ArticlePDF Available

Belief in the unstructured interview: The persistence of an illusion

Authors:

Abstract and Figures

Unstructured interviews are a ubiquitous tool for making screening decisions despite a vast literature suggesting that they have little validity. We sought to establish reasons why people might persist in the illusion that unstructured interviews are valid and what features about them actually lead to poor predictive accuracy. In three studies, we investigated the propensity for "sensemaking" - the ability for interviewers to make sense of virtually anything the interviewee says- and "dilution"-the tendency for available but non-diagnostic information to weaken the predictive value of quality information. In Study 1, participants predicted two fellow students' semester GPAs from valid background information like prior GPA and, for one of them, an unstructured interview. In one condition, the interview was essentially nonsense in that the interviewee was actually answering questions using a random response system. Consistent with sensemaking, participants formed interview impressions just as confidently after getting random responses as they did after real responses. Consistent with dilution, interviews actually led participants to make worse predictions. Study 2 showed that watching a random interview, rather than personally conducting it, did little to mitigate sensemaking. Study 3 showed that participants believe unstructured interviews will help accuracy, so much so that they would rather have random interviews than no interview. People form confident impressions even interviews are defined to be invalid, like our random interview, and these impressions can interfere with the use of valid information. Our simple recommendation for those making screening decisions is not to use them. © 2013. The authors license this article under the terms of the Creative Commons Attribution 3.0 License.
Content may be subject to copyright.
Judgment and Decision Making, Vol. 8, No. 5, September 2013, pp. 512–520
Belief in the unstructured interview: The persistence of an illusion
Jason DanaRobyn DawesNathanial Peterson
Abstract
Unstructured interviews are a ubiquitous tool for making screening decisions despite a vast literature suggesting that
they have little validity. We sought to establish reasons why people might persist in the illusion that unstructured inter-
views are valid and what features about them actually lead to poor predictive accuracy. In three studies, we investigated
the propensity for “sensemaking” - the ability for interviewers to make sense of virtually anything the interviewee says—
and “dilution”—the tendency for available but non-diagnostic information to weaken the predictive value of quality
information. In Study 1, participants predicted two fellow students’ semester GPAs from valid background information
like prior GPA and, for one of them, an unstructured interview. In one condition, the interview was essentially nonsense
in that the interviewee was actually answering questions using a random response system. Consistent with sensemak-
ing, participants formed interview impressions just as confidently after getting random responses as they did after real
responses. Consistent with dilution, interviews actually led participants to make worse predictions. Study 2 showed that
watching a random interview, rather than personally conducting it, did little to mitigate sensemaking. Study 3 showed
that participants believe unstructured interviews will help accuracy, so much so that they would rather have random in-
terviews than no interview. People form confident impressions even interviews are defined to be invalid, like our random
interview, and these impressions can interfere with the use of valid information. Our simple recommendation for those
making screening decisions is not to use them.
Keywords: unstructured interview, random interview, clinical judment, actuarial judgment.
1 Introduction
In 1979, an act of legislature suddenly forced the Univer-
sity of Texas Medical School at Houston to admit 50 more
applicants late in the admissions season. The additional
applicants were initially rejected for admission, based
largely on impressions from unstructured interviews in
which each interviewer could ask different questions of
different applicants in whatever way he or she saw fit.
Apparently, the expense of having faculty interview ev-
ery applicant was wasted: at the conclusion of medical
training and one postgraduate year, there were no mean-
ingful differences between the initially rejected and ini-
tially accepted groups of students in terms of attrition,
academic performance, clinical performance, or honors
earned (Devaul et al., 1987). Several large-scale field
studies have provided similar examples of the embarrass-
ingly poor validity of unstructured interviews for screen-
ing decisions (e.g., Bloom & Brundage, 1947; Milstein,
Wilkinson, Burrow, & Kessen, 1981; Carroll, Wiener,
Coates, Galegher, & Alibrio, 1982). More systematic re-
Dawes was thankful for support from NSF Grant SES-0136259.
His unfortunate death in 2010 prevented him from seeing the present
version of this article.
Copyright: © 2013. The authors license this article under the terms
of the Creative Commons Attribution 3.0 License.
Yale University, 135 Prospect St., New Haven CT 06511, Email:
jason.dana@yale.edu.
Carnegie Mellon University.
views in the area of employment decisions likewise show
that unstructured interviews are poor predictors of job
performance, with structured interviews faring somewhat
better (Wiesner & Cronshaw, 1988; Wright, Lichtenfels,
& Pursell, 1989; Huffcutt & Arthur, 1994; McDaniel,
Whetzel, Schmidt, & Maurer, 1994).
Despite the evidence, unstructured interviews remain a
ubiquitous and even predominant tool for many screening
decisions. Studies of human resource executives suggest
that they believe more in the validity of unstructured in-
terviews than other screening methods, even when they
are aware that the evidence suggests that structured as-
sessment is superior (Highhouse, 2008). Academics,
though not professional interviewers, may decide to ac-
cept graduate students or hire faculty based on an infor-
mal 20 minute chat, countermanding substantial aggre-
gated and/or statisticized data comparing the candidate to
others (test scores and GPAs in the case of students, C.V.’s
in the case of faculty). Recently, Wake Forest Univer-
sity stopped requiring standardized tests for undergradu-
ate admissions, moving to a system in which every appli-
cant is eligible for an unstructured interview that figures
into the admissions decision in a “holistic”, non-numeric
manner (Highhouse & Kostek, 2013).
Since Meehl’s (1954) seminal book on the clinical-
statistical controversy for making predictions, a large
literature studying human vs. statistical prediction has
shown that the statistical method is nearly always equal
512
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
513
or superior to clinical judgment (see Grove, Zald, Lebow,
Snitz, & Nelson, 2000). Further, many common objec-
tions to the interpretation of this evidence have been thor-
oughly discussed and refuted (see, e.g., Dawes, Faust, &
Meehl, 1989; Grove & Meehl, 1996). We do not seek to
rehash these debates or provide merely another example
of clinical judges failing to outperform a statistical rule.
Rather, we seek to establish some reasons why people
might persist in the illusion that unstructured interviews
are valid and why they can harm predictions. Specifically,
we explore how the inevitably noisy signals in an inter-
view dilute the decision maker’s potential use of valid in-
formation and how interviewers can form falsely coherent
impressions from virtually anything the interviewee says
or does.
Extending the literature in these ways is important for
at least two reasons. First, it is reasonable to think that,
while unstructured interviews are not particularly predic-
tive, they will not hurt accuracy. At least, we are aware
of no prior evidence that unstructured interviews decrease
accuracy, e.g., by way of studying the same decision mak-
ers with and without access to interviews. This point be-
comes important because the issue is often raised that
the interview conveys benefits beyond its predictive va-
lidity. For example, candidates who go through an in-
terview process may have an increased sense of commit-
ment and more likelihood to accept an offer. Thus, if
one can get a benefit from conducting unstructured in-
terviews and the interviews do not make one’s judgment
worse, it would seem riskless to use them for certain pur-
poses. Indeed, this point was made without rebuttal in a
discussion of the Wake Forest decision on the Society for
Judgment and Decision Making’s mailing list (2008–9,
see the “search” at http://www.sjdm.org). We will pro-
vide evidence that exposure to unstructured interviews
can indeed harm judgment.
Second, many have the feeling that the unstructured in-
terview is the best way to uncover important information
that is special to a candidate. Particularly, candidates may
possess some personality traits at the extremes of the dis-
tribution that might make them ideal or unsuitable. These
unusual cues, akin to what Meehl (1954) called “broken
legs”, could immediately remove a candidate from con-
sideration or catapult a candidate ahead of others. In-
deed, even a structured interview might not help for this
purpose if the interviewer does not have in mind what
this broken leg factor would be ahead of time. The logic
of broken leg cues was addressed by Meehl, who pointed
out that, if people were actually good at spotting broken
legs that statistical rules miss, then they would be more
accurate than statistical rules. This argument, however,
has an important flaw in that it assumes that all errors and
successes are equally important. If an interviewer is espe-
cially concerned with making some kinds of errors, such
as missed broken legs, unstructured interviews could, in
theory, be highly valuable in avoiding the most important
errors, which may outweigh making a few more errors on
the more mundane cases.
Basic psychological research, however, gives us rea-
son to believe that unstructured interviews can harm judg-
ment and reason to doubt that interviewers will be suffi-
ciently adept at spotting special information, and not false
alarms, about a candidate.
1.1 Can interviews hurt?
Access to an interview could hurt predictive accuracy be-
cause exposure to non-diagnostic information is known to
dilute valuable information. Unstructured interviews ex-
pose interviewers to so many casual observations about
the interviewee that have little or unknown diagnostic-
ity that interviewers cannot help but get more informa-
tion than they can use and thus, they must ignore some
cues. Research on the “dilution effect” (e.g., Nisbett,
Zukier, & Lemley, 1981; Zukier, 1982; Peters & Roth-
bart, 2000) shows that rather than just being ignored, ex-
traneous information reduces reliance on good informa-
tion. It is perhaps no coincidence that the stimuli for
the earliest dilution effect studies, which included am-
ple material judged non-diagnostic by study participants,
came from interview snippets (Nisbett, Zukier, & Lem-
ley, 1981). Because making good social judgments of-
ten requires ignoring information and relying on simple
rules, cognitive traits that might normally be construed
as positive, such as complexity of thought and need for
cognition, can actually be detrimental to accuracy (Rus-
cio, 2000). In the presence of quality cues, most of the
interview could serve as a distraction.
1.2 Can interviewers reliably extract spe-
cial information from unstructured in-
terviews?
Although too much irrelevant information dilutes the pre-
diction process, it can also lead to unwarranted confi-
dence due to sensemaking. People seek to impose or-
der on events, so much so that they often see patterns in
random sequences (Gilovich, 1991). As such, even the
noisiest interview data are readily translated into a good
story (see Dawes, 2001, Chapter 7) about the intervie-
wee. Just as one can, post hoc, fit a “significant” sta-
tistical model to pure noise, interviewers have too many
degrees of freedom to build a coherent story of intervie-
wees’ responses. If the interviewee gives a response that
is inconsistent with the interviewer’s impression, the in-
terviewer can dynamically reformulate that impression,
perhaps asking follow up questions until hearing a set of
responses that confirm an impression. Without structure,
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
514
interviewers may not ask questions intended to discon-
firm these impressions. because people are inclined to
seek information that confirms their hypotheses or avoid
what might disconfirm them (Devine, Hirt, & Gehrke
1990; Sanbonmatsu, Posavac, Kardes, & Mantel, 1998).
As an example, someone known to one of the authors
was given a panel interview for a potential job. Arriving 5
minutes early, she was immediately called into the room,
where the interview went quite succesfully and she was
offered the job on the spot. During the postmortem dis-
cussion, one of the interviewers was impressed by how
well she composed herself after showing up 25 minutes
late to the interview! Apparently, she had been misin-
formed that the time of the interview was 30 minutes after
the hour, rather than on the hour as the panel expected,
and she remained composed because she did not know
she was late. Interestingly, nothing that the panel asked
effectively tested this impression that the candidate was
unusually composed under (what they believed were) the
circumstances—they never learned that she thought she
was early. Further, there are many other less flattering im-
pressions than “composed” that could also have explained
a lack of concern over being 25 minutes late, including
flippant or arrogant.
The ability to sensemake combined with the tendency
for biased testing allows unstructured interviewers to feel
they understand an interviewee almost regardless of the
information they receive. Unfortunately, a feeling of un-
derstanding, while reassuring and confidence-inspiring,
is neither sufficient nor necessary for making accurate as-
sessments (Trout, 2002). Further, there is empirical evi-
dence that confidence and accuracy are often poorly re-
lated in interpersonal prediction contexts (Dunning, Grif-
fin, Milojkovic, & Ross, 1990; Swann & Gill, 1997) and
confidence has been shown to increase with information
even in situations where accuracy does not (e.g., Ander-
sson, Edman, & Ekman, 2005; Hall, Ariss, & Todorov,
2007). We suggest that people can feel confident in
the validity of unstructured interview impressions even
if they are worthless.
We experimentally tested the roles of dilution and
sensemaking in the context of using unstructured inter-
views to predict social outcomes. Study participants
predicted the semester GPAs of other students based
on biographical information including GPA prior to the
semester in question and in some cases, an unstructured
interview. In some conditions, the interviews were non-
sense for the task at hand because the interviewee secretly
used a random responding system to answer questions,
literally providing random answers to questions that were
independent of the interviewee’s natural response. Con-
sistent with dilution, participants’ GPA predictions were
more accurate without the unstructured interview and less
accurate than had they simply predicted that semester
Table 1: Interviewees’ prior and obtained GPAs.
Interviewees
Study1 1 2 3 4 5 6 7
Prior GPA 3.32 3.28 3.24 3.23 2.95 2.84 2.81
Obtained GPA 3.80 3.08 3.71 3.34 2.68 2.69 3.35
Study 2 1 2 3 4 5 6 7 8
Prior GPA 3.69 3.38 3.29 3.29 3.23 3.05 2.83 2.65
Obtained GPA 3.83 3.80 4.00 2.83 2.65 3.59 3.00 3.31
GPA would be equal to prior GPA, a strong cue that they
were given before making predictions. Consistent with
sensemaking, participants who unknowingly conducted
random interviews were just as likely to indicate in post-
interview surveys that they got good information as those
who conducted accruate interviews. This is the first ev-
idence we know of that unstructured interviews can be
worse than invalid; they can actually decrease accuracy.
Yet, while interviews were harmful in this context, even
our nonsense interviews promoted a feeling of confidence
in the interview impression.
2 Study 1
To explore whether interviews could dilute judgments
and make them worse, we had student participants predict
the semester grade point average (GPA) of two other stu-
dents, one prediction with biographical information (de-
scribed below) and an interview, the other with just bio-
graphical information. To explore whether interviewers
sensemake, we developed a random responding system
that the interviewees could use during the interview to
see whether it would perturb predictive accuracy or sub-
jective confidence in interview impressions.
2.1 Method
2.1.1 Interviewers and interviewees
Interviewers were 76 undergraduate students at Carnegie
Mellon University who were recruited through campus
advertising and paid for their participation. We employed
five Carnegie Mellon undergraduates (two female) as per-
manent interviewees. The interviewees ranged in age
from 18 to 22, and represented multiple races, majors,
and class standings. Two of the interviewees worked for
two semesters, creating a total of 7 different semester
GPAs to be predicted. Their prior cumulative GPAs and
GPAs for the semester to be predicted are listed in Table
1.
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
515
2.1.2 Procedures
Participants were introduced to a randomly assigned in-
terviewee and asked to conduct a 20 minute interview
with the goal of predicting the interviewee’s GPA for a
given semester. An experimenter remained in the room
during the interview to track time and answer any ques-
tions about the task. Prior to interviewing, participants
were told the interviewee’s age, major, class standing,
and course schedule for the semester to be predicted. Par-
ticipants were offered a break 10 minutes into the inter-
view, during which they could formulate more questions
to ask.
After the interview, the interviewee was excused and
participants made their GPA predictions, which were to
be kept confidential from the interviewee. Before mak-
ing their predictions, participants were given the intervie-
wee’s cumulative GPA prior to the target semester and in-
formed that prior cumulative GPA by itself was the best
statistical model for predicting GPAs at this institution
(Lewis-Rice, 1989). After the GPA prediction, partici-
pants answered a brief questionnaire (Table 3) probing
whether they got to know the interviewee and whether
the interview provided useful information. Finally, 68
participants predicted the semester GPA for another tar-
get whom they did not interview using only the target’s
background information and cumulative GPA prior to the
semester in question.
2.1.3 Interview conditions
The structure of the interview varied according to the par-
ticipant’s random assignment to one of three conditions.
In the accurate condition (n= 25), participants could ask
only closed-ended questions, i.e., “yes or no” or “this or
that” questions. Interviewees answered these questions
accurately. The random condition (n= 26) was similar
except that after the midway break, the interviewee se-
cretly responded on a pseudo-random basis. Interviewees
noted the first letter in the last two words of each question
and classified them as category 1 (letters A through M) or
category 2 (N through Z). If both letters belonged to the
same category, the interviewee answered yes (or took the
first option of a “this” or “that” question) and otherwise
answered no. This system tends to equalize the frequency
of yes and no answers as follows: Call the proportion of
words that the interviewer samples from category 1. A
yes answer occurs if both of the last 2 words are category
1, which occurs with probability p2, or both category 2,
which occurs with probability (1 p)2. The total proba-
bility of a yes response, p2+ (1 p)2, is always closer to
.5 than pitself. By employing random response, whether
an interviewee’s response did or did not match the in-
terviewer’s expectations or confirm the interviewer’s im-
pression was simply a matter of chance.
A lack of a significant difference in accuracy or sur-
vey answers between the accurate and random conditions
might not reflect sensemaking on the part of the inter-
viewer, but rather the deficient quality of all closed-ended
interviews. That is, if closed-ended interviews are too low
in quality for this task, any differences between random
and accurate interviews might be muted. In that case,
we would expect predictions to be better and ratings to
be higher if participants could ask questions and demand
answers in any way they wanted. To rule out this expla-
nation, we also conducted a natural condition (n= 25)
in which no closed-ended constraint was placed on the
interviewer’s questions.
2.2 Results
The validity (correlation with actual outcomes) of GPA
predictions following interviews (r=.31) was indeed
significantly lower than the validity of using prior cumu-
lative GPA alone (r=.65, t(73) = 3.77, p < .05,d
=.43; Hotelling’s method for dependent r with Williams
correction), information participants had when making
their predictions. As dilution predicted, our unstructured
interview did not prove helpful in light of an already
strong cue of prior GPA. While worse than using prior
cumulative GPA, a validity of .31 compares favorably
to that of unstructured employment interviews for pre-
dicting job performance (Campion, Palmer, & Campion,
1997). Comparing the success of our interviewers with
employment screeners is not totally appropriate—GPAs
could be easier to predict becuase GPA is more reliable
than measures of job performance or because job screen-
ers do not have information as valid as prior GPA when
making decisions. Still, the validities our interviewers
were able to obtain provide at least some evidence that
they were not completely deficient at the task.
Some of our interviewees were initially concerned that
the random interview would break down and be revealed
to be nonsense. No such problems occurred and random
interviews proceeded much as interviews in our other
conditions. Further, random responding did not make
interviewers less accurate: Only the validity in the ran-
dom condition (r=.42) was significantly different from
zero, while validities in the accurate (r=.20) and natu-
ral (r=.29) conditions were not, though validities from
these 3 conditions did not differ significantly from each
other. A plausible concern is that random condition par-
ticipants might have relied more on prior GPA because
the interview was bad, thus inflating accuracy in the ran-
dom condition because prior GPA was a strong predic-
tor. This was not the case; GPA predictions were no
more correlated with prior GPAs in the random condi-
tion (r=.54) than in the accurate (r=.53) or natural
condition (r=.67).
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
516
Table 2: Regression analyses of the accuracy of GPA predictions. (Dependent Variable: Predicted GPA.)
Study 1 Study 2 All Study 1 Study 2 All
Predictors: (1) (2) (3) (4) (5) (6)
Actual GPA 0.51∗∗ 0.190.09 0.21 22.67 21.07
(0.08) (0.09) (0.016) (1.72) (17.42) (19.63)
Access to interview 0.99∗∗ 0.08 0.54
(0.35) (0.42) (0.22)
Actual GPA×interview 0.280.02 0.15
(0.11) (0.12) (0.07)
Q1 0.37 0.73∗∗ 0.42
(0.56) (0.27) (0.25)
Q2 0.11 0.91∗∗ 0.64
(0.48) (0.28) (0.25)
Actual GPA×Q1 0.13 0.190.11
(0.17) (0.08) (0.07)
Actual GPA×Q2 0.03 0.26∗∗ 0.18
(0.15) (0.08) (0.07)
Target dummies No No Yes1No No Yes1
Clustering at subject level Yes Yes Yes No No No
p < 0.05,∗∗ p < 0.01.
1. Two interviewees obtained the same GPA when samples are combined, thus dummies represent-
ing each interviewee are included.
2. Eight participants who did not make a prediction without an interview are excluded.
Although participants judged the interview to be some-
what informative, GPA predictions were actually less
accurate with interviews (r=.31) than without them
(r=.61). Because these correlations involved different
judgments by the same participant, we tested the differ-
ence using regression with participant random effects, re-
gressing GPA predictions on actual GPA, a dummy = 1 if
an interview was conducted, and the interview×GPA in-
teraction. The results in column 1 of Table 2 indicate that
the interaction term was negative and significant, mean-
ing that predictions were indeed significantly less corre-
lated with outcomes when an interview was performed.
Table 3 shows that the mean agreement with the state-
ments “I am able to infer a lot about this person given the
amount of time we spent together” (accurate = 2.72, nat-
ural = 2.80, random = 2.85) and “From the interview,
I got information that was valuable in making a GPA
prediction” (accurate = 3.00, natural = 3.12, random
= 3.31) was similar across all conditions, with no sig-
nificant differences emerging (F(2,73) =.233 and 1.714,
respectively). While comparisons of accuracy and sub-
jective impressions yielded null results between random
and truthful interviews, in both cases the direction was
Table 3: Study 1 post prediction questionnaire. Mean
agreement with statements on a 4-point Likert scale (1 =
disagree, 4 =agree) with standard errors in parentheses.
Accurate Random Natural
I am able to infer a lot about this person given the
amount of time we spent together.
2.72 (.68) 2.85 (.47) 2.80 (.58)
From the interview, I got information that was
valuable in making a GPA prediction.
3.00 (.65) 3.31 (.55) 3.12 (.60)
“wrong”—prediction accuracy and impressions of use-
fulness trended higher for random interviews. Agreement
with these questions did not significantly modulate accu-
racy. Column 4 of Table 2 reports the results of regress-
ing GPA predictions on obtained GPA, questions 1 and 2,
and their interactions. Neither question interacted signif-
icantly with GPA.
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
517
2.3 Discussion
Consistent with sensemaking, a random interview did not
perturb either GPA predictions or subjective impressions
about the quality of the interview or the extent to which
they got to know the interviewee. Consistent with di-
lution, a single, strong cue—past GPA—predicted bet-
ter than participants themselves, even though they had
this information. Further supporting dilution, participants
made better predictions without an interview than with
one. While participants generally agreed that they got
useful information from interviews, interviews signifi-
cantly impaired accuracy in this environment.
Perhaps one reason that participants felt interviews
were useful and made sense of them even when they were
random is that they conducted them. The person conduct-
ing the interview controls the questions, which could be
important to at least the sensemaking part of our results.
If participants merely watched the interviews, rather than
conducting them, would they be less prone to either or
both effects? By having participants watch pre-recorded
interviews, we could also directly assess whether they can
tell random from accurate by informing them of the pos-
sibility that the interview they watched was random and
asking them to guess which type they saw.
3 Study 2
Rather than conducting the interview themselves, partic-
ipants in Study 2 watched a pre-recorded interview that
another student had conducted. Because this procedure
did not allow participants to ask their own questions, they
could be less prone to confirming their own theories of the
interviewee and thus less prone to sensemaking. If so, we
might expect participants to be able to discern random
from accurate interviews.
3.1 Method
3.1.1 Participants and interviewees
Participants were 64 undergraduate students at Carnegie
Mellon University who were recruited through cam-
pus advertising and paid for their participation. Eight
Carnegie Mellon undergraduates (5 female) participated
as interviewees and consented to having two interview
sessions recorded (one random, one accurate) as stimuli
for the study. Interviewees ranged in age from 19 to 21,
and again represented multiple races, majors, and class
standings. Table 1 lists their prior and obtained GPAs.
3.1.2 Procedures
Procedures were the same as in Study 1, with the fol-
lowing exceptions. Prior to conducting the experimental
Table 4: Study 2 mean Likert responses (5 =strongly
agree) to post experimental questions by condition (stan-
dard errors in parentheses).
Accurate Random
I am able to infer a lot about this person given the
interview I just watched.
3.47 (0.92) 3.47 (1.08)
From the interview I just watched, I got information
that was valuable in making a GPA prediction.
3.66 (0.94) 3.75 (0.98)
sessions, we video-recorded 16 interviews (one accurate
and one random for each interviewee, natural interviews
were not used) conducted similarly to Study 1, except
that the random interview was now random responding
throughout instead of after the break. Participants were
randomly assigned to watch one of the 16 interviews via
computer interface and predict the interviewee’s GPA for
a given semester. Each interview was randomly assigned
to four different participants. The post interview ques-
tion wording was amended slightly (Table 4) to reference
the interview that was watched and the Likert-type scale
now ranged from 1 to 5 and included a “neither agree
nor disagree” point. After the post-interview question-
naire, participants were informed that their interview was
randomly drawn from a pool containing half random in-
terviews and asked to guess whether it was random or
accurate.
3.2 Results
GPA predictions were about equally correlated with ac-
tual GPAs as they were in Study 1 (r=.28). For
this sample of interviewees, however, prior GPA was
not as predictive of semester GPAs as it was in Study
1 (r=.37) and was not significantly more accurate
than participant predictions. Thus, this form of dilution
was not present in Study 2. Though the procedure in
Study 2 is somewhat different, it is informative to com-
bine results with Study 1 to see if, overall, dilution is
present, especially considering that the sample of inter-
viewees on which this result somewhat depends is small.
Combining both studies, prior GPA alone predicts sig-
nificantly better than our participants do with interviews
(t(137) = 2.59, p < .05,d=.44).
Even though participants did not control the course
of the interview in Study 2, subjective impressions were
again unperturbed by random responding. Table 4 shows
that mean agreement with the statements “I am able to
infer a lot about this person given the interview I just
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
518
Table 5: Frequency of accurate/random guesses by inter-
view type.
Random Accurate Total
Guess Random 13 3 16
Guess Accurate 19 29 48
Total 32 32 64
watched” (accurate = 3.47, random = 3.47) and “From
watching the interview, I got information that was valu-
able in making a GPA prediction” (accurate = 3.66, ran-
dom = 3.75) was again similar across conditions, with
agreement in the random condition again being equal or
higher. As in Study 1, GPA predictions relied on prior
GPA about the same for random (r=.58) and accurate
(r=.55) interviews.
We again tested for interactions between answers to the
post-experimental questionnaire and predictive accuracy.
As can be seen in column 5 of Table 2, both interactions
were significant in Study 2. Interestingly, the coefficients
on each question and on each interaction had opposite
signs, such that feeling one is able to infer a lot about the
interviewee negatively impacted accuracy, while feeling
one had gotten good information from the interview posi-
tively impacted accuracy. When studies 1 and 2 are com-
bined, shown in column 6 of Table 2, only the interaction
between the valuable information question and accuracy
remained significant. This result seems somewhat para-
doxical: Access to interviews overall decreased accuracy,
but given that a participant had access to an interview,
greater agreement that one had gotten valuable informa-
tion from the interview increased accuracy. This result
raises the question of whether our finding of poorer accu-
racy following interviews is driven by a subset of partic-
ipants who did not feel they got useful information from
the interview, but used it anyway. The effects are not so
simple, however. For example, looking at only those par-
ticipants who agreed with the valuable information ques-
tion (answers of 4 or 5), the validity of predictions was
only .29. Thus, there is no simple main effect such that
those who felt they got valuable information from the in-
terview were more accurate.
Table 5 tabulates participants’ judgments of whether
they saw an accurate or random interview across in-
terview type. Participants correctly classified 66% of
the interviews, significantly better than chance (χ2
(1) =
8.33, p < .01). This result, however, was largely driven
by the participants judging all interviews to be accurate:
accurate interviews were nearly always judged to be ac-
curate (29/32), and more than half of random interviews
were judged accurate (19/32). Indeed, the tendency to
judge all interviews accurate was significantly stronger
than the tendency to be correct (McNemar’s test, χ2
(1) =
11.63, p < .001). Thus, while participants have some
skill in identifying accurate from random interviews, they
also see most interviews as probably being accurate, in-
dicating some degree of sensemaking. Whether partic-
ipants were accurate in this judgment, whether partici-
pants judged their interview to be accurate, and whether
participants correctly judged their interviews to be accu-
rate all did not interact with accuracy of GPA predictions
(all p > .30).
Although participants who merely watched interviews
were still prone to sensemaking, their predictions were
not more accurate without an interview, inconsistent with
dilution (see column 2 of Table 2). Dilution did not
hold for these participants, however, largely because no-
interview predictions, which were not handled differently
in this study, were much less accurate (r=.26) than in
Study 1, while predictions following all interviews were
about as accurate as in Study 1 (r=.28). Of course,
while interviews did not make predictions worse, they
also did not make them significantly better. Our two stud-
ies thus fail to indicate any incremental validity from in-
terviews, and Study 1 suggests a decrement in validity. At
best, one can say that watching an interview did not hurt
but conducting one did. While the procedures are differ-
ent across the studies, it is again informative to combine
the data and repeat our test of predictive validity with
and without an interview. Column 3 of Table 2 shows
that the interview×GPA interaction is negative and sig-
nificant; thus, interviews are overall negatively associated
with accuracy.
3.3 Discussion
Watching interviews did little to mitigate sensemaking;
participants’ predictive accuracy and subjective impres-
sions were similar after watching random and accurate
interviews, and they were more likely to see interviews
as accurate whether they were or not. One objection to
our interpretation of Studies 1 and 2 is the presence of
experimental demand to use interviews. Because we took
the trouble of having participants conduct or watch inter-
views for the majority of the study’s duration, it is not
unreasonable to assume that participants felt they should
use the interview, regardless of their feelings about its va-
lidity. Of course, such implicit demands are also present
in real-world settings in which one is forced to conduct an
interview for screening purposes. Still, one may wonder
whether participants believed that interviews aided accu-
racy, a question we explore in Study 3.
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
519
Table 6: Dominance matrix in which cell frequencies
are the number of participants who ranked the column
method better than the row method.
Natural Accurate Random No interview Total
Natural – 36 12 13 61
Accurate 128 28 22 178
Random 153 136 68 357
No interview 152 142 96 390
4 Study 3
4.1 Method
One hundred sixty nine Carnegie Mellon University stu-
dents completed this task as part of a longer session. Par-
ticipants were given descriptions of the methods and con-
ditions used in Study 1 (except that the random condition
was full random as in Study 2), including the information
that participants were given prior GPA and then asked to
predict a student’s GPA from a given semester. Partici-
pants in Study 3 were then asked to rank the interview
types (including no interview) in terms which they would
like to have to make their predictions as accurately as
possible. That is, they were essentially asked about the
incremental validity of each type of interview.
4.2 Results
The modal accuracy rankings for first through last place
were natural interview first, followed by accurate, ran-
dom, and no interview, respectively, making the predic-
tion type for which participants were the most accuate
in Study 1 the least favored. This ranking was also the
single most common, chosen by 57 (33%) of our par-
ticipants. No participant ranked the natural condition
last, while 56% of participants ranked no interview last.
The dominance matrix in Table 6 depicts all aggregate
pairwise preferences by reporting how many participants
ranked the interview type in the column over the type in
the row. Even random interviews, which by definition
contain misleading information, were preferred to no in-
terview by 96 participants (57%). By ranking the random
interview ahead of no interview, a simple majority of our
participants showed that they did not anticipate a dilution
effect: Apparently, they believed that random interviews
contained some useful information that all of the useless
information would not drown out. Thus, while interviews
do not help predict one’s GPA, and may be harmful, our
participants believe that any interview is better than no
interview, even in the presence of excellent biographical
information like prior GPA.
5 Discussion
We set out to examine whether unstructured interviews
could harm predictive accuracy and whether interview-
ers would believe they garnered useful information from
the interview regardless of its quality. Consistent with
dilution, Study 1 showed that participants were better
at predicting other students’ GPAs when they were not
given access to an unstructured interview in addition to
background information. Further, participants predicted
worse than if they had used prior GPA alone, information
they were given before making their predictions. Con-
sistent with sensemaking, participants were just as able
to make coherent impressions when the interviewee re-
sponded randomly, both in terms of the accuracy of their
predictions and their confidence in their subjective im-
pressions. Study 2 showed that even when watching
rather than conducting an interview, participants were
still somewhat prone to sensemaking. Finally, Study 3
showed that participants believe that interviews will help
in this context, so much so that they rate random inter-
views as being more helpful than no interview, which
was, in fact, the best way to make predictions in Study
1 and as good as other methods in Study 2.
Our findings suggest a rethinking of the meaning of
interview validity. The validity of predictions made by
interviewers or by numerically incorporating interviews
into a model is uninformative unless it can be directly
compared to predictions made by the same methods with-
out an interview. On its face, the validity of our par-
ticipants’ predictions following unstructured interviews
looks respectable (r=.31 in Study 1 and r=.28 in
Study 2), yet these same participants were able to pre-
dict better when they did not have access to an inter-
view, and could have predicted better still if they just used
prior GPA. It may be the case that for many screening
decisions, there are one or two cues that are very impor-
tant and could be garnered from nearly any interview (or
without one), and that these cues predict better by them-
selves than the clinical judges who have access to them.
The substantial literature on interviews for employment
screening, which already indicates that unstructured in-
terviews are not particularly good, may thus even be over-
stating the validity of unstructured interviews. Our evi-
dence is experimental and compares the same judges with
and without access to an interview. To our knowledge,
there is little prior evidence of this kind.
In addition to the vast evidence suggesting that un-
structured interviews do not provide incremental validity,
we provide direct evidence that they can harm accuracy.
Because of dilution, this finding should be especially ap-
plicable when interviewers already have valid biograph-
ical information at their disposal and try to use the un-
structured interview to augment it. Because of sensemak-
Judgment and Decision Making, Vol. 8, No. 5, September 2013 Belief in unstructured interviews
520
ing, interviewers are likely to feel they are getting use-
ful information from unstructured interviews, even when
they are useless. Because of both of these powerful cog-
nitive biases, interviewers probably over-value unstruc-
tured interviews. Our simple recommendation for those
who make screening decisions is not to use them.
References
Andersson, P., Edman, J., & Ekman, M. (2005). Predict-
ing the world cup 2002 in soccer: Performance and
confidence of experts and non-experts. International
Journal of Forecasting, 21, 565–576.
Bloom, R. F., & Brundage, E. G. (1947). Predictions of
success in elementary school for enlisted personnel. In
D. B. Stuit (Ed.), Personnel research and test develop-
ment in the Naval Bureau of Personnel, pp. 233–261..
Princeton: Princeton University Press.
Campion, M. A., Palmer, D. K., & Campion, J. E. (1997).
A review of structure in the selection interview. Per-
sonnel Psychology, 50, 655–702.
Carroll, J. S., Wiener, R. L., Coates, D., Galegher, J., &
Alibrio, J. J. (1982). Evaluation, diagnosis, and predic-
tion in parole decision making. Law & Society Review,
17, 199–228.
Dawes, R. M. (2001). Everyday irrationality: How pseu-
doscientists, lunatics, and the rest of us systematically
fail to think rationally. Boulder, Colorado: Westview
Press.
DeVaul, R., Jervey, F., Chappell, J., Caver, P., Short, B.,
& O’Keefe, S. (1987). Medical school performance
of initially rejected students. Journal of the American
Medical Association, 257, 47–51.
Devine, P., Hirt, E., & Gehrke, E. (1990). Diagnostic
and confirmation strategies in trait hypothesis testing.
Journal of Personality and Social Psychology, 58, 952–
963.
Garfinkel, H. (1967). Common sense knowledge of social
structures: The documentary method of interpretation
in lay and professional fact finding. In H. Garfinkel
(Ed.), Studies in ethnomethodology, pp. 76–103. En-
glewood Cliffs, NJ: Prentice-Hall.
Gilovich, T. (1991). How we know what isn’t so: The
fallibility of human reason in everyday life. New York:
Free Press.
Hall, C. C., Ariss, L., & Todorov, A. (2007). The illusion
of knowledge: When more information reduces accu-
racy and increases confidence. Organizational Behav-
ior and Human Decision Processes, 103, 277–290.
Highhouse, S. (2008). Stubborn reliance on intuition and
subjectivity in employee selection. Industrial and Or-
ganizational Psychology: Perspectives on Science and
Practice, 1, 333–342.
Highhouse, S., & Kostek, J. A. (2013). Holistic assess-
ment for selection and placement. In K. F. Geisinger,
B. A. Bracken, J. F. Carlson, J-I. C. Hansen, N. R. Kun-
cel, S. P. Reise, & M. C. Rodriguez (Eds.), APA Hand-
book of Testing and Assessment in Psychology. Vol.
1: Test theory and testing and assessment in industrial
and organizational psychology, pp. 565–577. Wash-
ington, DC: American Psychological Association.
Lewis-Rice, M. (1989). Marketing post-secondary edu-
cational programs with implications of higher educa-
tion administration. Doctoral dissertation, School of
Urban and Public Affairs, Carnegie Mellon University.
Meehl, P. E. (1954). Clinical versus statistical predic-
tion; a theoretical analysis and review of the evidence.
Minneapolis: University of Minnesota Press.
Milstein, R. M., Wilkinson, L., Burrow, G. N., & Kessen,
W. (1981). Admission decisions and performance dur-
ing medical school. Journal of Medical Education, 56,
77–82.
Nisbett, R., Zukier, H., & Lemley, R. (1981). The di-
lution effect: Nondiagnostic information weakens the
implications of diagnostic information. Cognitive Psy-
chology, 13, 248–277.
Ruscio, J. (2000). The role of complex thought in clinical
prediction: Social accountability and the need for cog-
nition. Journal of Consulting and Clinical Psychology,
68, 145–154.
Sanbonmatsu, D., Posavac, S., Kardes, F., & Mantel, S.
(1998). Selective Hypothesis Testing, Psychonomic
Bulletin & Review, 5, 197–220.
Swann, W. B., & Gill, M. J. (1997). Confidence and accu-
racy in person perception: Do we know what we think
we know about our relationship partners? Journal of
Personality and Social Psychology, 73, 747–757.
Trout, J. D. (2002). Scientific explanation and the sense
of understanding. Philosophy of Science, 69, 212–233.
... Predictions based on clinical judgment have been shown to be poorer than those based on actuarial approaches (e.g., Canivez, 2013;Dana et al., 2013;Dawes et al., 1989;Grove et al., 2000). Despite this support for actuarial modeling, it has rarely been applied to SLD identification in research or practice. ...
Article
Full-text available
The Dual Discrepancy/Consistency (DD/C) pattern of strengths and weaknesses (PSW) method is the most prominent “other alternative research-based procedure” used to identify a specific learning disability (SLD). The DD/C method is intended to be used with multiple data sources, but no research has examined identification decision making using the DD/C PSW analysis of cognitive and academic achievement test scores with other assessment data (e.g., pre-referral intervention data, parent/teacher reports, observations, evaluation of exclusionary factors). This study used an experimental vignette methodology to investigate school psychologists’ (N = 343) identification decision making using the DD/C method with multiple data sources. Results showed that school psychologist participants made highly accurate (97.50%) identification decisions when both cognitive and achievement test score patterns and other assessment data were clearly indicative of SLD, but that overall identification accuracy was only 65.60%. Participants made less accurate identification decisions when examining test score data that did not meet DD/C PSW criteria for SLD than when examining test score data that did. Participants made less accurate decisions when other assessment data (interpreted in conjunction with test scores) did not meet DD/C PSW criteria for SLD or were ambiguous than when they did. In addition, male school psychologists were less likely to make accurate identification decisions than female school psychologists, and participants who were Nationally Certified School Psychologists were less likely to make accurate identification decisions than those who were not. Last, more experience with DD/C resulted in less accurate identification decisions, and participants who made accurate decisions reported higher levels of confidence, although the effect size was negligible. Implications for research and practice are discussed.
... Indeed, the failure to adopt more effective selection practices has been widely documented (Rynes et al., 2002). This rejection of more effective methods may stem from personal preference, practitioners' beliefs, convenience, reluctance to change, and costs related to switching to alternative methods (Dana et al., 2013;Rynes et al., 2002). In regard to fairness perceptions of different selection methods, practitioners may have unique views in this respect. ...
Article
Full-text available
This research examines the perceived fairness of two types of job interviews: robot-mediated and face-to-face interviews. The robot-mediated interview tests the concept of a fair proxy in the shape of a teleoperated social robot. In Study 1, a mini-public (n=53) revealed four factors that influence fairness perceptions of the robot-mediated interview and showed how HR professionals' perception of fair personnel selection is influenced by moral pragmatism despite clear moral awareness of discriminative biases in interviews. In Study 2, an experimental survey (n=242) conducted at an unemployment center showed that the respondents perceived the robot-mediated interview as fairer than the face-to-face interview. Overall, the studies suggest that HR professionals and jobseekers exhibit diverging fairness perceptions and that the business case for the robot-mediated interview undermines its social case (i.e., reducing discrimination). The paper concludes by addressing key implications and avenues for future research.
... Answers were recorded through note-taking. Unstructured interviews are one of the best ways for producing qualitative results which provide us an in-depth understanding of the present scenario and let us interpret experience, perception of the people involved [18]. Again the qualitative data collection took place after we observed the growth rate of the plants on the module and analyzed them on the basis of growth rate. ...
Conference Paper
Full-text available
Vertical gardening is a common concept nowadays for sustainable design and space quality enrichment. With the flow of time, these modules and concepts are becoming more efficient and functional day by day. For different contexts, weather conditions and surrounding facts the choice of plants and soil mixture differs. This study mainly aims to introduce an easy and cost effective installation module for the context of Khulna City which is not only eco-friendly, energy saving and temperature controlling but also has a high aesthetic value and timeless texture for simple domestic buildings especially in warm humid climate. In this study we have tried to develop the module and taken care of every step such as frame making, soil mixing, plantation etc. and observed them closely day by day in natural weather and sun conditions. By trying and observing different plant growth rate, soil retention level and need of care we have selected a particular plant which gives the best result for plantation on a vertical tile module. Through the steps of research we have studied various literatures, reviews, theories and modular systems. This study involves contextual materials, plants and ingredients to develop a module of vertical green tile which is comparatively cheaper than most of the quality tile modules we usually use as finish materials on outdoor, semi-outdoors. This unique module and its cost effectiveness compared to normal tiles will encourage people to develop mass production and use them as finish material in the context of Khulna city in future which will not only bring them weather protection and aesthetic value but also will enrich the space quality, thermal comfort and architectural values.
Article
The focus group interview is a strategy that researchers use across all disciplines from marketing to education and cultural studies. Facilitating a focus group interview is challenging even for experienced researchers. This paper shares the authors’ perspectives on the focus group interview with arguments from the literature along with practical applications. The authors use examples from several of their research studies to discuss a scholarly yet practical approach to plan, conduct, and analyze successful focus groups for the intended audience of doctoral students. In addition to suggestions for facilitating data collection and data analysis for the focus group interview, this paper includes strategies to select the participants, ensure participants show up, engage all the focus group participants, address powerful or authoritative personalities, enhance the synergy of the group, and reach consensus.
Article
We analyze 11 years of admission decisions to a highly selective post‐graduate visual arts program. Our unique, longitudinal data includes detailed information about the selection procedures as well as the later artistic and economic achievements for both accepted and rejected applicants (n = 8,557). Regarding the predictive value of the selection procedures on later performance, our analyses show that the largest gain made is in the first, more cursory, pre‐selection round. We find that weeding out less promising applicants is easier than identifying later top‐performers, in terms of ex‐ante expectations. The subsequent or final selection round is resource intensive and consists of several interviews with multiple jury members. Here we uncover an implicit structure of criterion variables that helps explain the ex‐post admission decisions. Our simulations show that actual selection during the final round is slightly better than alternative decision rules (including a decision by lottery) in identifying applicants' potential. Nevertheless, measured by future performance, the gain from rigorous jury assessments relative to modest and cheaper selection methods is minimal. Our conclusions are maintained under several robustness checks.
Chapter
Psychological tele-assessment is the provision of psychological assessment through the use of remote tele-health (predominantly video-based) platforms. Although some assessment methods (like clinical interviews and surveys) have a robust and long-standing history of research validating the data that emerge from tele-assessments, other methods (like cognitive and academic tests) have much less research, and still others (like play-based autism evaluations) are not feasible to accomplish through tele-assessment techniques. This paper presents the current state of the literature on conducting valid and ethical tele-assessments.
Article
We perform the first tests of individual-level preferences for “blinding” in decision making: purposefully restricting the information one sees in order to form a more objective evaluation. For example, when grading her students’ papers, a professor might choose to “blind” herself to students’ names by anonymizing them, thus evaluating the papers on content alone. We predict that curiosity will shape blinding preferences, motivating people to seek out (vs. be blind to) irrelevant, potentially biasing information about a target of evaluation. We further predict that decision frames that reduce or satisfy curiosity about potentially biasing information will encourage choices to be blind to that information. We find support for these hypotheses across seven studies (N = 4,356) and multiple replications (N = 9,570), demonstrating consequences for bias and accuracy across a variety of evaluation contexts. We discuss implications for research on mental contamination as well as the “dark side” of curiosity.
Article
Deliberative analysis enables us to weigh features, simulate futures, and arrive at good, tractable decisions. So why do we so often eschew deliberation, and instead rely on more intuitive, gut responses? We propose that intuition might be prescribed for some decisions because people's folk theory of decision-making accords a special role to authenticity, which is associated with intuitive choice. Five pre-registered experiments find evidence in favor of this claim. In Experiment 1 (N = 654), we show that participants prescribe intuition and deliberation as a basis for decisions differentially across domains, and that these prescriptions predict reported choice. In Experiment 2 (N = 555), we find that choosing intuitively vs. deliberately leads to different inferences concerning the decision-maker's commitment and authenticity—with only inferences about the decision-maker's authenticity showing variation across domains that matches that observed for the prescription of intuition in Experiment 1. In Experiment 3 (N = 631), we replicate our prior results and rule out plausible confounds. Finally, in Experiment 4 (N = 177) and Experiment 5 (N = 526), we find that an experimental manipulation of the importance of authenticity affects the prescribed role for intuition as well as the endorsement of expert human or algorithmic advice. These effects hold beyond previously recognized influences on intuitive vs. deliberative choice, such as computational costs, presumed reliability, objectivity, complexity, and expertise.
Article
In a preregistered prediction-task experiment, we investigated the effect of narrative versus quantified information on decision-maker perceptions, confidence, predictor weighting, and predictive accuracy when making performance predictions. We also investigated the effect of who quantifies information (the decision maker or someone else). As expected, we found higher perceived informativeness and use intentions for narrative than quantified information. Information presented narratively was also weighted somewhat more heavily than quantified information. Using quantitative information quantified by decision makers themselves yielded higher perceived autonomy and use intentions than quantitative information quantified by someone else. However, no differences in prediction confidence were found and self- and other-produced quantifications received identical weight. Moreover, unexpectedly, differences in weighting did not translate to differences in predictive accuracy. Practitioner points • Several authors suggested that narrative information is perceived as richer, and hence, is more influential in judgments and predictions, than quantitative information. • Additionally, quantified information was suggested to be more influential when the decision-makers quantify information themselves. • We found that narrative information was perceived as more informative, yielded higher use intentions, and was weighted somewhat more heavily than quantified information. We also found higher perceived autonomy and use intentions for self-quantified, than for other-quantified information, but no differences in assigned weight. • Notably, we found no differences in confidence or predictive accuracy. • Since we did not find the expected effects on predictive accuracy, the implications for selection practice remain unclear.
Article
The role of diagnostic and confirmation strategies in trait hypothesis testing is examined. The present studies integrate theoretical and empirical work on qualitative differences among traits with the hypothesis-testing literature. Ss tested trait hypotheses from 2 hierarchically restrictive trait dimensions: introversion-extraversion and honesty-dishonesty. In Study 1, Ss generated questions to test trait hypotheses, and diagnosticity was theoretically defined (e.g., questions associated with nonrestrictive ends of trait dimensions). In Study 2, Ss selected questions from an experimenter-provided list in which diagnosticity was empirically defined. In Study 3, Ss chose between 2 equally diagnostic questions. In each of the studies, Ss showed a primary preference for diagnostic information and a secondary preference for confirmatory information. Ss' preference for diagnostic information suggests that they prefer to ask the most informative questions. The explanation for the confirmation bias is less obvious, and possible reasons for this effect are discussed.
Article
Scientists and laypeople alike use the sense of understanding that an explanation conveys as a cue to good or correct explanation. Although the occurrence of this sense or feeling of understanding is neither necessary nor sufficient for good explanation, it does drive judgments of the plausibility and, ultimately, the acceptability, of an explanation. This paper presents evidence that the sense of understanding is in part the routine consequence of two well-documented biases in cognitive psychology: overconfidence and hindsight. In light of the prevalence of counterfeit understanding in the history of science, I argue that many forms of cognitive achievement do not involve a sense of understanding, and that only the truth or accuracy of an explanation make the sense of understanding a valid cue to genuine understanding.
Article
At the University of Texas Medical School at Houston we had a unique opportunity to examine performance through the medical curriculum and one year of postgraduate training of 50 students initially rejected for medical school. Each had been interviewed by the same Admissions Committee, which earlier had selected 150 students through the traditional process. In contrasting the initially accepted and initially rejected groups, academic and demographic variables accounted for only 28% of group difference. The 72% of group difference not accounted for by the variables examined was presumed to relate to Admissions Committee preference. In attrition and in both preclinical and clinical performance through medical school and one year of postgraduate training, there were no meaningful differences between the groups. The observations suggest that the traditional interview process probably does not enhance the ability to predict performance of medical school applicants.(JAMA 1987;257:47-51)
Article
Discretionary legal decisions have become a recent focus of theory development and policy-oriented applied research. We investigated parole release decision making in Pennsylvania from both orientations. Analyses of post-hearing questionnaires and case files from 1,035 actual parole decisions revealed that the Parole Board considers institutional behavior and predictions of future risk and rehabilitation in the decision to release on parole. Predictions seem also to be based on diagnostic judgments identifying causes of crime such as personal dispositions, drugs, alcohol, money, and environment. A one-year follow-up of 838 released parolees showed that predictions were virtually unrelated to known post-release outcomes. An actuarial prediction device was developed that is more predictive than subjective judgments. The use of decision guidelines to structure discretion is discussed, as well as the utilization of our research in guideline development by Pennsylvania.
Article
This monograph is an expansion of lectures given in the years 1947-1950 to graduate colloquia at the universities of Chicago, Iowa, and Wisconsin, and of a lecture series delivered to staff and trainees at the Veterans Administration Mental Hygiene Clinic at Ft. Snelling, Minnesota. Perhaps a general remark in clarification of my own position is in order. Students in my class in clinical psychology have often reacted to the lectures on this topic as to a protective technique, complaining that I was biased either for or against statistics (or the clinician), depending mainly on where the student himself stood! This I have, of course, found very reassuring. One clinical student suggested that I tally the pro-con ratio for the list of honorific and derogatory adjectives in Chapter 1 (page 4), and the reader will discover that this unedited sample of my verbal behavior puts my bias squarely at the midline. The style and sequence of the paper reflect my own ambivalence and real puzzlement, and I have deliberately left the document in this discursive form to retain the flavor of the mental conflict that besets most of us who do clinical work but try to be scientists. I have read and heard too many rapid-fire, once-over-lightly "resolutions" of this controversy to aim at contributing another such. The thing is just not that simple. I was therefore not surprised to discover that the same sections which one reader finds obvious and over-elaborated, another singles out as especially useful for his particular difficulties. My thesis in a nutshell: "There is no convincing reason to assume that explicitly formalized mathematical rules and the clinician's creativity are equally suited for any given kind of task, or that their comparative effectiveness is the same for different tasks. Current clinical practice should be much more critically examined with this in mind than it has been." (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
The role of diagnostic and confirmation strategies in trait hypothesis testing is examined. The present studies integrate theoretical and empirical work on qualitative differences among traits with the hypothesis-testing literature. Ss tested trait hypotheses from 2 hierarchically restrictive trait dimensions: introversion–extraversion and honesty–dishonesty. In Study 1, Ss generated questions to test trait hypotheses, and diagnosticity was theoretically defined (e.g., questions associated with nonrestrictive ends of trait dimensions). In Study 2, Ss selected questions from an experimenter-provided list in which diagnosticity was empirically defined. In Study 3, Ss chose between 2 equally diagnostic questions. In each of the studies, Ss showed a primary preference for diagnostic information and a secondary preference for confirmatory information. Ss' preference for diagnostic information suggests that they prefer to ask the most informative questions. The explanation for the confirmation bias is less obvious, and possible reasons for this effect are discussed. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Defines irrationality as adhering to beliefs that are inherently self-contradictory, not just incorrect, self-defeating, or the basis or poor decisions. Such beliefs are unfortunately common. Two examples are given: the belief that child sexual abuse can be diagnosed by observing symptoms typically resulting from such abuse, rather than symptoms that differentiate between abused and non-abused children; and the belief that a physical or personal disaster can be understood by studying it alone in-depth rather than by comparing the situation in which it occurred to similar situations where nothing bad happened. This book first demonstrates how such irrationality results from ignoring obvious comparisons. Such neglect is traced to associational and story-based thinking, while true rational judgment requires comparative thinking. Strong emotion—or even insanity—is one reason for making automatic associations without comparison, but as the author demonstrates, a lot of everyday judgment, unsupported professional claims, and even social policy is based on the same kind of irrationality. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Virtually every previous review has concluded that structuring the selection interview improves its psychometric properties. This paper reviews the research literature in order to describe and evaluate the many ways interviews can be structured. Fifteen components of structure are identified that may enhance either the content of the interview or the evaluation process in the interview. Each component is explained in terms of its various operationalizations in the literature. Then, each component is critiqued in terms of its impact on numerous forms of reliability, validity, and user reactions. Finally, recommendations for research and practice are presented. It is concluded that interviews can be easily enhanced by using some of the many possible components of structure, and the improvement of this popular selection procedure should be a high priority for future research and practice.