Академический Документы
Профессиональный Документы
Культура Документы
Geochemistry: Exploration, Environment, Analysis, Vol. 6 2006, pp. 69–78 1467-7873/06/$15.00 2006 AAG/ Geological Society of London
70 C. R. Stanley
y = arcsin√x (5) The resulting x’s and y’s can then be averaged, and the mean x
and y values (x̄ and ȳ) can then be converted back into the
(Zubin 1936; Eisenhart et al. 1947; Anscombe 1948; Ghurye average azimuth using:
1949; Freeman & Tukey 1950; Claringfold et al. 1953; Stevens
1953; Biggers & Claringfold 1954; Blom 1954; Fisher 1954); and
reciprocal SD
= arctan ȳ
x̄
(8)
y=1 (6) Equations 7 and 8 are thus transformations that convert the
x original azimuths into and back from an additive set of variables
(Hoyle 1973) transformations. In these transformation func- that can be operated on (averaged) to determine the azimuth of
tions, the ‘s are commonly unconstrained, and therefore the ice flow direction (Fisher et al. 1987).
represent arbitrarily chosen parameters. Another example of additivity involves determination of
Historically, many of these transformations have been sampling error in geochemistry. Standard deviations of replicate
employed to create normally distributed variables (e.g. Hoyle samples collected in the field typically are used to estimate the
1973; Joseph & Bhaumik 1997; Box & Cox 1964); however, the combined magnitude of sampling and analytical (i.e. total) error.
commonly multi-modal nature of geochemical concentration Standard deviations of replicate analyses of the samples, in
data makes it virtually impossible to transform most geo- contrast, estimate only the magnitude of analytical error.
chemical concentrations into new variables with distributions Unfortunately, one may not simply subtract the analytical error
approximating normality. As a result, applying the above standard deviation (a) from the total error standard deviation
transformations to convert a multi-modal distribution into a (t) to determine the sampling error standard deviation (s),
normal one may not be realizable, and thus may be misguided. because errors are additive as variances (not as standard
Other reasons exist for transforming geochemical data, and deviations). As a result, we must square the total error and
these may impart, to the geochemical data, desirable features analytical error standard deviations, to produce the correspond-
that facilitate interpretation. Historically, and generally outside ing variances (Davis 1986). Then, by subtraction, we determine
the field of geochemistry, data have been transformed for three the sampling error variance, and by taking the square root we
different and distinct reasons (Tukey 1958; Dolby 1963; Kruskal determine the sampling error standard deviation. These power
1968; Hoyle 1973): (1) to create a new variable that is ‘additive’ (square and square root) transformations are used to create
(Elston 1961; Hoyle 1973); (2) to create a new variable that has new variables that are additive (the variances), and can be
‘constant variance’ (i.e. homoscedastic; Kendall & Stuart 1966; manipulated using the sum-of-variances equation:
Hoyle 1973); and (3) to create a new variable that is normally
distributed (Hoyle 1973; Joseph & Bhaumik 1997; Box & Cox s2 + a2 = t2 (9)
1964).
to determine sampling error.
A third example of additivity involves the inherent nature
Additivity of geochemical concentrations in rock samples, which vary
‘Additivity’ is the linear property of a variable and allows it to be because geological material transfer processes (fractional
numerically operated on in a linear manner that often has crystallization, diagenesis, metamorphism, weathering, hydro-
physical meaning (Elston 1961; Hoyle 1973). Additivity is not thermal alteration, etc.) have added and removed material to
commonly understood by geologists, because it is a data feature and from the rocks over time. Because the absolute amounts of
rarely sought, mostly because many geological variables are each element in the rocks vary in proportion to how much of
already ‘additive’. The several examples below illustrate geo- each element is added or removed, these element amounts are
chemical applications where transformations may be used to already linear (additive). For example, adding 1 g of Na to a
create additivity. 100 g rock containing 10% Na (i.e. containing 10 g of Na)
The first example involves a transformation required to produces a new rock with an amount of Na equal to the amount
obtain ‘additivity’ in till geochemical exploration, where knowl- initially present plus the amount that was added (11 g).
edge of the glacial flow directions is critical to data interpret- Unfortunately, the absolute amounts of elements in samples
ation. Glacial flow directions can be determined by measuring are not generally known in geochemistry because when the
the azimuths of the long dimensions of clasts within a till. samples are collected, information is lost about the size of the
Because elongate clasts line up parallel to the glacial flow rock. As a result, we normally describe rock compositions using
direction due to shear forces imparted by the moving ice (Mark element concentrations (proportions), and these are not neces-
1973), these measurements can be used to determine the ice sarily proportional to the material transfers that take place to
dispersal direction. As a result, after collecting an adequate cause compositional variability (Stanley & Madeisky 1995). This
number of till clast major axis azimuth measurements (generally is because addition or loss of an element can cause the
>50), the geochemist typically determines the average direction concentrations of other elements to change (via closure), even
of the major axes of the clasts to deduce the glacial flow though these elements were neither added nor removed (i.e.
Numerical transformations of geochemical data: 1 71
they are conserved; Chayes 1962; Stanley & Madeisky 1995). As be used to determine the relative change in the element
a result, a transformation may be used to convert the concen- concentration (dx/x). For the major element:
trations into molar Pearce element ratios (PERs; Pearce 1968;
Russell & Nicholls 1988; Russell & Stanley 1990a; 1990b; dx = dX dS = 0.1 0.1 = 0.010 0.001 = 0.009 (14)
Stanley & Madeisky 1995), which are molar ratios with a x X S 10 100
conserved element denominator that acts as a standardizing
variable to circumvent the effects of closure. The PERs have for the minor element:
values that change linearly with the amount of material
dx = dX dS = 0.1 0.1 = 0.100 0.001 = 0.099 (15)
transfer that the numerator element of the ratio undergoes x X S 1 100
during the material transfer process (Pearce 1968; Stanley &
Madeisky 1995). As a result, PERs are proportionally ‘additive’, and for the trace element:
and the equation used to obtain a PER from concentration
data is: dx = dX dS = 0.1 0.1 = 1.000 0.001 = 0.999 (16)
x X S 0.1 100
x/gx
PERx/z = (10) In all cases, the closure term (dS/S)=0.001, but this closure
z/gz
effect is proportionally largest, relative to the material transfer
where the element concentrations (x and z) are in mass units effect (dX/X), for the major element and smallest for trace
(wt%, ppm, ppb, etc.), and the gram formula weights (g) are in element because the material transfer effects differ simply
grams/mole (Stanley & Madeisky 1995). This conversion because of the different concentration levels. Clearly, closure
formula is an ‘additive transformation’. Note that Equation 10 introduces significant amounts of additional variation to major
is also a multi-variate transformation, because it involves two elements, and has much less impact, introducing much less
variables (x and z). As a result, Equation 10 may not be variation, in elements with lower concentrations. In practice,
considered by some to be strictly a transformation because it is the effect of closure in trace elements is generally negligible, and
a function of two variables. Nevertheless, it does produce a new use of a transformation to create an additive geochemical
variable that is additive. variable from a trace element concentration is generally unnec-
One should note that although major element concentrations essary.
are typically not ‘additive’, due to closure, trace element Although the above three examples illustrate the relevance of
concentrations commonly are. This can be demonstrated by additivity transformations in geochemical data analysis, trans-
considering the definition of a concentration: formations to create new variables that have ‘additivity’ are
otherwise rarely undertaken in most geochemical applications.
x=X (11)
S
Error variance stabilization
where X is the mass of an element, S is the mass of a rock, and Another possible reason for data transformation is called ‘error
x is the mass concentration of that element. Taking the variance stabilization’. This has historically not been a tradi-
derivative of Equation 11 yields: tional motivation for data transformation in geochemical appli-
cations, but is a desirable trait for geochemical data that are to
dx = dX X2 dS (12) be numerically or statistically evaluated. Error variance stabiliz-
S S ation is used to treat data that are ‘heteroscedastic’, having
and dividing both sides by x and rearranging, gives: magnitudes of variations (measurement errors) that change
across the range of the data. Most geochemical concentration
dx = dX dS data are ‘heteroscedastic’, as analytical errors commonly
(13) increase with concentration (Thompson & Howarth 1973,
x X S
1976a, b, 1978; Thompson 1973, 1982; Fletcher 1981; Stanley &
This indicates that the relative change in an element concen- Sinclair 1986; Stanley 2003b). This makes data evaluation
tration is a function of the relative change in the amount of the difficult, because the variations attributable to measurement
element (dX/X; the effect of material transfer) and the relative error vary across the range of the data, and the criteria used to
change in the size of the rock (dS/S; the effect of closure). When determine whether observed concentration differences are real
material transfer does not involve a change in the size of the thus change across the range of concentrations as well.
rock (i.e. it is a perfect exchange process and dS=0), then the Transformation to stabilize the error in a geochemical
relative change in concentration is equal to the relative change variable creates a new variable that is ‘homoscedastic’ (has
in the amount of the element, and closure has no impact. As a constant error variance across the range of the new transformed
result, the magnitude of the dS/S term in Equation 13 describes variable), and this attribute can significantly simplify subsequent
the magnitude of the closure effect. This allows Equation 13 statistical tests, numerical procedures and the graphical presen-
to be used to illustrate why closure affects major element tation of the data. The transformation function used is specific
concentrations much more than trace element concentrations. to the model describing how measurement error changes with
Consider three rocks each containing a major element with a concentration, and propagation of the original heteroscedastic
concentration of 10 wt%, a minor element with a concentration error into the new variable results in new (transformed) error
of 1 wt%, and a trace element with a concentration of 0.1 wt%. that is homoscedastic. Evaluation of homoscedastic data is far
Each of these rocks undergoes a different type of hydrothermal simpler and leads to more compelling conclusions because the
metasomatism; in the first rock the major element is added, in data can be evaluated on an ‘even playing field’.
the second rock the minor element is added, and in the third Examples of error stabilizing transformations commonly
rock the trace element is added. Let us assume that the amount used in other scientific applications (Hoyle 1973) include:
of addition of each element is equal (0.1 g), and that the mass (1) the logarithmic transform (for proportional error; Equation
of each rock was initially 100 g. For each rock, Equation 13 can 1); (2) the square root transform (for Poisson error;
72 C. R. Stanley
Equation 2); and (3) the angular transform (for binomial error;
Equation 5).
To illustrate how these error-stabilizing transformations
work, consider Case 1, above. If proportional measurement
error is 10%, then =y=0.10 c, where c is an element concen-
tration. Propagation of this error into the logarithm of c (y=log
c) requires use of the generative error propagation equation
(Stanley 1990):
p p1 p
In this case, there is only one variable (c), which we will assign
to x1, so Equation 17 simplifies significantly (the second term
equals zero). The only required partial derivative is thus: Fig. 1. Histogram of the 80 mesh (177 µm) fraction Cu
concentrations (in ppm) from 247 B-horizon soil samples collected
Syc D = 1c (18)
over the Daisy Creek strata-bound Cu–Ag prospect, Montana, and
analysed after an aqua regia digest with atomic absorption spec-
trometry (Stanley 1984). The minimum, maximum, mean, standard
deviation, skewness and kurtosis of these data are 10 ppm, 973 ppm,
and Equation 17 becomes: 74.964 ppm, 123.411 ppm, 3.830 and 17.595, respectively. Data have
been rescaled to the interval 0 ↔ 1 and are plotted with 40 class
2
SD
y2 = 1 (0.10c)2 = 0.01
c
(19)
intervals (note that this does not change the skewness and kurtosis).
Clearly, regardless of the magnitude of c, the error in y (the Unfortunately, most geochemical data derive from several
transformed variable) equals 0.01 and is constant (homo- geological sources. Because multiple rock types or geological
scedastic) across the range of y. materials may be represented in a geochemical survey dataset,
To date, no published geochemistry-related paper is known the frequency distribution of the geochemical data is commonly
to the author that transforms geochemical data for this purpose, multi-modal (Sinclair 1976; Stanley 1988). As a result, no matter
although ‘homoscedasticity’ would significantly facilitate a large what transformation is used, it is generally impossible to obtain
number of efforts to evaluate geochemical data. a truly normally transformed distribution, unless the transfor-
mation is not monotonic (in which case the fundamental
structure of the data is destroyed, and the information in the
Normality data is significantly altered). Application of the Box–Cox
In many geochemical applications, transformations are com- transform to the soil Cu concentration data from the Daisy
monly undertaken to convert, or attempt to convert, the Creek stratabound Cu–Ag prospect, Montana (Stanley 1984)
observed frequency distribution into one that is normal, illustrates this problem (Fig. 1). The frequency distribution of
approximately normal, or at least symmetric (e.g. Bartlett 1947; this dataset has been modelled using probability plot analysis
Draper & Cox 1969; Hoyle 1973). This is typically done for (Sinclair 1976; Stanley 1988), and is interpreted to consist of a
theoretical reasons, because subsequent statistical procedures mixture of three lognormal populations (Fig. 2), with means and
either require data that are normally distributed, or are substan- mean one standard deviation concentrations (in square
tially simplified if the data are normally distributed (Hoyle brackets) of 26 [17, 38], 78 [55, 111] and 313 [198, 496] ppm
1973). A common transformation that has been used to convert that comprise 71, 19 and 10% of the data, respectively. These
data into a new variable that is at least ‘nearly normal’ is the component distributions can be related spatially to background,
Box–Cox transform (Equation 4). Typically, is adjusted until chalcopyrite mineralized and chalcocite+bornite-mineralized
the resulting distribution has a skewness of 0 and a kurtosis or zones at the prospect, and so have validated geological causes
3 (or nearly so; Joseph & Bhaumik 1997; Box & Cox 1964; (Stanley 1984; Sinclair 1991).
0 and 3 are the expected values of these parameters for a normal Application of the maximum likelihood Box–Cox transform
distribution). A maximum likelihood objective function can be to these data significantly reduces the skewness and kurtosis of
maximized: this soil concentration variable (from 3.830 and 17.595, to 0.161
and 0.564, respectively). However, even though these par-
S D
ameters can be made more consistent with the expected values
n
for a normal distribution (0 and 3, respectively) by adjusting
<max n ln2 + ( 1)
2 y &lnx
i=1
i (20) to 0.58 using the maximum likelihood approach of Joseph &
Bhaumik (1997), the resulting distribution is still multi-modal,
has a non-ideal skewness and kurtosis, and exhibits other
to determine the optimal value of that produces a transformed significant departures from normality (Fig. 3). As a result, the
distribution with skewness and kurtosis closest to 0 and 3, frequency distribution from this transformation in no way
respectively (Joseph & Bhaumik 1997). Computer software in approximates a normal distribution, even though the transform
the form of a MATLAB procedure (BOXCOX.M) is available employed is designed to create a distribution approximating
that determines the maximum likelihood value of for a normality. Nevertheless, the transform radically reduced the
Box–Cox transform that converts data into a newly trans- skewness of the data, because it stretched the data in concen-
formed variable with a distribution that comes closest to tration ranges where data frequencies are high, and compressed
normality (this procedure can be downloaded from the author’s the data in concentration ranges where data frequencies are low
website at http://www.acadiau.ca\cstanley\software.htm). (Fig. 4).
Numerical transformations of geochemical data: 1 73
yi = SS xi min
max min
D
(1 2ε) D +ε (21)
Data presentation Fig. 9. Bubbleplot of raw Cu concentrations from the Daisy Creek
Presentation of the above transformed data in geographic space soil survey (Stanley 1984). Values were rescaled to the interval 0 ↔
may lead to recognition of spatial patterns that can be associated 1, and the bubble diameters are proportional to this rescaled value.
with geological features of interest. However, like the selection
of an appropriate transformation, selection of an appropriate
format to present the transformed concentration data is very patterns. Unfortunately, there is no theoretical reason or
important. Geochemical data are typically punctual data, col- guarantee for any spatial patterns to exist a priori (e.g. some
lected from finite points in space. Most geochemical samples surveys result in ‘salt-and-pepper’ type patterns reflecting
are collected from small volumes, but the results of geochemical measurement of a geochemical variable that lacks regional
analysis of these samples are typically expected to represent correlation). As a result, because geochemical data are not
significantly larger regions of a survey area (e.g. on a square soil fundamentally regionalized, and this may be in part due to
grid, an actual sample may be derived from a single soil pit of inadequate sample support (either the sizes of the samples are
25 cm 25 cm area, but the sample spacing may be 50 m too small, or the samples are not truly representative of the
50 m). As a result, many geochemical concentrations do not areas from which they have been collected; Clark 1979; Sinclair
represent an integrated signal from across a large sample region, & Blackwell 2002), contouring of geochemical data is generally
but rather are typically grab samples from specific finite not justified before spatial examination proves that patterns or
locations. trends actually exist. Consequently, an alternative method for
Nevertheless, experience has shown that geochemical displaying geochemical data to allow recognition of a spatial
samples collected in this way may exhibit discernible spatial trend, other than contouring, is necessary so that contouring is
76 C. R. Stanley
Fig. 10. Bubbleplot of raw Cu concentrations from the Daisy Fig. 11. Bubbleplot of Box–Cox transformed Cu concentrations
Creek soil survey (Stanley 1984). Values were rescaled to the interval from the Daisy Creek soil survey (Stanley 1984). Values were rescaled
0 ↔ 1, but in contrast to Figure 9, these bubble areas are to the interval 0 ↔ 1, and the bubble diameters are proportional to
proportional to this rescaled value. this rescaled value.
REFERENCES
AHRENS, L.H. 1954a. The lognormal distribution of the elements, I. Geochimica
et Cosmochimica Acta, 5, 49–73.
Fig. 12. Bubbleplot of ‘maximum data variance’ transformed Cu AHRENS, L.H. 1954b. The lognormal distribution of the elements, II.
concentrations from the Daisy Creek soil survey (Stanley 1984). Geochimica et Cosmochimica Acta, 6, 121–131.
Values were rescaled to the interval 0.001 ↔ 0.999, raised to the AHRENS, L.H. 1957. The lognormal distribution of the elements, III. Geochimica
power of 0.44, and the bubble diameters are proportional to this et Cosmochimica Acta, 11, 205–212.
rescaled value. AITCHISON, J. & BROWN, J.A.C. 1957. The Lognormal Distribution. Cambridge
University Press, Cambridge.
ANSCOMBE, F.J. 1948. The transformation of Poisson, binomial and negative
the concentration ranges defined by thresholds determined binomial data. Biometrika, 35, 246–254.
from the probability plot analysis depicted in Figure 2. This AUBREY, K.V. 1954. Frequency distributions of the concentrations of
dual representation bestows these bubbleplots with both an elements in rocks. Nature, 174 (4420), 141–142.
AUBREY, K.V. 1956. Frequency distributions of elements in igneous rocks.
interpretation of population membership (the bubble colours) Geochimica et Cosmochimica Acta, 9 (1-2), 83–89.
and the means to evaluate the extent to which this interpret- BARTLETT, M.S. 1936. The square-root transformation in analysis of variance.
ation is founded by the data. As a result, these bubble plots Journal of the Royal Statistical Society, Supplement, 3, 68–78.
facilitate rigorous scientific inquiry and hypothesis testing in the BARTLETT, M.S. 1947. The use of transformations. Biometrics, 3, 39–52.
interpretation of the geochemical patterns under examination. BARRETT, E.C. & CURTIS, L.F. 1992. Introduction to Environmental Remote Sensing
3rd. Chapman and Hall, New York.
When comparing these bubbleplots, the reader is reminded BENNETT, B.M. 1956. On the variance stabilizing properties of certain
to resist founding any judgement regarding the relative success logarithmic transformations, I. Trabajos de Estadistica, 7, 295–297.
of these transformations based on the coherence of the pattern BENNETT, B.M. 1957. On the variance stabilizing properties of certain
observed. Rather, judgement should be based on the charac- logarithmic transformations, II. Trabajos de Estadistica, 8, 69–74.
BIGGERS, J.D. & CLARINGFOLD, P.J. 1954. The angular transformation in
teristics that the various transformations imparted to the quantal analysis. Biometrics, 10, 570.
transformed data. BLOM, G. 1954. Transformation of the binomial, negative binomial, Poisson
and chi-square distribution (correction: Biometrika 43, 235). Biometrika, 41,
302–316.
CONCLUSIONS BOX, G.E.P. & COX, D.R. 1964. An analysis of transformations. Journal of the
Transformation of concentration data into a new variable which Royal Statistical Society, Series B, 26, 211–243.
CHAO, T.T. 1984. Use of partial dissolution techniques in geochemical
exhibits a normal, near-normal or even symmetric frequency exploration. Journal of Geochemical Exploration, 20, 101–135.
distribution is not generally possible because of the multi-modal CHAPMAN, R.P. 1977. Log transformations in exploration geochemistry.
character of many/most geochemical datasets. As a result, Mathematical Geology, 9, 194–198.
transformation for this purpose is generally misguided and CHAYES, F. 1962. Numerical correlation and petrographic variation. Journal of
Geology, 70, 440–452.
unlikely to be successful. CLARINGFOLD, P.J., BIGGERS, J.B. & EMMENS, C.W. 1953. The angular
However, transformation of concentration data to obtain transformation in quantal analysis. Biometrics, 9, 467–483.
new variables that exhibit other optimal properties is warranted CLARK, I. 1979. Practical Geostatistics. Applied Science Publishers, London.
on a variety of theoretical grounds. Use of a power transform to COHEN, A.C. 1991. Truncated and Censored Samples. Marcel Dekker, New York.
CURTIS, J.H. 1943. On transformations used in analysis of variance. Annals of
maximize the variance of concentration data that has been Mathematical Statistics, 14, 107–132.
rescaled to within the interval 0 ↔ 1 is: (1) theoretically DAVIS, J.C. 1986. Statistics and Data Analysis in Geology. John Wiley & Sons, New
desirable, (2) can be undertaken using a simple spreadsheet York.
78 C. R. Stanley
DE WIJS, H.J. & MEISCH, A.T. 1979. Universal data transformation. Professional RICHARDS, J.C. 1993. Remote Sensing Digital Image Analysis: An Introduction 2nd
Paper. US Geological Survey. edn. Springer-Verlag, New York.
DOLBY, J.L. 1963. A quick method for choosing a transformation. ROSE, A.W., HAWKES, H.E. & WEBB, J.S. 1979. Geochemistry In Mineral
Technometrics, 5, 317–326. Exploration 2nd edn. Academic Press, New York.
DRAPER, N.R. & COX, D.R. 1969. On distributions and their transformations RUSSELL, J.K. & NICHOLLS, J. 1988. Analysis of petrologic hypotheses with
to normality. Journal of the Royal Statistical Society of London, Series B, 31, Pearce element ratios. Contributions to Mineralogy and Petrology, 99, 25–35.
472–476. RUSSELL, J.K. & STANLEY, C.R. 1990a. Theory and application of Pearce element ratios
DRAPER, N.R. & HUNTER, W.G. 1969. Transformations: some examples to geochemical data analysis. Geological Association of Canada, Short Course
revisited. Technometrics, 11, 23–40. Notes, 8.
EISENHART, C., HASTAY, H.W. & WALLIS, W.A. 1947. Inverse sine transfor- RUSSELL, J.K. & STANLEY, C.R. 1990b. A theoretical basis for the development
mation of proportions. In: Techniques of Statistical Analysis. McGraw-Hill, and use of chemical variation diagrams. Geochimica et Cosmochimica Acta, 54
New York. (9), 2419–2431.
ELSTON, R.E. 1961. On additivity in the analysis of variance. Biometrics, 17, SCHNEIDER, H. 1988. Truncated and Censored Samples from Normal Populations.
209–219. Marcel Dekker, New York.
FINNEY, D.J. 1941. On the distribution of a variate whose logarithmic is SINCLAIR, A.J. 1976. Applications of Probability Graphs in Mineral Exploration.
normally distributed. Journal of the Royal Statistical Society of London, Supplement, Special Publication, 4. Association of Exploration Geochemists, Rexdale,
7, 155–161. Ontario, Canada.
FISHER, R.A. 1954. The analysis of variance with various binomial SINCLAIR, A.J. 1991. A fundamental approach to threshold estimation in
transformations. Biometrics, 10, 130–151. exploration geochemistry; probability plots revisited. Journal of Geochemical
FISHER, N.I., LEWIS, T. & EMBLETON, B.J.J. 1987. Statistical Analysis of Spherical Exploration, 41 (1–2), 1–22.
Data. Cambridge University Press, New York. SINCLAIR, A.J. & BLACKWELL, G.H. 2002. Applied Mineral Inventory Estimation.
FLETCHER, W.K. 1981. Analytical Methods in Geochemical Prospecting. Handbook Cambridge University Press, New York.
of Exploration Geochemistry, 1. Elsevier, Amsterdam. STANLEY, C.R. 1984. The Geology and Geochemistry of the Daisy Creek Prospect, A
FREEMAN, M.F. & TUKEY, J.W. 1950. Transformations related to the angular Stratabound Copper–Silver Occurrence in Western Montana. MSc Thesis. The
and the square root. Annals of Mathematical Statistics, 21, 607–611. University of British Columbia.
GADDUM, J.H. 1945. Lognormal distributions. Nature, 156, 463–466. STANLEY, C.R. 1988. PROBPLOT - An interactive computer program to fit mixtures
GHURYE, S.G. 1949. Transformation of a binomial variate for the analysis of of normal (or log-normal) distributions using maximum likelihood optimization
variance. Journal of the Indian Society of Agricultural Statistics, 2, 94–109. procedures. Special Publication, 14. Association of Exploration Geochemists,
HALL, G.E.M. 1992. Geoanalysis (special issue). Journal of Geochemical Rexdale, Ontario, Canada.
Exploration, 44, 1–349. STANLEY, C.R. 1990. Error propagation and regression on Pearce element
HALL, G.E.M. & BONHAM CARTER, G.F. 1998. Selective extractions (special ratio diagrams. In: RUSSELL, J.K. & STANLEY, C.R. (eds) Theory and Application
issue). Journal of Geochemical Exploration, 61, 1–232. of Pearce Element Ratios to Geochemical Data Analysis. Short Course Notes, 8.
HEALY, M.J.R. & TAYLOR, L.R. 1962. Tables for power-law transformations. Geological Association of Canada, 179–216.
Biometrika, 49, 557–559. STANLEY, C.R. 2003a. Estimating sampling errors for major and trace
HOFFMAN, D.D. 1998. Visual Intelligence: How we Create what we See. Norton & elements in geological materials using a propagation of variance approach.
Co., New York. Geochemistry: Exploration, Environment, Analysis, 3 (2), 169–178.
HOWARTH, R.J. & EARLE, S.A.M. 1979. Application of a generalized power STANLEY, C.R. 2003b. THPLOT.M: A MATLAB function to imple-
transformation to geochemical data. Mathematical Geology, 11, 45–62. ment generalized Thompson-Howarth error analysis using replicate data.
HOYLE, M.H. 1973. Transformations - an introduction and a bibliography. Computers and Geosciences, 29 (2), 225–237.
International Statistical Reviews, 41, 203–223. STANLEY, C.R. & MADEISKY, H.E. 1995. Pearce element ratio analysis in
JIZBA, Z.V. 1959. Frequency distributions of elements in rocks. Geochimica et lithogeochemical exploration. In: 17th International Geochemical Exploration
Cosmochimica Acta, 16, 79–82. Symposium. Annual Meeting, Short Course. Society of Exploration
JOSEPH, L. & BHAUMIK, B.K. 1997. Improved estimation of the Box-Cox Geochemists, Townsville, Queensland, Australia.
transform parameter and its application to hydrogeochemical data. STANLEY, C.R. & SINCLAIR, A.J. 1986. Relative error analysis of replicate
Mathematical Geology, 29, 963–976. geochemical data: advantages and applications. In: GeoExpo – 1986.
Exploration in the North American Cordillera. Programs and Abstracts. Regional
KENDALL, M.G. & STUART, A. 1966. The Advanced Theory of Statistics, 3. Griffin,
London. Symposium. Association of Exploration Geochemists, Vancouver, Canada,
KRUSKAL, J.B. 1968. Statistical analysis: transformations of data. In: SILLS, D.L. 77–78.
& MERTON, R.L. (eds) International Encyclopaedia of the Social Sciences. STANLEY, C.R., CARTER, J., CAAZLET, P. & CARROLL, B. 2003. Light hydro-
MacMillan, New York. carbon gas and heavy metal lithogeochemistry as an indicator for
mineralisation in the Irish upper Carboniferous. In: 21st International
LAURENT, A.G. 1963. The lognormal distribution and the translation method:
description and estimation problems. Journal of the American Statistical Geochemical Exploration Symposium. Program with Abstracts, 21. Association
Association, 58, 231–235. of Exploration Geochemists, Dublin, 80.
STEVENS, W.L. 1953. Tables of the angular transformation. Biometrika, 40,
LEVINSON, A.A. 1980. Introduction to Exploration Geochemistry. Applied 70–73.
Publishing, Wilmette, Illinois. THOMPSON, M. 1973. DUPAN 3, A subroutine for the interpretation of
MARK, D.A. 1973. Analysis of axial orientation data, including till fabrics. duplicated data in geochemical analysis. Computers and Geosciences, 4, 333–340.
Geological Society of America Bulletin, 84, 1369–1374. THOMPSON, M. 1982. Regression methods and the comparison of accuracy.
MCLACHLAN, G.J. & BASFORD, K.E. 1988. Mixture Models. Marcel Dekker, New The Analyst, 107, 1169–1180.
York. THOMPSON, M. & HOWARTH, R.J. 1973. The rapid estimation and control of
MIESCH, A.T. 1977. Log-transformation in geochemistry. Mathematical Geology, precision by duplicate determinations. The Analyst, 98, 153–160.
9, 191–194. THOMPSON, M. & HOWARTH, R.J. 1976a. Duplicate analysis in practice – Part
MIESCH, A.T. & KORK, J.O. 1978. Frequency distributions of elements. US 1. Theoretical approach and estimation of analytical reproducibility. The
Geological Survey Professional Paper, 188–189. Analyst, 101, 690–698.
MOORE, C.C., STANLEY, C.R. & GOODWIN, T. 2003. Distribution of zinc and THOMPSON, M. & HOWARTH, R.J. 1976b. Duplicate analysis in practice – Part
cadmium in stream sediments from the West Barney’s River area. 2. Examination of proposed methods and examples of its use. The Analyst,
Antigonish highlands, Nova Scotia. In: Nova Scotia Mining Matters Symposium, 101, 699–709.
Abstracts with Program, 10. THOMPSON, M. & HOWARTH, R.J. 1978. A new approach to the estimation of
MOORE, P.G. 1957. Transformations to normality using fractional powers of analytical precision. Journal of Geochemical Exploration, 9, 23–30.
the variables. Journal of the American Statistical Association, 52, 237–246. TREISMAN, A.M. & GORMICAN, S. 1988. Feature analysis in early vision:
OERTEL, A.C. 1969a. Frequency distributions of element concentrations; 1. Evidence from search asymmetries. Psychological Review, 95, 15–48.
Theoretical aspects. Geochimica et Cosmochimica Acta, 33 (7), 821–831. TUKEY, J.W. 1957. On the comparative anatomy of transformations. Annals of
OERTEL, A.C. 1969b. Frequency distributions of element concentrations; 2. Mathematical Statistics, 28, 602–632.
Surface soils and ferromagnesian minerals. Geochimica et Cosmochimica Acta, TUKEY, J.W. 1958. The true purpose of transformations. Biometrika, 24, 1041.
33 (7), 833–839. VERBYLA, D.L. 1995. Satellite Remote Sensing. Lewis Publishers, New York.
PEARCE, S.C. 1945. Lognormal distributions. Nature, 156, 747. VISTELIUS, A.B. 1960. The skew frequency distribution and the fundamental
PEARCE, T.H. 1968. A contribution to the theory of variation diagrams. law of geochemical processes. Journal of Geology, 68 (1), 1–22.
Contributions to Mineralogy and Petrology, 19, 142–157. VISTELIUS, A.B. 1964. Informational characteristics of frequency distributions
RAZUMOVSKY, N.K. 1941. On the role of the logarithmically normal law of in geochemistry. Nature, 202 (4938), 1206.
frequency distribution in petrology and geochemistry. Doklady Akademii ZUBIN, J. 1936. Note on a transformation function for proportions and
Nauk SSSR, 33 (1), 48–49. percentages. Journal of Applied Psychology, 19, 213–220.