David A. Kenny
April 5, 2021

Chapter 6 of Interpersonal Perception: The Foundation of Social Relationships. See also pages 63-71 in Chapter 3.

Target accuracy is measured by the degree of association between how a perceiver sees a target with how the target really is (i.e., the truth). In Laing notation it is symbolized by A(B) = B. Very often it is called just accuracy, but it is helpful to call it target accuracy to distinguish it from other forms of accuracy, i.e., meta-accuracy and self-accuracy.

A major obstacle in target accuracy research is obtaining a measure of the target's actual standing on the trait: If John says that Mark is friendly how do we know whether Mark is friendly or not? Some researchers believed that it is so difficult to measure the target's actual standing on the trait that accuracy research should be abandoned. Fortunately, as argued by William Swann (1984) and others, this has not happened.

The major ways of measuring the truth are self-reports of the target, reports by knowledgeable informants (e. g., spouse or close friends), observations of the target's behavior, and the average rating of others. It is generally agreed that behavioral ratings are best, but they are difficult to gather. What is generally recommended is multimethod approach to measuring the truth. For instance, Borkenau and colleagues in 2004 used observer ratings, self-ratings, and ratings by close friends.

Regardless of how the truth is measured, it can be argued that the truth itself is a rating. Even behavior measures are the ratings of an observer. Kruglanski in 1989 has argued that accuracy is special type of consensus; that is, accuracy represents the correspondence between two sets of ratings. One set of ratings is seen as the "standard" to which the other is compared.

If we think of accuracy as a correlation, we can compute that correlation in two very different ways. First, we can compute the correlation across targets, which is called a nomothetic correlation. For instance, we might ask if perceivers think a target is friendly, is that target friendly? Second, we might look within a perceiver's ratings of a target on multiple traits and ask if those judgments line up with the truth, which is called a ideographic correlation. Note the unit for this correlation is trait and accuracy can be measured for a single perceiver. This page discusses the nomothetic approach and a later page discusses individual differences in accuracy.

Beginning with a paper in 1955 by Lee Cronbach, accuracy researchers had to face the issue that ratings of the target should be decomposed into components and that accuracy is the correspondence between those components. Cronbach proposed four types of accuracy of a perceiver who rates several targets on several traits:

ELEVATION: the correspondence between the average rating of the perceiver and the average score on the criterion measure (the truth).
DIFFERENTIAL ELEVATION: the correspondence between the differences of the average ratings across traits of two targets with the corresponding difference in criterion.
STEREOTYPE ACCURACY: the correspondence between the differences of the average ratings across targets of two traits with the corresponding difference in criterion.
DIFFERENTIAL ACCURACY: the correspondence between the rating of the perceiver and the criterion with means of target and trait removed.
If we assume that all measures are scaled where more means something good, then in essence, elevation assesses the degree to which the perceiver is overly positive or negative, differential elevation assesses whether the perceiver knows which targets are good and bad, stereotype accuracy assesses whether the perceiver know people in general vary on the traits, and differential accuracy assesses whether the perceiver knows how the targets differ from each other on the traits. Most analysts treat differential elevation and differential accuracy as "real" accuracy and the other two as artifactual. However, there are situations where stereotype accuracy, which has come to be called normative accuracy, and elevation are meaningful.

The Social Relations Model decomposition of variance is different from that of Cronbach. It looks at the ratings of perceivers and targets for each trait whereas the Cronbach decomposition is of the ratings of targets and traits for each perceiver. Using the Social Relations Model to study accuracy, the focus is not on who is more accurate (i.e., ideographic accuracy), but rather what is the level of accuracy in person perception and when are people accurate (i.e., nomothetic accuracy).

The Cronbach critique of accuracy research had a devastating effect on the area. In essence, it killed accuracy research for a generation. Researchers felt either that accuracy was too complicated to measure (something that may have been true at the time but no longer is), that it was impossible to measure (which is not true), or that accuracy did not exist (which may or may not be true, but probably is not). The essence of the Cronbach critique is that accuracy is not global but it is the correspondence between components.

Within the Social Relations Model, there are four types of target accuracy: perceiver, generalized, dyadic, and constant target accuracy.

PERCEIVER ACCURACY: the correlation between how a person tends to see others in general with how others generally behave with the person.
GENERALIZED ACCURACY : the correlation between how a person is generally seen by others and how that person generally behaves.
DYADIC ACCURACY: the correlation between how the person is uniquely seen by someone with how that person uniquely behaves with that someone.
ELEVATION ACCURACY: the match between average perceptions and average behavior.
For perceiver accuracy, if Alice sees others as friendly, is Alice a friendly person? For generalized accuracy, if A is seen by others as friendly, is A a friendly person? For dyadic accuracy, if person A sees B as especially friendly, is B especially friendly when interacting with A? For elevation accuracy, if people think others are friendly, are they in fact friendly? Note to measure elevation accuracy, the judgment and the truth need to be on the same scale, something this is atypical in accuracy research.

William Swann (1984) argued that perceivers should be especially accurate at predicting how others would behave when interacting with the perceiver, something Swann called circumscribed accuracy. In terms of the SRM, circumscribed accuracy implies that there should be dyadic accuracy.

There have been three SRM studies measuring target accuracy using a behavioral measure of the truth. The first was Levesque and Kenny (1993) who looked at extraversion at zero acquaintance. They found generalized accuracy of .56, but no evidence of dyadic or perceiver accuracy. These findings are consistent with work by Ambady and Rosenthal (1993) that judgments based on "thin slices" of behaviors has surprising validity.

The second study, Kenny, Kieffer, Smith, Ceplenski, and Kulo (1996) examined competiveness and extroversion in groups of fraternity members who were playing a competitive game. Again, generalized accuracy was found for both competitiveness (.43) and extroversion (.63), but not dyadic or perceiver accuracy.

The third study by Kenny, West, Cillessen, Coie, Dodge, Hubbard, and Schwartz (2007) examined 10 year-old boys and their perceptions of aggressiveness. Aggressive behavior was measured from behavioral observations in a controlled setting for an hour of free play. Results showed generalized accuracy of .56, and no evidence of dyadic or perceiver accuracy.

The failure of all three of the studies to find dyadic accuracy surprised us. However, as described on pages 226-227 in Chapter 8 of Interpersonal Perception: The Foundation of Social Relationships, there is evidence, albeit indirect, of for dyadic accuracy.

Using the PERSON Model, target accuracy measured for an individual asymptotes at .63, regardless of the value of the kernel-of-truth parameters. Where it starts at zero acquaintance, depends on the kernel of truth. Given the values of S and R, the estimate of the kernel-of-truth parameter is .36. However, in some cases the kernel of truth is better called a grain of falsehood; that is, the stereotype is false, leading to a negative correlation between the stereotype and the truth. Examine Figure 6.1 on page 166 of Interpersonal Perception: The Foundation of Social Relationships, to see the different relationships between acquaintance and accuracy for different values of m, the kernel of truth parameter. With a strong kernel of truth the relationship is relatively flat, but when there is a grain of falsehood, accuracy is initially negative. Note that regardless of the kernel of truth after about 15 or so act, there is little or no effect of the kernel of truth. Note too at that point, increasing acquaintance has very little effect on accuracy.

The PERSON model's estimate of accuracy at long-term acquaintance is .63, a value considerably higher than the observed value. The difference can be explained by unreliability of behavioral ratings, cross-context differences in the P component, these are insufficient to explain the gap, and multiple determinants of behavior besides the trait of interest.

What then is the correlation between a perceiver's judgment and the truth. If we take .60 as the estimate of generalized accuracy and .40 as the estimate of consensus that translate to a correlation of .32. That roughly translate to being accurate 66 percent of the time, where 50 is chance. That value is very close to the estimate of 68 percent found by Dunning and colleagues (1990), when they asked people to make predictions about someone's behavior.

Borkenau, P., Mauer, N. Riemann, R. Spinath, F. M. & Angleitner, A.  (2004). Thin Slices of behavior as cues of personality and intelligence.  Journal of Personality and Social Psychology, 86, 599-614.

Cronbach, L. J. (1955). Processes affecting scores on "understanding others" and "assumed similarity." Psychological Bulletin, 52, 177-193

Dunning, D., Griffin, D. W., Milojkovic, J. D., & Ross, L. (1990). The overconfidence effect in social prediction. Journal of Personality and Social Psychology58, 568-581.

Funder, D. C. (1995). On the accuracy of personality judgment: A realistic approach. Psychological Review, 102, 652-670.

Kenny, D. A., Kieffer, S., Smith, J., Ceplenski, P., & Kulo, J. (1996). Circumscribed accuracy among well-acquainted individuals. Journal of Experimental Social Psychology, 32, 1-12.

Kenny, D. A., West, T. V., Cillessen, A. H. N., Coie, J. D., Dodge, K. A., Hubbard, J. A., & Schwartz, D. (2007). Accuracy in judgments of aggressiveness. Personality and Social Psychology Bulletin, 33, 1225-1236.

Kruglanski, A. W. (1989). The psychology of being "right": The problem of accuracy in social perception and cognition. Psychological Bulletin, 106, 395-409.

Levesque, M. J., & Kenny, D. A. (1993). Accuracy of behavioral predictions at zero acquaintance: A social relations analysis. Journal of Personality and Social Psychology, 65, 1178-1187.

Swann, W. B., Jr., (1984). Quest for accuracy in person perception: A matter of pragmatics. Psychological Review, 91, 457-477.

Go to the next Interpersonal Perception page.

Book logo
Go to the Interpersonal Perception menu.