Analytic Culture in the U.S. Intelligence Community (PDF) - CIA
Analytic Culture in the U.S. Intelligence Community (PDF) - CIA
Analytic Culture in the U.S. Intelligence Community (PDF) - CIA
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
INTEGRATING METHODOLOGISTS<br />
good as, or better than, an expert at mak<strong>in</strong>g calls about <strong>the</strong> future. In fact, <strong>the</strong><br />
expert does not tend to outperform <strong>the</strong> actuarial table, even if given more specific<br />
case <strong>in</strong>formation than is available to <strong>the</strong> statistical model. 18<br />
There are few exceptions to <strong>the</strong>se research f<strong>in</strong>d<strong>in</strong>gs, but <strong>the</strong>se are <strong>in</strong>formative.<br />
When experts are given <strong>the</strong> results of <strong>the</strong> Bayesian probabilities, for<br />
example, <strong>the</strong>y tend to score as well as <strong>the</strong> statistical model if <strong>the</strong>y use <strong>the</strong> statistical<br />
<strong>in</strong>formation <strong>in</strong> mak<strong>in</strong>g <strong>the</strong>ir own predictions. 19 In addition, if experts<br />
have privileged <strong>in</strong>formation that is not reflected <strong>in</strong> <strong>the</strong> statistical table, <strong>the</strong>y<br />
will actually perform better than does <strong>the</strong> table. A classic example is <strong>the</strong> case<br />
of <strong>the</strong> judge’s broken leg. Judge X has gone to <strong>the</strong> <strong>the</strong>ater every Friday night<br />
for <strong>the</strong> past 10 years. Based on a Bayesian analysis, one would predict, with<br />
some certa<strong>in</strong>ty, that this Friday night would be no different. An expert knows,<br />
however, that <strong>the</strong> judge broke her leg Thursday afternoon and is expected to<br />
be <strong>in</strong> <strong>the</strong> hospital until Saturday. Know<strong>in</strong>g this key variable allows <strong>the</strong> expert<br />
to predict that <strong>the</strong> judge will not attend <strong>the</strong> <strong>the</strong>ater this Friday.<br />
Although hav<strong>in</strong>g a s<strong>in</strong>gle variable as <strong>the</strong> determ<strong>in</strong><strong>in</strong>g factor makes this case<br />
easy to grasp, analysis is seldom, if ever, this simple. Forecast<strong>in</strong>g is a complex,<br />
<strong>in</strong>terdiscipl<strong>in</strong>ary, dynamic, and multivariate task where<strong>in</strong> many variables<br />
<strong>in</strong>teract, weight and value change, and o<strong>the</strong>r variables are <strong>in</strong>troduced or omitted.<br />
Dur<strong>in</strong>g <strong>the</strong> past 30 years, researchers have categorized, experimented, and<br />
<strong>the</strong>orized about <strong>the</strong> cognitive aspects of forecast<strong>in</strong>g and have sought to<br />
expla<strong>in</strong> why experts are less accurate forecasters than statistical models.<br />
Despite such efforts, <strong>the</strong> literature shows little consensus regard<strong>in</strong>g <strong>the</strong> causes<br />
or manifestations of human bias. Some have argued that experts, like all<br />
humans, are <strong>in</strong>consistent when us<strong>in</strong>g mental models to make predictions. That<br />
is, <strong>the</strong> model an expert uses for predict<strong>in</strong>g X <strong>in</strong> one month is different from <strong>the</strong><br />
model used for predict<strong>in</strong>g X <strong>in</strong> a later month, although precisely <strong>the</strong> same case<br />
and same data set are used <strong>in</strong> both <strong>in</strong>stances. 20 A number of researchers po<strong>in</strong>t<br />
17<br />
R. Dawes, D. Faust, and P. Meehl, “Cl<strong>in</strong>ical Versus Actuarial Judgment”; W. Grove and P.<br />
Meehl, “Comparative Efficiency of Informal (Subjective, Impressionistic) and Formal (Mechanical,<br />
Algorithmic) Prediction Procedures.”<br />
18<br />
R. Dawes, “A Case Study of Graduate Admissions”; Grove and Meehl; H. Sacks, “Promises,<br />
Performance, and Pr<strong>in</strong>ciples”; T. Sarb<strong>in</strong>, “A Contribution to <strong>the</strong> Study of Actuarial and Individual<br />
Methods of Prediction”; J. Sawyer, “Measurement and Prediction, Cl<strong>in</strong>ical and Statistical”; W.<br />
Schofield and J. Garrard, “Longitud<strong>in</strong>al Study of Medical Students Selected for Admission to<br />
Medical School by Actuarial and Committee Methods.”<br />
19<br />
L. Goldberg, “Simple Models or Simple Processes?”; L. Goldberg, “Man versus Model of<br />
Man”; D. Leli and S. Filskov, “Cl<strong>in</strong>ical-Actuarial Detection of and Description of Bra<strong>in</strong> Impairment<br />
with <strong>the</strong> Wechsler-Bellevue Form I.”<br />
20<br />
J. Fries, et al., “Assessment of Radiologic Progression <strong>in</strong> Rheumatoid Arthritis.”<br />
65