, or may not involve ownership of
equity stakes. A network agship like IBM or Intel
breaks down the value chain into a variety of discrete
functions and locates them wherever they can be
carried out most effectively, where they improve the
rms access to resources and capabilities and where
they are needed to facilitate the penetration of impor-
tant growth markets.
The main purpose of these networks is to provide the
agship with quick and low-cost access to resources,
capabilities and knowledge that are complementary to
its core competencies. In other words, transaction cost
savings matter. Yet, the real benets result from the
dissemination, exchange and outsourcing of knowl-
edge and complementary capabilities.
GPN typically combine a breath-taking speed of ge-
ographic dispersion with spatial concentration: much
of the recent cross-border extension of manufacturing
and services has been concentrated on a growing, but
still limited number of specialized lower-cost clusters.
Apart from the usual suspects in Asia (Korea, Tai-
wan, Singapore, China, Malaysia, Thailand, and now
also India), this includes once peripheral locations in
Europe (e.g. Ireland, Central and Eastern Europe and
Russia), Brazil, Mexico, and Argentina in Latin Amer-
ica, some Carribbean locations (like Costa Rica), and
a few spots elsewhere in the so-called RoW (rest of
the world).
The degree of dispersion differs across the value
chain: it increases, the closer one gets to the nal prod-
uct, while dispersion remains concentrated especially
for critical precision components. Concentration of
dispersion increases, the more we move toward more
complex, capital-intensive precision components:
memory devices and displays are sourced primarily
from Japan, Korea, Taiwan and Singapore; and hard
disk drives from a Singapore-centered triangle of lo-
cations in Southeast Asia. Finally, dispersion becomes
most concentrated for high-precision, design-intensive
components that pose the most demanding require-
ments on the mix of capabilities that a rm and its
cluster needs to master: microprocessors for instance
are sourced from a few globally dispersed afliates
of Intel, two secondary American suppliers, and one
recent entrant from Taiwan.
2.2.1. Flagships
GPN typically consist of various hierarchical layers
that range from network agships that dominate such
networks, down to a variety of usually smaller, local
specialized network suppliers. This taxonomy helps to
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1421
Fig. 1. The nodes of a global production network.
assess the different capacities of these rms to benet
from knowledge diffusion and to upgrade local capa-
bility formation.
We distinguish two types of global agships: (i)
brand leaders (BL), like Cisco, GE, IBM, Compaq
or Dell; and (ii) contract manufacturers (CM), like
Solectron or Flextronics, that establish their own GPN
to provide integrated global supply chain services
to the global brand leaders. Cisco is an interest-
ing example of a brand leader: its GPN connects
the agship to 32 manufacturing plants worldwide.
These suppliers are formally independent, but they
go through a lengthy process of certication to en-
sure that they meet Ciscos demanding requirements.
Outsourcing volume manufacturing and related sup-
port services enable brand leaders to combine cost
reduction, product differentiation and time-to-market.
Equally important are nancial considerations:
getting rid of low-margin manufacturing helps the BL
to increase shareholder returns.
Contract manufacturers have rapidly increased in
importance since the mid-1990s. This represents an
acceleration of a long-standing trend towards vertical
specialization that is particularly pronounced in the
electronics industry (Ernst, 2002a). During the 1990s,
global brand leaders have put up for sale a growing
number of their overseas facilities, and in some cases
whole chunks of their GPN. BL from North Amer-
ica like HP, Dell, Compaq, Motorola, Intel, IBM, Lu-
cent, Nortel were rst in pursuing such divestment
strategies. But European BL (e.g., Philips, Ericcson,
Siemens and Nokia) and, more recently, Japanese ones
(e.g., NEC, Fujitsu, Sony) have followed suit. Out-
sourcing based on contract manufacturing became the
panacea of the 1990s (Lakenan et al., 2001, p. 3).
CM have aggressively seized this opportunity: through
1422 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
acquisition and capacity expansion they have devel-
oped, within a few years, their own GPN that now
complement the networks established by the global
brand leaders. This gave rise to an extremely rapid
growth of the CM industry. From 1996 to 2000, capi-
tal expenditures grew 11-fold (50% CAGR), and rev-
enues increased by almost 400% (81% CAGR).
The agship is at the heart of a network: it pro-
vides strategic and organizational leadership beyond
the resources that, from an accounting perspective,
lie directly under its management control (Rugman,
1997, p. 182). The strategy of the agship company
thus directly affects the growth, the strategic direc-
tion and network position of lower-end participants,
like specialized suppliers and subcontractors. The lat-
ter, in turn, have no reciprocal inuence over the
agship strategy (Rugman and DCruz, 2000, p. 84).
The agship derives its strength from its control over
critical resources and capabilities that facilitate inno-
vation, and from its capacity to coordinate transac-
tions and knowledge exchange between the different
network nodes.
2.2.2. Local suppliers
Greatly simplifying, we distinguish two types
of local suppliers: higher-tier lead suppliers and
lower-tier suppliers. Higher-tier suppliers, like for
instance Taiwans Acer group (Ernst, 2000b), play an
intermediary role between global agships and local
suppliers. They deal directly with global agships
(both brand leaders and contract manufacturers);
they possess valuable proprietary assets (including
technology); and they have developed their own
mini-GPN (Chen, in press). With the exception of
hard-core R&D and strategic marketing that remain
under the control of the network agship, the lead
supplier must be able to shoulder all steps in the value
chain. It must even take on the coordination functions
necessary for global supply chain management.
Lower-tier suppliers are in a much more precar-
ious position. Their main competitive advantages are
low cost and speed, and exibility of delivery. They
are typically used as price breakers and capacity
buffers, and can be dropped at short notice. This
second group of local suppliers rarely deals directly
with the global agships; they interact primarily with
local higher-tier suppliers. Lower-tier suppliers nor-
mally lack proprietary assets; their nancial position is
weak; and they are highly vulnerable to abrupt changes
in markets and technology, and to nancial crises.
3. GPN and knowledge diffusion
The agships can exert considerable pressure on
local suppliers, especially in small developing coun-
tries: they can discipline suppliers by threatening to
drop them from the networks whenever they fail to
provide the required services at low price and world
class quality.
At the same time, GPN also act as powerful carriers
of knowledge. First, agships need to transfer techni-
cal and managerial knowledge to the local suppliers.
This is necessary to upgrade the suppliers technical
and managerial skills, so that they can meet the tech-
nical specications of the agships. Second, once a
network supplier successfully upgrades its capabili-
ties, this creates an incentive for agships to transfer
more sophisticated knowledge, including engineer-
ing, product and process development. This reects
the increasingly demanding competitive requirements
that we referred to earlier. In the electronics industry
for instance, product-life-cycles have been cut to six
months, and sometimes less (Ernst, 2002a). Over-
seas production thus frequently occurs soon after the
launching of new products. This is only possible if
agships share key design information more freely
with overseas afliates and suppliers. Speed-to-market
requires that engineers across the different nodes of
a GPN are plugged into the agships design debates
(both on-line and face-to-face) on a regular basis.
Of course, knowledge transfer is not a sufcient
condition for effective knowledge diffusion. Diffusion
is completed only when transferred knowledge is in-
ternalized and translated into the capability of the lo-
cal suppliers (e.g. Kim, 1997, and Ernst et al., 1998).
Much depends on the types of knowledge involved
and the mechanisms that agships use to disseminate
different types of knowledge.
3.1. The categories of knowledge
Knowledge may be classied into various categories
depending on the purpose of its use. Polanyis (1962)
classied knowledge into explicit and tacit knowledge.
Explicit knowledge refers to knowledge that is codied
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1423
in formal, systematic language (encoded knowledge).
It is knowledge that can be combined, stored, retrieved,
and transmitted with relative ease and through various
mechanisms.
But explicit knowledge is useful only when tacit
knowledge enables individuals and organizations to
make sense of and utilize it. Tacit knowledge refers to
knowledge that is so deeply rooted in the human body
and mind that it is hard to codify and communicate. It
is knowledge that can only be expressed through ac-
tion, commitment, and involvement in a specic con-
text and locality. Tacit knowledge is based on experi-
ence: people acquire it through observation, imitation,
and practice. Its diffusion requires apprentice-type
training and face-to-face interaction. It can also be
transferred, however, through the movement of hu-
man carriers of such knowledge, a fact that much of
the literature on industrial districts used to neglect.
Many have attempted to unpackage the blackbox
of tacit knowledge (e.g. Sparrow, 1998; Spender,
1996). For our purpose, the following classication,
rst coined by Collins (1993) and later expanded
by Blackler (1995), appear to be most useful. Tacit
knowledge may become part of the human body as
skills (embodied knowledge); part of human being as
cognitive capacity (embrained knowledge); routinized
in organizational practice (embedded knowledge); and
inculcated in the organization as basic assumptions,
beliefs and norms (encultured knowledge). Different
types of tacit knowledge are associated with different
aspects of organizational activities and with different
degree of difculties in transferring it.
3.2. Knowledge transfer mechanisms
Flagships transfer knowledge across borders
through various mechanisms. First, the transfer may
be mediated through the market, involving a formal
contract for terms and conditions between the knowl-
edge supplier and the knowledge buyer with payment
involved. Knowledge may also be transferred infor-
mally without any payment involved. Second, the
agship may play an active role, exercising signif-
icant control over the way in which knowledge is
disseminated to and used by the local supplier. Alter-
natively, the agship may play a passive role, having
almost nothing to do with the way the local supplier
takes advantage of available knowledge that is either
embodied in or disembodied from the physical items.
These two dimensionsmarket-mediation and the
role of agshipsoffer a useful two-by-two matrix,
as shown in Fig. 2, to identify different mechanisms
of knowledge transfer through GPN (Kim, 1991).
First, network agships use largely formal mecha-
nisms such as FDI, FL, technical consultancies, etc.
in quadrant 1 to transfer knowledge to local suppliers,
if the latter are subsidiaries or joint venture partners.
For instance, when such agships as Intel, Motorola,
Texas Instruments, and Fairchild decided to outsource
assembly operations of their semiconductor devices,
they took the mechanisms of FDI, FL, and techni-
cal consultancies to establish their subsidiaries in the
Philippines (Antonio, 2001) and other countries in
Southeast Asia. They insisted on majority ownership
in the subsidiaries, licensed and transferred a complete
production system.
Second, independent local suppliers rely heavily
on standard machinery in quadrant 2 to improve their
productivity in production operations. Machinery is
a major source of process innovation for their users
(Abernathy and Townsend, 1975). Flagships are not
necessarily the suppliers of the machinery, but they
can play an important indirect role, by forcing inde-
pendent local suppliers to purchase more sophis-
ticated equipment to improve their production
capabilities. For instance, Mando, one of the major
auto components suppliers from Korea, purchased
a series of robots to automate their production pro-
cesses. Each of the robots embodied state-of-the-art
production knowledge. The equipment suppliers,
however, had little inuence over the way Mando em-
ployees assimilate knowledge by using these robots.
Third, a more direct way for agships to transfer
knowledge to independent local suppliers are infor-
mal mechanisms in quadrant 3, largely through the
original equipment manufacturing (OEM) arrange-
ments. As in the quadrant 1, agships actively transfer
knowledge in the form of blue prints, technical spec-
ications, and technical assistance, mostly free of
charge, to independent local suppliers to ensure that
products and services produced by the latter meet the
formers technical specications. For instance, Boeing
outsources some parts of fuselage from independent
local suppliers in Japan, Taiwan, and Korea. In doing
so, Boeing actively provides the local suppliers with
1424 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
Fig. 2. Knowledge transfer mechanisms.
technical literature, product specications, and tech-
nical assistance to help them meet its specications.
Fourth, independent local suppliers can also rely on
knowledge transfer mechanisms in quadrant 4. Like in
quadrant 2, agships exert little direct inuence over
the way independent local suppliers use such mech-
anisms as reverse engineering, observations, and hu-
man mobility to expedite upgrading their capabilities.
In Korea, for instance, the Small Industry Promotion
Corporation and industry-related SME associations
frequently organizes observation tour of foreign rms
as a way to acquire new knowledge. Human mobil-
ity in quadrant 4 includes not only the repatriation
of top-rated engineers trained abroad (Saxenian, in
press) but also the active use of experienced foreign
engineers who are hired for short periods as so-called
moonlighters.
To what degree do the agships use the knowl-
edge transfer mechanisms? The shift from MNCs to
global network agships has expanded both the mech-
anisms and the volume of knowledge transfer. MNCs
relied heavily on the mechanisms in quadrant 1 of
Fig. 2 in setting up their plants either for the penetra-
tion of protected markets or for exploiting differential
factor costs. In contrast, agships transfer knowledge
not only through mechanisms in quadrant 1 but also
through mechanisms in quadrant 3. Flagships also tend
to transfer more knowledge to local suppliers than ver-
tically integrated MNCs. These transfers are neces-
sary to enable local suppliers to provide the agship
with competitive products and services, in line with
the changing requirements of markets and technology.
4. Local capability formation
Local suppliers can only effectively absorb knowl-
edge disseminated by global network agships, if
they have developed their own capabilities. Knowl-
edge internalization and capability building require
individual and organizational learning. Individuals are
the primary actors in learning and knowledge creation
(Hedberg, 1981). They constitute local capabilities
that may be combined at the organizational level. Or-
ganizational learning, however, is not the simple sum
of individual learning. Only effective organizations
can translate individual learning and capabilities into
organizational learning and capabilities.
4.1. Concepts
Firms create knowledge primarily through the dy-
namic process of conversion between explicit and
tacit knowledge (Nonaka, 1991). Tacit-to-tacit con-
version (called socialization) takes place when tacit
knowledge of one individual is shared with others
through training, whereas explicit-to-explicit conver-
sion (combination) takes place when an individual or
a group combines discrete pieces of explicit knowl-
edge into a new whole. Tacit-to-explicit conversion
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1425
(externalization) occurs when an individual or a group
is able to articulate the foundations of individual tacit
knowledge. Finally, explicit-to-tacit conversion (inter-
nalization) takes place when new explicit knowledge
is shared throughout the rm and other members
begin to use it to broaden, extend, and reframe their
own tacit knowledge. Such conversion tends to be-
come faster in speed and larger in scale in a spiral
process, as more actors in and around the rms be-
come involved in knowledge conversion (Nonaka and
Takeuchi, 1995). For effective knowledge conversion
to lead to productive learning, two important elements
are required: an existing knowledge base or compe-
tence (most of it tacit knowledge), and the intensity
of effort or commitment. Cohen and Lavinthal (1990)
call this absorptive capacity. How fast and success-
fully the local suppliers internalize and translate trans-
ferred knowledge into their own capability through
learning will be largely determined by their absorptive
capacity and their ability to upgrade it continuously.
Tacit knowledge enables the individual as well as
the organization to use both explicit and tacit knowl-
edge available elsewhere and to create new knowl-
edge through various knowledge conversion activities
in production and R&D. The intensity of effort, on
the other hand, determines the speed of knowledge
conversion. It represents the amount of emotional,
intellectual, and physical energy that members of
an organization invest in acquiring and converting
knowledge. Exposure of individuals and rms to rel-
evant external knowledge is insufcient, unless they
make a conscious effort to internalize and use it.
Hence, considerable time and effort must be directed
to learning (Kim, 1998). Of the two, the intensity
of effort or commitment is more important than the
knowledge base, as the former creates the latter, but
not vice versa (Ullrich, 1998).
4.2. GPN as mediators of local capability
formation
Let us rst look at explicit knowledge. Flagships
typically provide the local suppliers with encoded
knowledge, such as machinery that embodies new
knowledge, blueprints, production and quality con-
trol manuals, product and service specications, and
training handouts. This is done to assist the suppliers
in building capabilities that are necessary to produce
products and services with the expected quality and
price. Personnel at the local suppliers read and try
to assimilate the transferred explicit knowledge into
their tacit knowledge (internalization in Fig. 3). In
most cases, the acquisition of explicit knowledge
alone is not sufcient for the local suppliers to as-
similate and use it in production, as the translation
of explicit knowledge into actual operations requires
a signicant amount of tacit knowledge. Thus, to
augment the explicit knowledge, network agships
typically invite engineers and managers of the local
suppliers to its best-practice plants, so that they can
observe how actual production systems work and to
receive a systematic training.
This can help to translate knowledge gained from
the literature into actual operations (internalization).
It also enables local engineers to internalize how the
agships organization and production systems are
managed (internalization of embedded knowledge),
and to absorb tacit knowledge directly transferred
from foreign engineers through training (socializa-
tion). Once they return home, however, these engi-
neers confront various unforeseen problems in their
attempts to translate what they have learned at the
agships into the operational systems that exist at
home. For this reason, the agships also send their
own engineers (embodied and embrained knowledge)
to help local engineers debug problems in engineering
and manufacturing systems (socialization).
Take the case of subsidiaries or joint ventures in
quadrant 1 of Fig. 2. When Sony established Hwashin
Electronics Company in Korea as a joint venture
to outsource its consumer electronics products, it
supplied not only machinery and equipment for the
mass-production system of its joint venture partner.
Sony also provided blue prints of products, product
specications, and production and quality control
manuals (encoded knowledge). In addition, the ag-
ship invited a number of Korean engineers, techni-
cians, and managers to undergo training at Sonys
plant in Japan on production, organization, and human
resource management (HRM), transferring embed-
ded and encultured knowledge. Sony also dispatched
a number of engineers and technicians to Korea to
help Korean engineers debug problems encountered
in operating and maintaining the production system
and controlling the quality of products to ensure that
Hwashin meet the technical specications of Sonys
1426 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
Fig. 3. The process of local capability formation.
products (embodied and embrained knowledge). Sony
had done these knowledge transfer activities formally
as part of its FDI and FL to Hwashin.
In the case of independent local suppliers in quad-
rant 3 of Fig. 2, when General Electric decided to
outsource its microwave ovens from Samsung under
the OEM arrangements, it sent its engineers to Sam-
sung to explain its technical specications (encoded
knowledge) and taught Samsung engineers master the
engineering details of the product (embrained knowl-
edge), (Magaziner and Patinkin, 1989). GE had done
all these knowledge transfer activities free of charge
to ensure that Samsungs products meet GEs technical
specications.
Second, local suppliers may attempt to translate
such explicit knowledge as production and quality
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1427
control manuals, HRM handbooks, and other litera-
ture transferred from agships into their own produc-
tion and quality control manuals and HRMhandbooks,
which may be more compatible with local institutions
and business behavior. Then a combination takes place
froma set of explicit knowledge to a newset of explicit
knowledge at the local suppliers. In this process, ex-
ternalization of knowledge also takes place from tacit
knowledge of local engineers and managers to explicit
knowledge in the form a new set of manuals and hand-
books. For instance, when Volvo took over the owner-
ship of Samsungs heavy machinery division after the
Asian crisis to turn it into its Asian supplier, Volvo in-
troduced its own management systems, which reects
both Volvos requirements and those shaped by local
institutions. In developing a new set of manuals and
handbooks, the ground was laid for internalization,
combination and externalization.
Third, the link with GPN also induces knowledge
conversions within local suppliers. The key is the
diffusion of locialized and internalized knowledge ac-
cumulated by a limited number of engineers and man-
agers of the local suppliers through training provided
by the network agship. This knowledge needs to
be diffused within local suppliers through spiral pro-
cesses of socialization, as more actors in and around
the rms get involved in knowledge conversion ac-
tivities. Externalization and internalization take place
internally, as actors convert from/to explicit to/from
tacit knowledge within the local supplying rms,
gradually developing embedded knowledge. For in-
stance, Samsung Electronics recently sent a group of
HRM specialists to GE to learn the latter HRM sys-
tem. Upon return, these specialists have conducted a
series of seminars for HRM specialists in the rm to
share the knowledge, leading to the development of
new HRM policy and procedures and to the gradual
formation of new embedded knowledge.
Fourth, the effectiveness and speed of knowledge
conversion process will be determined not so much by
quantity and quality of the knowledge transferred by
the agships as by the absorptive capacity of the lo-
cal suppliers. This holds true regardless of the knowl-
edge transfer mechanisms. Tacit knowledge plays a
central role in this process. This is true even for the
conversion from explicit knowledge to explicit knowl-
edge. Once again, this highlights how important it is
for local suppliers to develop a rich tacit knowledge
base. The strength of the domestic knowledge base
determines the level of sophistication of the converted
knowledge, while the intensity of effort accelerates the
speed of the conversion processes. In turn, spiral pro-
cesses of knowledge conversion determine the level of
the companys internal knowledge base. Leading local
suppliers thus invest heavily in recruiting the cream of
the crop from universities; they also develop intensive
training programs to upgrade the existing knowledge
base.
5. Conclusions
Liberalization, digital convergence, and intensify-
ing global competition have produced a major organi-
zational innovation: a transition from multinational
corporations that exploit labor cost differentials in
different countries to global network agships that
integrate their dispersed supply, knowledge, and cus-
tomer bases into global (or regional) production net-
works. The paper demonstrates that these networks
have boosted international knowledge diffusion, pro-
viding new opportunities for capability formation by
local suppliers in developing countries. Under pres-
sure from agships, local suppliers have a strong in-
centive to internalize transferred knowledge through
various forms of knowledge conversion. The base-
line, however, is the absorptive capacity of the local
suppliers: it determines the effectiveness of capability
formation.
There is nothing automatic about these processes.
Local suppliers need to take an active approach to
maximize their benets from network participation.
Flagships place business orders and transfer valuable
knowledge to local suppliers with only one objective
in mind: to strengthen the competitiveness of their
GPN. In response to intensifying global competition,
the agships outsourcing requirements have become
more demanding. Typically, suppliers are selected by
three criteria: a solid nancial standing; high ratings
on a quarterly scoreboard measuring performance in
delivery, quality etc. and speed of response. The latter
is of critical importance: suppliers are expected to re-
spond within hours with a price, a delivery time, and
a record on their recent performance on reliability and
product quality. This implies that local suppliers can
only upgrade or perish.
1428 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
To stay on the GPN, local suppliers must constantly
upgrade their absorptive capacity. Local suppliers,
therefore, must tap, develop, and retain highly skilled
human resources. Adequate incentives are required
to sustain a high intensity of effort. Ruthless global
competition implies that typically only a limited num-
ber of higher-tier local suppliers is strong enough to
cope with these challenges. But it is unclear what will
happen to the great majority of local rms that are
outside this charmed circle of higher-tier suppliers.
In short, participation in GPN is no substitute for
domestic upgrading efforts. Without the latter, net-
work integration of some higher-tier suppliers may
well increase the divide between rms and districts
that have and those that do not have access to the in-
formation and knowledge that is necessary to reap the
benets of network participation. Many people are un-
derstandably concerned that this may lead to a decline
in economic growth and welfare. This paper demon-
strates that there is cause for cautious optimism: net-
work participation may provide new opportunities for
effective knowledge diffusion to local rms and in-
dustrial districts in developing countries, provided ap-
propriate policies and support institutions are in place
that enable local suppliers to exploit the opportunities
and pressures that result from network participation.
The nature of these policies and institutions will be
the subject of a separate article.
References
Abernathy, W.J., Townsend, P.L., 1975. Technology, productivity,
and process change. Technological Forecasting and Social
Change 7, 377396.
Antonelli, C., 1992. The Economics of Information Networks.
Elsevier, Amsterdam.
Antonio, E.T., 2001. Country Sector Study: Philippines, a paper
presented at the Progress Review Meeting of the International
Competitiveness of Asian Economies: A Cross-Country Study,
Asian Development Bank, Manila.
Blackler, F., 1995. Knowledge, knowledge work, and organizations:
an overview and interpretation. Organization Studies 16, 121
146.
Borrus, M., Ernst, D., Haggard, S., 2000. (Eds.) International
Production Networks in Asia. Rivalry or Riches? Routledge,
London.
Brynjolfson, E., Hitt, L.M., 2000. Beyond Computation:
Information Technology, Organizational Transformations and
Business Performance, A Manuscript, Sloan School of
Management, MIT, Cambridge, MA.
Chandler, A.D., 1977. The Visible Hand: The managerial
Revolution in American Business, Harvard University Press,
Cambridge, MA.
Chandler, A.D., Cortada, J.W., 2000. The information age:
continuities and differences. In: Chandler, A.D., Cortada,
J.W. (Eds.), A Nation Transformed by Information. Oxford
University Press, New York. pp. 281300.
Chen, S.H. Global production networks and information
technology: the case of Taiwan. Industry and Innovation 9, in
press.
Christensen, C.M., 1997. The Innovators Dilemma: When New
Technologies Cause Great Firms to Fail. Harvard Business
School Press, Boston.
Cohen, W.M., Lavinthal, D.A., 1990. Absorptive capacity: a new
perspective on learning and innovation. Administrative Science
Quarterly 35, 128152.
Collins, H., 1993. Structure of knowledge. Social Research 60,
95116.
Dicken, P., 1992. Global Shift: The Internationalization of
Economic Activity. Paul Chapman, London.
Dunning, J., 1981. International Production and the Multinational
Enterprise. George Allen & Unwn, London.
Ernst, D., 1997. From Partial to Systemic Globalization.
International Production Networks in the Electronics Industry,
The Data Storage Industry Globalization Project Report
19972002, Graduate School of International Relations and
Pacic Studies, University of California at San Diego.
Ernst, D., 2000a. Inter-organizational knowledge outsourcing: what
permits small taiwanese rms to compete in the computer
industry. Asia Pacic Journal of Management 17, 223255.
Ernst, D., 2000b. Placing the Networks on the Internet: Challenges
and Opportunities for Managing in Developing Asia. In:
Lundvall, B.A., Smith, K. (Eds.), Proceedings of the Paper
Presented at the Second Asia Academy of Management
Conference, 1518 December 2000, Knowledge Creation in the
Learning Economy, Edward Elgar, Cheltenham.
Ernst, D., 2002a. The Economics of Electronics Industry:
Competitive Dynamics and Industrial Organization. In: Warner,
M., Lazonick, W. (Eds.), The International Encyclopedia of
Business and Management (IEBM), International Thomson
Business Press, London, in press.
Ernst, D., 2002b. Global production networks and the changing
geography of innovation systems. implications for developing
countries. Journal of the Economics of Innovation and New
Technologies 12, 127.
Ernst, D., 2002c. The new mobility of knowledge: digital
information systems and global agship networks. In: Latham,
R., Sassen, S. (Eds.), Cooperation and Conict in a Connected
World. Routledge, London, in press.
Ernst, D., OConnor, D., 1992. Technology and Global
Competition: The Challenge for Newly Industrializing
Economies. OECD Development Centre Studies, Paris.
Ernst, D., Ganiatsos, T., Mytelka, L. 1998. Technological
Capabilities and Export Success: Lessons from East Asia.
Routledge, London.
Flaherty, T., 1986. Coordinating International Manufacturing and
Technology. In: Porter, M. (Ed.), Competition in Global
Industries. Harvard Business School Press, Boston, pp. 83110.
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1429
Ghoshal, S., Bartlett, C.A., 1990. The multinational corporation
as an interorganizational network. Academy of Management
Review l15, 603625.
Hagstrm, P., 2000. New wine in old bottles: information
technology evolution in rm strategy and structure. In:
Birkinshaw, J., Hagstrm, P. (Eds.), The Flexible Firm:
Capability Management in Network Organizations. Oxford
University Press, Oxford, pp. 194206.
Hedberg, B., 1981. How organizations learn and unlearn. In:
Nystrom, P.C., Starbuck, W.H. (Eds.), Handbook of Organi-
zational Design: Adapting Organizations to Their Environments,
Vol. 1. Oxford University Press, New York, pp. 326.
Kim, L., 1991. Pros and cons of international technology transfer:
a developing country view. In: Agmon, T., von Glinow, M.A.
(Eds.), Technology Transfer in International Business. Oxford
University Press. New York, pp. 223239.
Kim, L., 1997. Imitation to Innovation: The Dynamics of
Koreas Technological Learning. Harvard Business School
Press, Boston.
Kim, L., 1998. Crisis construction and organizational learning:
dynamics of capability building in catching-up at hyundai motor.
Organization Science 9, 506521.
Kogut, B. 1985. Designing Global Strategies: Proting from Opera-
tional Flexibility. Sloan Management Review, Fall, pp. 2739.
Kogut, B., Kulatilaka, N., 1994. Operating exibility, global
manufacturing, and the option value of a multinational network.
Management Science 40, 123139.
Kobrin, S.J., 1997. The architecture of globalization: state
sovereignty in a networked global economy. In: Dunning, J.H.
(Ed.), Governments, Globalization and International Business,
Oxford University Press, London, pp. 146172.
Lakenan, B., Boyd, D., Frey, E., 2001. Outsourcing and its perils.
Strategy + Business 24, 112.
Magaziner, I., Patinkin, M., 1989. Fast Heat: How Korea Won the
Microwave War. Harvard Business Review, January, pp. 8393.
Nelson, R.R., 1995. The Agenda for Growth Theory: A Different
Point of View. IIASA Working Paper.
Nonaka, I. 1991. The Knowledge-Creating Company. Harvard
Business Review, NovemberDecember, pp. 96104.
Nonaka, I., Takeuchi, H., 1995. The Knowledge Creating
Company: How Japanese Companies Creates the Dynamics of
Innovation. Oxford University Press, New York.
Polanyi, M., 1962. Personal Knowledge: Towards a Post-Critical
Philosophy. University of Chicago Press, Chicago.
Porter, M., 1990. The Competitive Advantage of Nations.
Macmillan, London.
Reddy, N.M., Zhao, L., 1990. International technology transfer: a
review. Research Policy 19, 285307.
Rugman, A.M., 1997. Canada, In: Dunning, J.H. (Ed.).
Governments, Globalization And International Business. Oxford
University Press, London, pp. 175202.
Rugman, A.M., DCruz, J.R., 2000. Multinationals as Flagship
Firms, Regional Business Networks. Oxford University Press,
Oxford.
Saxenian, A. Transnational communities and the evolution of
global production networks: the case of Taiwan, China and
India. Industry and Innovation 9, in press.
Sparrow, J. 1998. Knowledge in Organizations: Access to Thinking
at Work. Sage Press, London.
Spender, J.C., 1996. Making knowledge the basis of a dynamic
theory of the rm. Strategic Management Journal 17, 45
62.
Ullrich, D. 1998. Intellectual capital = competence commitment.
Sloan Management Review, Winter, pp. 1526.
Westphal, L., Kim, L., Dahlman, C., 1985. Reections on
the Republic of Koreas Acquisition of Technological
Capability. In: Rosenberg, N., Frischtak, C. (Eds.), International
Technology Transfer: Concepts, Measures, and Comparisons.
Praeger, New York, pp. 167221.
Wilkins, M., 1970. The Emergence of Multinational Enterprise.
Harvard University Press, Cambridge, MA.
Zander, U., Kogut, B., 1995. Knowledge and the speed of
the transfer and imitation of organizational capabilities: an
empirical test. Organizational Science 6, 7692.
Research Policy 31 (2002) 569588
Market versus technology drive in R&D internationalization:
four different patterns of managing research and development
Maximilian von Zedtwitz
a,
, Oliver Gassmann
b
a
International Institute for Management Development (IMD), P.O. Box 915, CH-1001 Lausanne, Switzerland
b
R&D Technology Management, Schindler Corporation, CH-6031 Ebikon, Switzerland
Received 13 January 2000; received in revised form 2 October 2000; accepted 24 April 2001
Abstract
Research and development are subject to different location drivers. The analysis of 1021 R&D units, each distinguished by
its main orientation towards either research or development work, reveals that research is concentrated in only ve regions
worldwide, while development is more globally dispersed.
Our research is based on 290 research interviews and database research in 81 technology-intensive multinational companies.
We identify two principal location rationalesaccess to markets and access to scienceas the principal determinants for four
trends that lead to four archetypes of R&D internationalization: national treasure, market-driven, technology-driven,
and global. Their organizational evolution is characterized by four trends. The model is illustrated with short cases of
international R&D organization at Kubota, Schindler, Xerox, and Glaxo-Wellcome.
Differences in R&Dinternationalization drivers lead to a separation of individual R&Dunits by geography and organization.
Current belief is to integrate R&D processes; separation seems to contradict this trend. We argue that this need not be the
case, for there are good reasons to maintain some independence between research and development. 2002 Elsevier Science
B.V. All rights reserved.
Keywords: Globalization; Research and development; Archetypes; Trends; Locations
1. Introduction
1.1. Difcult internationalization of R&D
The internationalization of industrial R&D in
recent years has been interpreted as the attempt of
technology-intensive companies to exploit location-
specic innovation advantages in order to compete in
an increasingly globalized environment. Our empirical
investigation conrms that in most multinational com-
panies, globalization of R&D is typically accepted
Tanaka Business School, Imperial College London, South Kensington Campus, London SW7 2AZ, UK
Received 1 December 2004; accepted 1 May 2005
Available online 30 August 2005
Abstract
This paper examines the roles of international assignments and other forms of researcher mobility inside the integrated R&D
networks of six of the largest European pharmaceutical companies. From in-depth interviews with R&D managers and scientists
it is found that, while the use of international assignments is both limited and often not aimed specically at the transfer of
knowledge, other forms of short-term mobility are widely employed. But what is transferred through such short-term visits is
narrow in focus and often related to specic projects whereas international assignments enable the transfer of broader and more
complex knowledge. The potential impact of these changes in the inter-unit mobility patterns of researchers may be to limit both
the exploitation of potential synergies across disciplines and the creation of enduring and strong personal ties among researchers
working in distant locations.
2005 Elsevier B.V. All rights reserved.
Keywords: Researcher mobility; R&D internationalisation; Multinational companies
1. Introduction
In recent years, considerable evidence has been
gathered that points to the increasing internationali-
sation of R&D activities by multinational enterprises
(MNEs) (see, for example, Dalton et al., 1999; Le Bas
and Sierra, 2002). As shown by a number of studies
(Chiesa, 1996b; Gassmann and von Zedtwitz, 1999)
this growth in R&D activities performed abroad is
being accompanied by a transformation in the way
a
m
m
a
t
i
o
n
Z
e
n
e
c
a
C
h
a
r
n
w
o
o
d
,
U
K
R
e
s
p
i
r
a
t
o
r
y
,
i
n
a
m
m
a
t
i
o
n
A
s
t
r
a
M
o
l
n
d
a
l
,
S
w
e
d
e
n
N
e
u
r
o
l
o
g
y
,
r
e
s
p
i
r
a
t
o
r
y
d
i
s
e
a
s
e
s
,
i
n
a
m
m
a
t
i
o
n
A
s
t
r
a
G
o
t
h
e
n
b
u
r
g
,
S
w
e
d
e
n
C
a
r
d
i
o
v
a
s
c
u
l
a
r
,
g
a
s
t
r
o
i
n
t
e
s
t
i
n
a
l
A
s
t
r
a
S
o
d
e
r
t
a
l
j
e
,
S
w
e
d
e
n
P
a
i
n
c
o
n
t
r
o
l
,
c
e
n
t
r
a
l
n
e
r
v
o
u
s
s
y
s
t
e
m
A
s
t
r
a
L
u
n
d
,
S
w
e
d
e
n
R
e
s
p
i
r
a
t
o
r
y
Z
e
n
e
c
a
W
i
l
m
i
n
g
t
o
n
,
D
e
l
a
w
a
r
e
,
U
S
A
C
e
n
t
r
a
l
n
e
r
v
o
u
s
s
y
s
t
e
m
N
e
w
B
o
s
t
o
n
,
U
S
A
O
n
c
o
l
o
g
y
,
i
n
f
e
c
t
i
o
n
A
v
e
n
t
i
s
a
H
o
e
c
h
s
t
B
r
i
d
g
e
w
a
t
e
r
,
N
e
w
J
e
r
s
e
y
,
U
S
A
R
e
s
p
i
r
a
t
o
r
y
d
i
s
e
a
s
e
s
,
r
h
e
u
m
a
t
o
i
d
a
r
t
h
r
i
t
i
s
,
c
e
n
t
r
a
l
n
e
r
v
o
u
s
s
y
s
t
e
m
,
i
m
m
u
n
o
l
o
g
y
R
h
o
n
e
-
P
o
u
l
e
n
c
P
a
r
i
s
,
F
r
a
n
c
e
A
l
z
h
e
i
m
e
r
s
a
n
d
P
a
r
k
i
n
s
o
n
s
d
i
s
e
a
s
e
s
,
i
n
f
e
c
t
i
o
u
s
d
i
s
e
a
s
e
s
,
o
n
c
o
l
o
g
y
H
o
e
c
h
s
t
F
r
a
n
k
f
u
r
t
,
G
e
r
m
a
n
y
C
a
r
d
i
o
v
a
s
c
u
l
a
r
d
i
s
e
a
s
e
s
,
m
e
t
a
b
o
l
i
c
d
i
s
e
a
s
e
s
,
o
s
t
e
o
-
a
r
t
h
r
i
t
i
s
G
S
K
S
m
i
t
h
K
l
i
n
e
B
e
e
c
h
a
m
U
p
p
e
r
M
e
r
i
o
n
,
P
h
i
l
a
d
e
l
p
h
i
a
,
U
S
A
C
a
r
d
i
o
v
a
s
c
u
l
a
r
,
u
r
o
g
e
n
i
t
a
l
,
m
i
c
r
o
b
i
a
l
,
o
n
c
o
l
o
g
y
G
l
a
x
o
W
e
l
c
o
m
e
R
e
s
e
a
r
c
h
T
r
i
a
n
g
l
e
P
a
r
k
,
N
o
r
t
h
C
a
r
o
l
i
n
a
,
U
S
A
M
e
t
a
b
o
l
i
c
a
n
d
a
n
t
i
v
i
r
a
l
G
l
a
x
o
W
e
l
c
o
m
e
S
t
e
v
e
n
a
g
e
,
U
K
N
e
u
r
o
l
o
g
y
,
r
e
s
p
i
r
a
t
o
r
y
d
i
s
e
a
s
e
s
,
i
n
a
m
m
a
t
i
o
n
G
l
a
x
o
W
e
l
c
o
m
e
V
e
r
o
n
a
,
I
t
a
l
y
P
s
y
c
h
i
a
t
r
y
,
n
e
u
r
o
l
o
g
y
N
o
v
a
r
t
i
s
S
a
n
d
o
z
T
s
u
k
u
b
a
,
J
a
p
a
n
O
n
c
o
l
o
g
y
,
a
r
t
h
r
i
t
i
s
N
e
w
C
a
m
b
r
i
d
g
e
,
M
a
s
s
a
c
h
u
s
e
t
t
s
,
U
S
A
C
a
r
d
i
o
v
a
s
c
u
l
a
r
d
i
s
e
a
s
e
s
,
m
e
t
a
b
o
l
i
s
m
,
i
n
f
e
c
t
i
o
u
s
d
i
s
e
a
s
e
s
S
a
n
d
o
z
(
s
i
n
c
e
1
9
6
4
)
E
a
s
t
H
a
n
o
v
e
r
,
N
e
w
J
e
r
s
e
y
,
U
S
A
O
n
c
o
l
o
g
y
,
a
r
t
h
r
i
t
i
s
,
f
u
n
c
t
i
o
n
a
l
g
e
n
o
m
i
c
S
a
n
d
o
z
V
i
e
n
n
a
,
A
u
s
t
r
i
a
D
e
r
m
a
t
o
l
o
g
y
,
i
m
m
u
n
o
l
o
g
y
S
a
n
d
o
z
a
n
d
C
i
b
a
-
G
e
i
g
y
B
a
s
e
l
,
S
w
i
t
z
e
r
l
a
n
d
N
e
r
v
o
u
s
s
y
s
t
e
m
,
t
r
a
n
s
p
l
a
n
t
a
t
i
o
n
,
o
n
c
o
l
o
g
y
,
a
r
t
h
r
i
t
i
s
/
b
o
n
e
,
f
u
n
c
t
i
o
n
a
l
g
e
n
o
m
i
c
,
o
p
h
t
h
a
l
m
i
c
C
i
b
a
-
G
e
i
g
y
H
o
r
s
h
a
m
,
U
K
R
e
s
p
i
r
a
t
o
r
y
,
c
h
r
o
n
i
c
p
a
i
n
R
o
c
h
e
b
A
c
q
u
i
r
e
d
i
n
1
9
9
4
P
a
l
o
A
l
t
o
,
C
a
l
i
f
o
r
n
i
a
,
U
S
A
C
e
n
t
r
a
l
n
e
r
v
o
u
s
s
y
s
t
e
m
,
i
n
a
m
m
a
t
o
r
y
d
i
s
e
a
s
e
s
/
b
o
n
e
,
g
e
n
i
t
o
u
r
i
n
a
r
y
d
i
s
e
a
s
e
s
,
v
i
r
a
l
d
i
s
e
a
s
e
s
N
u
t
l
e
y
,
N
e
w
J
e
r
s
e
y
,
U
S
A
M
e
t
a
b
o
l
i
c
d
i
s
o
r
d
e
r
s
,
o
n
c
o
l
o
g
y
,
v
a
s
c
u
l
a
r
d
i
s
e
a
s
e
s
B
a
s
e
l
,
S
w
i
t
z
e
r
l
a
n
d
M
e
t
a
b
o
l
i
c
d
i
s
o
r
d
e
r
s
,
c
e
n
t
r
a
l
n
e
r
v
o
u
s
s
y
s
t
e
m
,
v
a
s
c
u
l
a
r
d
i
s
e
a
s
e
s
P
e
n
z
b
e
r
g
,
G
e
r
m
a
n
y
O
n
c
o
l
o
g
y
S
c
h
e
r
i
n
g
R
i
c
h
m
o
n
d
,
C
a
l
i
f
o
r
n
i
a
,
U
S
A
N
e
u
r
o
l
o
g
y
,
i
m
m
u
n
o
l
o
g
y
,
c
a
r
d
i
o
v
a
s
c
u
l
a
r
,
o
n
c
o
l
o
g
y
B
e
r
l
i
n
,
G
e
r
m
a
n
y
I
n
v
i
v
o
d
i
a
g
n
o
s
t
i
c
,
r
a
d
i
o
p
h
a
r
m
a
c
e
u
t
i
c
a
l
s
,
n
e
u
r
o
l
o
g
y
,
i
m
m
u
n
o
l
o
g
y
,
c
a
r
d
i
o
v
a
s
c
u
l
a
r
,
d
e
r
m
a
t
o
l
o
g
y
,
g
e
n
d
e
r
h
e
a
l
t
h
c
a
r
e
,
o
n
c
o
l
o
g
y
A
c
q
u
i
r
e
d
i
n
2
0
0
0
M
o
b
a
r
a
,
J
a
p
a
n
O
n
c
o
l
o
g
y
,
n
e
u
r
o
l
o
g
y
a
A
t
t
h
e
t
i
m
e
o
f
t
h
e
i
n
t
e
r
v
i
e
w
s
A
v
e
n
t
i
s
a
n
d
S
a
n
o
w
e
r
e
s
t
i
l
l
t
w
o
s
e
p
a
r
a
t
e
c
o
m
p
a
n
i
e
s
.
b
A
l
t
h
o
u
g
h
R
o
c
h
e
h
a
s
n
a
n
c
i
a
l
c
o
n
t
r
o
l
o
v
e
r
G
e
n
e
n
t
e
c
h
,
t
h
i
s
c
o
m
p
a
n
y
c
a
n
n
o
t
b
e
c
o
n
s
i
d
e
r
e
d
a
s
a
R
o
c
h
e
s
u
b
s
i
d
i
a
r
y
.
S
o
f
a
r
,
R
o
c
h
e
h
a
s
a
n
o
p
t
-
i
n
r
i
g
h
t
f
o
r
c
o
-
d
e
v
e
l
o
p
i
n
g
c
o
m
p
o
u
n
d
s
d
i
s
c
o
v
e
r
e
d
b
y
G
e
n
e
n
t
e
c
h
f
r
o
m
p
h
a
s
e
2
o
r
3
o
f
c
l
i
n
i
c
a
l
d
e
v
e
l
o
p
m
e
n
t
.
1358 P. Criscuolo / Research Policy 34 (2005) 13501365
R&D facility engaged in activities which range from
earlydiscoverytolife-cycle management ina particular
therapeutic area. More than one R&D unit is normally
involved in different phases of the drug discovery pro-
cess supported by centres of excellence in enabling
technologies.
Usingthe taxonomyproposedbyGassmannandvon
Zedtwitz (1999), it can be seen that drug discovery is
organised as an integrated-network structure. In each
location a critical mass of scientists and technicians
specialised in a set of targets for a disease area are
brought together, facilitating personal interaction and
the exchange of tacit knowledge. The activity of each
centre is supervised by a central research unit, which
could either be in the home country or in the US (as, for
example, in Aventis and Novartis), to avoid duplication
and favour knowledge integration. As pointed out by
Ramirez (2003), the organisation of research activities
according to therapeutic areas has been made possible
by the new heuristic in drug discovery that has allowed
scientists to focus on specic groups of targets in par-
ticular centres of expertise. This international division
of labour enables the rm to access multiple external
knowledge sources from centres of excellence around
the world and/or internal knowledge sources from
research units with a strong technological background
within the organisation. Centres of excellence for a par-
ticular enabling technology (such as genomics, high-
throughput screening, combinatorial chemistry and
bio-informatics) support the activity of these research
units.
The drug development process appears, however, to
be organised as a polycentric decentralised structure,
in Gassmann and von Zedtwitzs (1999) taxonomy.
This structure is characterised by a geographically dis-
persed network of autonomous units with their own
assets and capabilities, which allows themto respond to
local demands and opportunities. Development activi-
ties are indeed concentrated in a few locations in order
to achieve critical mass and economies of scale and
tend to be located in both the home country and the
US to be near the largest markets and the regulatory
authorities. These units usually conne themselves to
developing the results from research units located in
the same region and are coordinated by a global devel-
opment centre (for example, Bridgewater for Aventis).
The degree of interaction among the development cen-
tres is not as intense as among the research network
although they frequently carry out studies for other
development centres whenever these have problems of
insufcient capacity.
6. The role of international assignements and
other forms of mobility
The existing organisation of research activities
in pharmaceuticals is, therefore, quite complex and
involves a high degree of communication among geo-
graphically dispersed research centres that try to oper-
ate in an integrated way during the drug discovery
process. A particular project will be handed-over from
one laboratory to another as it progresses through the
different phases of the drug discovery and develop-
ment process. Thus, results from the drug discovery
process must be transferred to the units carrying out
development activities. Moreover, knowledge acquired
in a particular therapeutic area also needs to be trans-
mitted not only to other units working in the same area
but also to other centres working in different therapeu-
tic areas where it could be employed.
The management and functioning of this complex
organisation relies heavily on ICT applications such as
intranets, portals, project websites, internal databases
of compounds, videoconferencingandemail. However,
face-to-face interaction is important not only for trans-
ferring less codied forms of knowledge but also for
buildingthe necessarytrust, andthus generatingknowl-
edge sharing through electronic communication. As
one scientist noted:
We coordinate our work through phone calls, emails,
but personal contact is crucial; you need to sit together
and you can work much better together if you have built
a personal relationship which is necessary for building
trust . . . the ow of information happens when there is
trust.
Given the apparent importance of face-to-face inter-
action and the creation of personal relationships it
seems natural to investigate what use is made of inter-
national assignments of researchers. Table 3 shows the
characteristics of the interviewed researchers and the
main purposes behind their international assignments.
Some important ndings should be noted. First, the
majority of researchers were seconded for a period of
P. Criscuolo / Research Policy 34 (2005) 13501365 1359
Table 3
Interviewee prole and international assignment characteristics
Job description Major activity R&D function Assignment Period Main purpose of
assignment
Proposed by
Research chemist Search for new compounds Discovery 6 months Personal development Management
Medicinal chemist Synthesis of compounds Discovery 1 year Organisational
development
Scientist
Medicinal chemist Combinatorial chemistry Discovery 6 months Knowledge acquisition Scientist
Senior scientist Development of
high-throughput screening
technology
Discovery 1 year Knowledge transfer Scientist
Senior scientist Synthesis of compounds Discovery 6 months Personal development Scientist
Toxicologist Toxicology studies Development 1 year Organisational
development
Scientist
Product line extension
manager
Creation of new formulations
and products from existing
drugs
Development 6 months Personal development Scientist
Bio-analyst Measurement of drug
metabolism in the plasma
Development 3 years Knowledge transfer Management
Genomics researcher Analysis of gene-expression
data
Discovery 6 months Knowledge acquisition Management
Bio-informatics
researcher
Modelling and simulation of
biological processes
Discovery 10 months Knowledge transfer Management
Clinical
pharmacologist
Analysis of how the drug
works in the human body and
how the body reacts to the
drug
Development 5 years Knowledge transfer Management
Head of therapeutic
area
Development of new
compounds
Discovery 2 years Knowledge acquisition Management
6 months or 1 year, which seems in line with the overall
trend in the international mobility of the work force in
MNEs mentioned in Section 2. Second, although the
most common reason is to transfer or acquire knowl-
edge, international assignments of researchers are also
used for personal and organisational development.
4
Third, even though international assignments appear
to be more frequently employed in the drug discovery
process, which reects the higher level of integration
and interaction among the research units, they are also
used in the drug development process. Finally, these
secondments can either be imposed by the manage-
ment or they can be proposed by the scientist because
he/she considers the experience to be a useful part of
his/her personal development.
4
Even though experience abroad is not articulated as a requirement
for advancement within the company and no promise of promotion is
made before the assignment, researchers know that gaining experi-
ence on how R&D activities are organised in another facility is very
benecial to their careers.
But beyond these general trends it is also use-
ful to look one-by-one at the individual assignments
described in Table 3. One scientist spent 6 months in the
US as a visiting scientist in a dedicated bio-informatics
group formed as a result of a joint venture with an
American company. This assignment was aimed at
acquiring new technological expertise in the eld of
genomics that was missing in the European R&Dfacil-
ity and was needed to undertake gene-chip experiments
in the European site. Similarly, another senior scientist
went for 1 year to a US research facility to implement
a new high-throughput screening technology, which
had been developed in the home country laboratory.
The company believed that the relocation of the key
scientist involved in the development of the newequip-
ment would not only facilitate the transfer of technical
knowledge but would also avoid the not invented here
syndrome (Katz and Allen, 1982). In another company,
sabbatical visits were used for transferring expertise in
combinatorial chemistry and automation technologies
from one site to another.
1360 P. Criscuolo / Research Policy 34 (2005) 13501365
However, this socialisation mechanism is perceived
not only as a way for channelling transfers of knowl-
edge but also as a way of improving relations between
R&D locations, and thus it is used as a coordination
mechanism. This emerges very clearly from the fol-
lowing quote:
Historically, communication and cooperation between
European and US sites has been difcult due to the
different research philosophies. During the mid 1990s
there was more competition than cooperation with
them. During that time it was even difcult to exchange
knowledge. To overcome cultural differences and to
increase a common understanding among colleagues
fromdifferent continents staff exchange programs have
been established. Now, after the rst exchange of staff,
things are improving . . . People taking part in the
exchange programme should be the bridge between the
two sites. It is important to knowpeople, it is much eas-
ier to deal with them if you have worked with them
(emphasis added).
Similarly in another company a scientist said:
The R&Dfacility where I went was part of a company
we just acquired the year before and people there were
not very happy about this unfriendly take over. One
of the purposes of this exchange was to overcome the
initial reservation and try to build a collaboration that
would make us colleagues rather than competitors.
In all instances, this experience enabled researchers
to have better interaction and communication with the
people they met in the other facility. Interaction with
colleagues inside the company appeared to be the most
important source of learning, while very little interac-
tion seemed to have taken place between the seconded
researchers and other host country rms, universities,
or scientists. This nding conrms that the main pur-
pose of these assignments is network building among
researchers working in geographical distant R&D cen-
tres and increasing awareness of who knows what
in other centres. These relationships are in most cases
maintained once the scientist has returned to his/her
home laboratory. As stated by one researcher:
This experience made me knowing people to whom
I have been in touch later on my job, once back in
Europe. I got a better understanding of the technology
they were using in the US and I was also able to record
the contact people in the US subsidiary.
Researchers who had been on secondment seemed
to have also gained a better understanding of the R&D
organisation and management adopted in the US cen-
tre and this appears to be one of the primary goals
attributed to this type of socialisation mechanism by
the top management of the companies. According to
the head of a therapeutic area:
For me, in terms of delivering projects globally, it
is important that people [when sent to another R&D
centre] understand what it takes to deliver things in
a different country and how to do it most effectively.
When these people go back to their original site they
actually strengthen the site they came from because
they know how this organisation works globally rather
than knowing only how it works based on what they
have seen in one site . . . when you send someone
in another R&D site you have to make sure that [the
international assignment] is seen as a family thing not
an individual thing (emphasis added).
Thus, while scientists perceive the foreign expe-
rience as a chance to broaden their knowledge and
acquire new skills, top management see it as an oppor-
tunity to increase the productivity of the company by
getting the researchers acquainted with how the R&D
function is managed in other facilities.
However, even though rms acknowledge the cru-
cial role of scientists in the international diffusion of
knowledge and rely heavily on the ability and private
initiative of individual scientists to span the bound-
aries of their own area of specialisation and interact
with other scientists, the use of international assign-
ments is very limited. Across all rms in 2003 only
1% of the R&D staff employed was on secondment to
another unit in the same company for a period longer
than 6 months and secondments are generally organ-
ised only when needed. Only in one division of one
particular company is there an organised exchange
programme where a scientist from the European divi-
sion spends 1 year in the US facility while a scientist
from the US replaces her/him at the European site.
This programme has only been in place for the past
2 years and is an attempt at replacing the less popular
P. Criscuolo / Research Policy 34 (2005) 13501365 1361
sabbatical option, where a scientist goes to a foreign
R&D laboratory but is not replaced by another scien-
tist. However, this low degree of long-term mobility
is accompanied by a high degree of short-term mobil-
ity associated with the implementation of cross-border
projects and/or temporary assignments (from 1 week
to 6 months) for the transfer of specic expertise. As
observed by an R&D manager: because people travel
more often for weekly periods, there is less need to
send them for longer periods. Cross-border projects
are widely used because of the organisational structure
of the R&Dactivities, which often requires researchers
from different locations at different stages of the drug
discovery project. They also aim to replace long-term
secondments of researchers. As explained by an R&D
manager:
Resources are managed globally but people stay at the
site. For example imagine you have a diabetes project
in the European site where the expertise is and you need
chemical expertises. It might happen that the chemist is
working in the US facility. Typically you do not need
to move the chemist from the US to Europe because
you just include this particular chemist in the research
team.
These international project teams allow individuals
from distant locations and from different cultural and
technological background to be brought together reg-
ularly. However, cross-border team projects are seen
more as an unavoidable consequence of the R&D
organisation than as a strategic means for increas-
ing knowledge diffusion inside the geographically dis-
persed R&D network. This clearly emerges from the
comments of an R&D manager:
The one-location teamis the preferable model because
it is the more efcient, but the realityof our organisation
is that most of our teams have members based in at least
two countries and some of them three. My personal
view is that if you can have one location team you are
going to be better off, if you can have all sitting in one
corridor is goingtoworkbetter. But this is the exception
to the rule.
Temporary assignments and short visits are also
quite often employed and they are easy to organize and
do not require senior manager authorization.
People will travel quite easily to have face-to-face
meetings. It is quite easy to bring someone over from
another facility. For example we had a resource short-
fall in the headquarters and we brought two people with
that particular skill from North America to work here
for three weeks. We do this quite often. It is resource
driven process, not part of individual personal devel-
opment.
This type of short-term assignment is not only
adopted when there is a shortfall of resources but is
also employed for transferring specic expertise as is
the case when a project passes fromthe discovery to the
development phase. At this stage of the drug discov-
ery process researchers from the development function
will work closely with the discovery team from 3 to 6
months before the compound has been identied.
Another way of bringing together people from
across the organisationwithsimilar expertise is through
seminars on specic scientic and technology areas.
In only one company is there an annual international
research meeting where all the researchers present their
work. This type of event does not take place in other
companies because of the size of the R&Dorganisation
and it has been replaced by smaller workshops focused
on a particular discipline. Similarly, companies pro-
mote and ofcially support communities of practice
(Wenger, 1998) and technology interest groups. These
communities are formed by people with expertise in a
target family, such as kinesis or protease, or in a dis-
cipline such as molecular biology, chemistry, pharma-
cology. The task of this group of experts is to integrate
and diffuse knowledge across different locations and
therapeutic areas.
5
The members of these communi-
ties interact regularly through intranet, meetings and
formal workshops. Technology councils are set up to
discuss problems faced in using particular technolo-
gies, such as those used in combinatorial chemistry or
high-throughput screening. These councils provide a
forum where individuals can share best practice and
some of the challenges they face in using a particular
technology. Another organisational measure adopted
by one company to gather people from different thera-
peutic areas and locations is cross-disciplinary project
proposal review boards.
5
They are also supposed to span outside the rms boundaries and
follow the developments of their scientic elds outside the rm.
1362 P. Criscuolo / Research Policy 34 (2005) 13501365
Thus, in line with what is found by B eret et al.
(2003) and with the general trend in international
labour mobility in MNEs, the adoption of an integrated
R&D network has lead to an increase in researchers
mobility but in the form of short-term visits which are
either connected to cross-border projects or organised
for transferring specic skills.
The general impression from the interviews with
R&Dmanagers is that the use of long-termmobility has
declined and is now very limited. Secondments abroad
were more often employed in the past when each centre
was organised as a completely self-sufcient R&Dunit
specialised in certain therapeutic areas where most of
the phases of the drug discovery and development pro-
cess were carried out. Because all the R&D staff from
research to development were in one place, there was
less short-term researcher mobility and cross-border
team projects were almost never implemented. Long-
term international assignments of researchers instead
were employed more often than in the current organi-
sation model as a mechanism for transferring complex
and location-specic knowledge.
There are two main factors explaining this low level
of long-term mobility of researchers in the form of
international assignments. First, it is very expensive
to send someone to another facility (according to an
R&D manager it doubles the cost of employment) and
the scientists themselves are reluctant, perhaps more
so than before, to leave lifestyles behind to take on
assignments abroad. Second, the R&D organisational
structure itself hampers long-term researcher mobil-
ity, a point emphasised also by B eret et al. (2003).
The high-degree of technological distance between
research sites implies that competences are not easily
transferable from one site to another, making mobility
difcult. As acknowledged by an R&Dmanager, scien-
tists do not move among different R&D units, because
the expertise they have is often on a specic therapeutic
area that cannot be applied in other research contexts.
This emerged very clearly from interviews with those
seconded researchers whose primary purpose was not
to transfer specic knowledge to the foreign R&D
unit.
When I was in the US I developed a completely new
class of drugs. I wanted to work more on that class of
compounds, but I did not, because it would have meant
implementing here a different program.
I was not able to apply much of what I have learnt in
the US because we use a different approach.
However, it is possible to identify a longer-term
potential for cross-fertilisation arising from exchanges
of personnel.
Coming from traditional medicinal chemistry it was
my rst chance to get in touch with automated chem-
istry. After staying three months in the automated
chemistry group in our US subsidiary I was aware of
the scope and limitations of the approach which helped
a lot in contributing to a global technology strategy.
While I was in the USI had started a collaboration with
a company expert in gene-chip technology. This expe-
rience improved my innovation activity because we are
trying to work more and more in my eld with these
micro-areas and this time in the US was a trigger for
me to think very hard on how I could use these tools.
7. Conclusion
This paper has analysed the roles of international
assignments and other forms of inter-unit researcher
mobility in six of the largest European pharmaceu-
tical MNEs as a coordination and technology trans-
fer mechanism. In these companies the innovative
efforts in the drug discovery phase are carried out
in an integrated network structure characterised by
inter-unit geographical, technological and organisa-
tional distance, which could hinder knowledge sharing
and coordination of the R&D activities. In this organ-
isational structure international assignments are used
to achieve three main goals. First, researchers are sec-
onded to other R&D units to transfer technical knowl-
edge from the headquarters or to acquire technical
knowledge fromother researchfacilities. Second, inter-
national assignments are used for individual leader-
ship development. Third, assigning researchers to other
R&D units is employed as a coordination mechanism,
i.e. as a means of reducing organisational distance
through social interaction among researchers. The sec-
ondment experience appears to increase researchers
overall understanding of R&D organisation and man-
agement in the other unit, helping themto identify who
knows what, and fostering the creation of personal
P. Criscuolo / Research Policy 34 (2005) 13501365 1363
relationships with other researchers. Thus, researcher
mobility seems to help overcoming some of the bar-
riers to knowledge sharing inside the R&D network
and fully exploiting the benets of this organisational
form.
However, it was found that long-term international
assignments are not widely used even though com-
panies rely heavily on the initiative of individual
researchers to interact with other colleagues and to
enhance R&D synergies across projects. They have
been replaced by other, short-term forms of mobility
which are less costly and do not require the relocation
of the family. Researchers from different R&D centres
meet each other quite regularly either because there
are involved in cross-border team projects or because
they participate in a technology interest group. Short-
termassignments are also used for transferring specic
expertise to other R&D facilities or for solving prob-
lems of critical-skill shortages.
This paper has provided some indicative evidence
on the frequency and purpose of inter-unit researcher
mobility but further and more systematic research is
needed. One issue to be addressed is whether the shift
towards repeated short-term visits is less conducive to
the formation of enduring and strong social relation-
ships among researchers than longer termassignments.
Building these social relationships is important not
only for transferring knowledge but also for reducing
inter-unit attrition and improving cooperation between
research facilities. A second issue relates to the impact
that these changes in the duration of international sec-
ondments have on the internal transfer of knowledge
rms innovative performance. It was found that short-
term assignments are mainly aimed at transferring
project and/or task-related knowledge; thus, one won-
ders whether they allowany opportunity for identifying
potential synergies across research areas. Short-term
visits may hinder serendipity in drug discovery, on the
one hand, and cross-fertilisation of knowledge within
and across therapeutic areas and thus innovative per-
formance on the other. As argued by Henderson
and Cockburn (1994, 1996), both are becoming cru-
cial for discovering new drugs. Thanks to the adop-
tion of the new science-deductive method the search
for new compounds is now being shaped by knowl-
edge of the fundamental physiological mechanisms,
and therefore knowledge acquired in one therapeutic
area can lead to advances in other areas. Similarly
Hollingsworths work (2005) shows that major dis-
coveries in biomedical research occur in organisations
characterised by a high degree of interdisciplinary and
integrated activity across diverse disciplines. One can
imagine a number of ways to explore these research
questions. One of them could be looking at the perfor-
mance of individual scientists and their social networks
and relate them to their intra-rm mobility patterns.
Results from such an investigation could be of great
interest for the management of R&D organisations.
If scientists with a long-term secondment experience
are more likely to make major breakthroughs because
they are more interdisciplinary, then R&D managers
should invest more resources in putting scientists on
the road.
Acknowledgements
The author would like to acknowledge the nan-
cial support received by the European Commission
(Marie Curie Fellowship) and to thank Stefano Bru-
soni, Rajneesh Narula, Lionel Nesta, Andrea Prencipe,
Matias Ramirez, Fergal Shortall, Ammon Salter, Ed
Steinmueller, two anonymous referees, and partici-
pants at the conference in honour of Keith Pavitt and
at the EIBA conference for helpful comments and sug-
gestions.
References
Allansdottir, A., Bonaccorsi, A., Gambardella, A., Mariani, M., Ors-
enigo, M., Pammolli, F., Riccaboni, M., 2002. Innovation and
Competitiveness in European Biotechnology, Enterprise Papers
No. 7.
Allen, T.J., 1970. Communication networks in R&D laboratories.
R&D Management 1 (1), 1421.
Allen, T.J., 1977. Managing the Flow of Technology. MIT Press,
Cambridge, MA.
Almeida, P., Kogut, B., 1999. Localization of knowledge and the
mobility of engineers in regional networks. Management Science
45 (7), 905917.
Arora, A., Gambardella, A., 1994. The changing technology of
technological change: general and abstract knowledge and the
division of innovative labour. Research Policy 23 (5), 523532.
Bartlett, C.A., 1986. Building and managing the transnational. The
new organizational Challenge, in: M.E., Porter (Ed.), Competi-
tion in global industries. Boston, MA, Harvard Business School
Press.
1364 P. Criscuolo / Research Policy 34 (2005) 13501365
Bartlett, C.A., Ghoshal, S., 1990. Managing innovation in the
transnational corporation. In: Bartlett, C., Doz, Y., Hedlund, G.
(Eds.), Managing the Global Firm. Routledge, London.
Bartlett, C.A., Ghoshal, S., 1989. Managing across borders: the
transnational solution. Boston, MA, Harvard Business School
Press.
B eret, P., Mendez, A., Paraponaris, C., Richez-Battesti, N., 2003.
R&D personnel and human resource management in multina-
tional companies: between homogenization and differentiation.
International Journal of Human Resource Management 14 (3),
449468.
Breschi, S., Lissoni, F., 2003. Mobility and Social Networks.
Localised Knowledge Spillovers Revisited, CESPRI Working
Paper No. 142.
Buckley, P.J., Casson, M., 1976. The Future of the Multinational
Enterprise. London, Macmillan.
Casson, M., Singh, S., 1993. Corporate research and development
strategies: the inuence of rm, industry and country factors on
the decentralization of R&D. R&DManagement 23 (2), 91107.
Chiesa, V., 1996a. Human resource management issues in global
R&D organisation: a case study. Journal of Engineering and
Technology Management 13 (2), 189202.
Chiesa, V., 1996b. Managing the internationalisation of R&D activ-
ities. IEEE Transactions on Engineering Management 43 (1),
723.
Chiesa, V., 1996c. Separating research from development: evidence
from the pharmaceutical industry. European Management Jour-
nal 14 (6), 638647.
Chiesa, V., Manzini, R., 1996. Managing the transfer of knowledge
within multinational rms. International Journal of Technology
Management 12 (4), 462476.
Daft, R.L., Lengel, R.H., 1986. Organizational information require-
ments, media richness and structural design. Management Sci-
ence 32 (5), 554571.
Dalton, D., Serapio, M., Yoshida, P., 1999. Globalizing Industrial
R&D. U.S. Department of Commerce, Technology Administra-
tion, Ofce of Technology Policy.
De Meyer, A., 1993. Internationalising R&D improves a rm tech-
nical learning. Research Technology Management 36 (4), 42
49.
De Meyer, A., Mizushima, A., 1989. Global R&D management.
R&D Management 19 (2), 135146.
Desanctis, G., Monge, P., 1999. Introduction to the special issue:
communication processes for virtual organizations. Organization
Science 10 (6), 693703.
Edstr om, A., Galbraith, J.R., 1977. Transfer of managers as a coor-
dination and control strategy in multinational organizations.
Administrative Science Quarterly 22 (2), 248268.
Gassmann, O., von Zedtwitz, M., 1999. New concepts and trends
in international R&D organization. Research Policy 28 (23),
231250.
Glaser, B.G., Strauss, A.L., 1967. The Discovery of Grounded The-
ory: Strategies for Qualitative Research. Aldine De Gruyter, New
York.
Grant, R., 1996. Toward a knowledge-based theory of the rm.
Strategic Management Journal 17 (Winter special issue), 109
122.
Gupta, A.K., Govindarajan, V., 2000. Knowledge ows within multi-
national corporations. Strategic Management Journal 21 (4),
473496.
Harris, H., 2002. Strategic management of international workers.
Innovations in International HR 28 (1), 15.
Harzing, A.W., 2000. Of bears, bumble-bees, and spiders: the role of
expatriates in controlling foreign subsidiaries. Journal of World
Business 36 (4), 366379.
Hedlund, G., 1986. The hypermodern MNC A heterarchy? Human
Resource Management 25, 935.
Henderson, R., Cockburn, I., 1994. Measuring competence? Explor-
ing rm effects in pharmaceutical research. Strategic Manage-
ment Journal 15 (Winter special issue), 6384.
Henderson, R., Cockburn, I., 1996. Scale, scope, and spillovers:
the determinants of research productivity in drug discovery. The
Rand Journal of Economics 27 (1), 3259.
Hollingsworth, R., Hollingsworth, E.J., Hage, J., 2005. Fostering
Scientic Excellence: Organizations Institutions and Major Dis-
coveries in Biomedical Science. Cambridge University Press,
New York.
Howells, J., 1995. Going global: the use of ICT networks in research
and developments. Research Policy 24, 169184.
Katz, R., Allen, T.J., 1982. Investigating the not invented here (NIH)
syndrome: a look at the performance, tenure, and communication
patterns of 50 R&D project groups. R&D Management 12 (1),
719.
Kogut, B., Zander, U., 1992. Knowledge of the rm, combinative
capabilities, and the replication of technology. Organization Sci-
ence 3 (3), 383397.
Kogut, B., Zander, U., 1993. Knowledge of the rm and the evo-
lutionary theory of the multinational corporation. Journal of
International Business Studies Fourth Quarter 625645.
Le Bas, C., Sierra, C., 2002. Location versus country advantages in
R&Dactivities: some further results onmultinationals locational
strategies. Research Policy 31 (4), 589609.
Nightingale, P., 2000. Economies of scale in experimentation: knowl-
edge and technology in pharmaceutical R&D. Industrial and
Corporate Change 9 (2), 315359.
Nightingale, P., Martin, P., 2004. The myth of the biotech revolution.
Trends in Biotechnology 22 (11), 564569.
Orlikowski, W., 2002. Knowing in practice: enacting a collective
capability in distributed organizing. Organization Science 13 (3),
249273.
Pearce, R.D., 1999. Decentralised R&D and strategic competitive-
ness: globalised approaches to generation and use of technology
in multinational enterprises (MNEs). Research Policy 28 (23),
157178.
Perlmutter, H., 1965. Lenterprise internationale trois conceptions.
Revue Economique et Sociale 23.
Persaud, A., Kumar, U., Kumar, V., 2001. Harnessing scientic and
technological knowledge for the rapiddeployment of global inno-
vations. Engineering Management Journal 13 (1), 1218.
PricewaterhouseCoopers, 1999. International Assignments. Euro-
pean Policy and Practice. P. Europe.
Ramirez, P., 2003. Globalisation, technology and organizational
change in the pharmaceutical industry, Ph.D. thesis, Manchester
School of Management.
P. Criscuolo / Research Policy 34 (2005) 13501365 1365
Reger, G., 2000. Internationalization of research and development in
pharmaceuticals. In: Jungmittag, A., Reger, G., Reiss, T. (Eds.),
Changing Innovation in the Pharmaceutical Industry. Springer-
Verlag, Berlin.
Reiss, T., Hinze, S., 2000. Innovation process and techno-scientic
dynamics. In: Jungmittag, A., Reger, G., Reiss, T. (Eds.), Chang-
ing Innovation in the Pharmaceutical Industry. Springer, Berlin.
Roberts, K., Kossek, E.E., Ozeki, C., 1998. Managing the global
workforce: challenge and strategies. Academy of Management
Executive 12 (4), 93106.
Rosenkopf, L., Almeida, P., 2003. Overcoming local search through
alliances and mobility. Management Science 49 (6), 751766.
Senker, J., 1998. BiotechnologyandCompetitive Advantage. Edward
Elgar, Cheltenham.
Shan, W., Song, J., 1997. Foreign direct investment and the sourc-
ing of technology advantage: an evidence from the biotechnol-
ogy industry. Journal of International Business Studies 28 (2),
267284.
Sharp, M., 1999. The science of nations: Europeanmultinationals and
American biotechnology. International Journal of Biotechnology
1 (1), 132159.
Singh, J., 2004. Social Networks as Determinants of Knowledge
Diffusion Patterns. Harvard Business School and Department of
Economics, Mimeo.
Stolpe, M., 2002. Determinants of knowledge diffusion as evidenced
in patent data: the case of liquid crystal display technologies.
Research Policy 31 (7), 11811198.
Szulanski, G., 1996. Exploring internal stickiness: impediments to
the transfer of best practice within the rm. Strategic Manage-
ment Journal 17 (Winter Special Issue), 2743.
Teigland, R., Fey, C., Birkinshaw, J., 2000. Knowledge dissemination
in global R&D operations: an empirical study of multinationals
in the high technology electronics industry. Management Inter-
national Review 40 (1), 4978.
Welch, D.E., Worm, V., Fenwick, M., 2003. Are virtual assignments
feasible? Management International Review 43 (special issue),
95114.
Wenger, E.C., 1998. Communities of Practice: Learning, Meaning
and Identity. Cambridge University Press, Cambridge.
Westney, D.E., 1993. Cross-pacic internationalization of R&D
by U.S. and Japanese rms. R&D Management 23 (2), 171
181.