Вы находитесь на странице: 1из 6

The Lens Model

What is the idiographic-statistical approach? Define and describe the following terms:
ra, Ys, Ye, cue(s), Rs, Re, G, ! Describe a lens model stud" in which the #udgment of
$arious e%perts is e$aluated, that is, choose the tas& and #udges so that there are
belie$able cues and both a #udgment and criterion a$ailable and show how the lens model
stud" can show the better from poorer #udges!
Background and Description
'he (ens )odel is a wa" of thin&ing about describing the relations between the
en$ironment and the beha$ior of organisms in the en$ironment! 'he fundamentals were
de$eloped b" *gon +runswi& and populari,ed b" -enneth .ammond through research
on social #udgments!
/ccording to .ammond, +runswi& thought that ps"cholog" should pro$ide descriptions
of beha$ior rather than attempt to disco$er laws of beha$ior! 'hat is, we should proceed
more li&e geographers than ph"sicists! +runswi& also thought t"pical /012/ designs
were b" nature unrepresentati$e of the world and inherentl" ungenerali,able! 3n the
word, things are correlated! 'he two ma#or tenets of the theor" are representati$e design
and the idiographic-statistical approach!
Representati$e Design is concerned with the selection and inclusion of stimulus
conditions in a stud"! 'he argument is that stimuli should be sampled from a population
#ust li&e sub#ects are sampled from a population! We generall" sample people from some
population that the" are to represent, but we onl" present the sub#ects in our stud" with a
$er" small number of stimulus situations! 4or e%ample, suppose we are interested in the
effects that the e%perimenter has on the sub#ects of the stud"! We are interested in the se%
of the e%perimenter on how people respond to a sur$e"! 5o we ha$e a male and a female
e%perimenter administer sur$e"s to groups of sub#ects and loo& for differences in the
sub#ects6 responses! +" ha$ing onl" a single male and a single female e%perimenter, all
the characteristics of these two indi$iduals are confounded with the se% effect! 'o
understand the effects of se% of e%perimenter, we should sample males and females as
e%perimenters, #ust as we sample sub#ects to be respondents!
'he idiographic-statistical approach refers to using models to represent indi$iduals in rich
en$ironments rather than groups in simple en$ironments! 'he idiographic approach
considers differences in beha$ior within indi$iduals as the ob#ects of interest, as opposed
to the nomothetic approach, which treats indi$idual differences as errors! 'he
idiographic statistical approach means that statistical tests should be applied to indi$idual
sub#ects6 beha$ior! 4or e%ample, suppose we gi$e a clinical ps"chologist 789 different,
real )):3 profiles and as& him or her to tell us about the person described b" each
profile! +" the time we are done, we should &now something about the relations between
the profile and the clinician6s comments! 'o ta&e a simple e%ample, we e%pect there to be
some relation between the $alue on the ;cannot sa"; scale and what the clinician sa"s
about the person! 3f we de$eloped this scenario, it would ma&e conceptual sense to tal&
<
about computing a t-test for scales abo$e =9 $erses those less than =9 or to tal& about the
correlation of a scale score with some description of the person (e!g!, the correlation
between the scale 7 >depression? and the probabilit" that the clinician predicts somatic
complaints)!
'he contemporar" $ersion of the (ens )odel loo&s li&e this:
Cues
Judgment Criterion
Achievement
X1
X2
X3
X4
Ys
Ye
rs,1
re,1
rs,4
re,4
3t6s called the lens model because it loo&s a little bit li&e light passing through a con$e%
lens! 3t will help to clarif" the items b" ha$ing a concrete e%ample in mind! 5uppose we
are interested in the salaries of facult" at @ni$ersit"! 5alar" should depend upon the #ob
performance and tenure of a facult" member as well as some other factors, such as
mar&et conditions! We could as& a person to loo& at some information about a facult"
member and to predict that person6s salar"! 3f we are at a public institution, we can also
loo& up the actual salar" in the librar"!
Here are the ideas in the model:
<! 'here are things in the en$ironment that should relate to our dependent $ariable!
'hese things (our independent $ariables) are called cues in lens model terminolog"! 3n
our e%ample, cues might be number of refereed publications, number of boo&s published,
number of boo& chapters, total grant dollars from e%ternal funding agencies, number of
grants applied for, student contact hours, college committee memberships, and so forth!
'he cues are shown in the center of the diagram, and labeled A< through AB!
7! 1ur sub#ect (participant) in the stud" ma&es a #udgment about a person or ob#ect gi$en
the information about the cues! 3n our e%ample, a member of a 5:/' committee (salar",
promotion and tenure) might be as&ed to #udge the salar" of an indi$idual gi$en a $ita!
'he #udgment is represented as Ys in the figure!
7
C! 'here is an actual $alue of the dependent $ariable in the en$ironment! 3n our case,
each facult" member has an actual salar"! 'his is shown as Ye in the figure!
B! 'here is a functional relation between each cue and each Y! 3n the lens model, it is
customar" to tal& about the linear correlation between each cue and the #udgment (e!g!,
rs,<) and also between each cue and the en$ironment (e!g!, re,BD find these on the abo$e
figure)!
8! 3t is customar" to tal& about the correlation between the #udgment and the
en$ironment (that is, rYs, Ye) as the achie$ement inde%, ra!
What we could do is to gi$e a #udge a series of facult" $itas and ha$e the #udge ma&e a
#udgment about the salar" for each facult"! We could also loo& up the salar" for each
facult" member in the librar"! 3f we correlate the #udgments of facult" salar" with the
actual facult" salaries, we would ha$e the achie$ement of the #udge!
4or a second e%ample of the lens model, we could e%amine the e%pert #udgments of
cancer pathologists who e%amine tissue slides for prognostic #udgments! 4or each #udge
we present a series of slides of cancer tissue from former patients who ha$e died! We as&
the #udge to predict how long the patient will li$e from the information in the slide!
'here are se$eral indications of the se$erit" of the disease in each slide, and these are the
cues that the #udge uses to ma&e his or her #udgment! +ecause we &now when the slides
were ta&en and how long the patient li$ed, we also ha$e data on the criterion side! We
can compute the correlation between the pathologist6s #udgment and the actual length of
life to calculate achie$ement!
E! 3f we do such studies for a series of #udges, we will find that some #udges ha$e higher
achie$ement indices than others, that is, some #udges are better calibrated to the
en$ironment than others!
=! +" loo&ing at the correlations between the cues and the #udgment, we can get some
idea of the ;importance; of the cues to the #udge! 3f A< has a high correlation with the
#udgment but A7 does not, we might infer that the #udge appears to pa" more attention to
A< than to A7! F5ome smart people re#ect this position!G /t an" rate, the correlations
describe the relations between the cues and the #udgment! 5imilarl", the correlations
between the cues and the criterion describe the relations between the cues and what
happens to the actual en$ironment!
H! We can use relations between the cues and the #udgment to describe differences in
different #udges! 4or e%ample, our first #udge might show a strong association between
salar" #udgments and number of refereed articles! 1ur second #udge ma" show a smaller
association between salar" and number of refereed articles, but a larger association
between salar" and grant dollars! 5uch differences are useful in showing how people6s
#udgments are similar and different! 'hin& about preferences for unionIlabor contracts
C
that $ar" in automatic salar" ad#ustments for inflation, #ob securit", merit pa", medical
benefits, and $acation time!
J! We can compare the relations between the cues and the #udgment to the relations
between the cues and the criterion! 'his can help us to understand where the human is
well and poorl" calibrated to the en$ironment! 4or e%ample, it ma" turn out that our
#udge shows a strong association between refereed publications and salar", but our
criterion side shows a strong association between grant dollars and actual salar"!
/lthough +runswi& had some good arguments, there was no such thing as multiple
regression when he was ali$e, so nobod" &new how to collect and anal",e the &ind of
data suggested b" the lens model! You, howe$er, are well eKuipped for such an anal"sis
at this point!
Relations between Regression and terms in the Lens Model.
We can collect data to suppl" numbers for each cue for each profile! / single facult"
member would be a profile! 'his facult" member would ha$e se$eral cues represented
(the total number of refereed publications, the total dollar $alue of e%ternal grants, the
a$erage number of committees ser$ed per "ear, etc!)! We would collect data on the
#udgment side and the criterion side! 'hen we could compute a correlation matri% of all
the cues and the two dependent $ariables! 5uch a matri% has all the information we need
to do the regressions of interest! 1n the #udgment side, we ha$e
Y a b X
s i i
i
k
6
= +
=

<
,
which is ordinar" multiple regression! We ha$e a $er" similar eKuation on the criterion
side:
Y a b X
e i i
i
k
6
= +
=

<
0ote that in general, the regression weights a and bi will not be the same on the #udgment
side and the criterion side!
1n the #udgment side, we can find an R-sKuare, which is the proportion of $ariance in the
#udgments due to the cues! 1r we can tal& about the unsKuared correlation,
R r
s Ys Y s
=
, 6
that is, the correlation between actual #udgments and predicted #udgments! 'his Kuantit"
(Rs) tells us about the cogniti$e consistenc" of the #udge! 'o the e%tent that the person
B
ma&es the same #udgment gi$en the same stimulus, Rs will be large! )ore accuratel",
the more consistentl" the cues account for the #udgment, the higher Rs will be!
1n the en$ironment side, we can compute similar Kuantities, such as
R r
e Ye Y e
=
, 6
which is the correlation between the actual criterion $alue and the predicted criterion
$alue! 'his is an inde% of how predictable the en$ironment or tas& is! 3f the en$ironment
is $er" predictable from the cues, Re will be high! 3f most of the $ariance in the
en$ironment comes from sources other than our cues, then Re will be small!
Recall that earlier we tal&ed about the achie$ement inde% ra , which is the correlation
between the #udgment and the criterion! 'he si,e of ra will be affected the the magnitude
of Rs and Re ! 3f there little cogniti$e control or if the en$ironment is not predictable
from the cues, the ra cannot be $er" large, e$en if the #udge is $er" well calibrated! 4or
e%ample, e$en if "ou are a world-class economist, "ou ma" be unable to predict the $alue
of the stoc& mar&et $er" well because most of the $ariance in prices is due to $ariables
"ou cannot measure!
.owe$er, we can remo$e the error or unpredictable $ariance from each side of the model
b" using predicted $alues of the #udgment and criterion instead of the actual $alues! 'hat
is, we can deal with Y6s and Y6e! 3f we compute the $alue of the correlation between the
two predicted $alues, we ha$e
G r
Y s Y e
=
6 , 6
G is called the matching inde%! 'his shows how well the model (or polic") of the #udge
corresponds to the model of the criterion! 3t shows how well the #udge is calibrated to the
criterion when there is perfect consistenc" of both #udge and criterion and in addition that
onl" the cues in the model are considered! 3t is possible that G is large e$en though ra is
small (our world-class economist might be an e%ample)!
)ore specificall", it turns out that when both sides of the model are well described b" a
linear model (no interactions, cur$es, etc!) then:

r GR R
a s e
=
that is, achie$ement (ra, the correlation between the #udgment and the actual en$ironment)
is eKual to the matching inde% (correlation between the perfectl" consistent model of the
#udge and perfectl" consistent model of the en$ironment) multiplied b" or discounted for
the consistenc" of the #udge and the consistenc" of the en$ironment!
3f there is some nonlinearit" in both the criterion side and the #udgment side (e!g!, "ou
ha$e to ha$e both lots of pubs and lots of grant dollars to get a raise, neither is sufficient
b" itself), and the #udge matches the en$ironment, there will be:
8
C r
err s err e
=
> , ?,> , ?
C r
Ys Y s Ye Y e
=
> 6 ?,> 6 ?
'hat is, , the configuralit" inde%, is the correlation between the residuals of the two
regression eKuations! 3f there is some configuralit" ( is not ,ero), then
r GR R C R R
a s e s e
= + < <
7 7
'he term to the right of the plus sign ad#usts the linear part for the configuralit"! We don6t
simpl" add the part because this was the correlation of the residuals, where the $ariance
due to the cues was alread" remo$ed! 'he part to the right of the plus sign is the
co$ariance of the residuals of the two regression eKuations!
E

Вам также может понравиться