Epstein proposed that perhaps personality traits were not able to predict the occurrence of specific behaviors, but that personality traits may be able to predict a person’s behavior ON THE AVERAGE. Epstein tested the hypothesis that stability in behavior (supportive of a trait position) could be demonstrated over a wide range of variables so long as the behavior in question was averaged over a sufficient number of occurrences. Using a sample of 19 college seniors and 15 1st year graduate students, Epstein had them record DAILY a variety of measures over a semester. Measures included self report of mood (e.g., happy-sad, kind-angry, calm-tense, worthy-unworthy), peer reports, objective measures of discrete items of behavior (e.g., social telephone calls made and received, letters written and received, # of headaches, etc) and various psychophysiological measures and personality measures. Correlations were then computed between behaviors recorded on odd and even days in the sample. Then data were aggregated (averaged) in the following fashion.

r day-1 with day-2

r days 1+3/2 with days 2+4/2

r days 1+3+5/3 with days 2+4+6/3, and so on up to a 12 day sample.

As more measures were taken and aggregated, reliability coefficients increased dramatically as more measures were taken and aggregated (averaged), from r = .3 to .4 over 2 measures to r = .8 to .9 over the maximum of averaged measures. When personality inventories were correlated with behaviors and data from both sources were averaged over many occasions, validity coefficients uniformly exceeded .3 and ranged from r = .4 to .6.

It is important to note that there are many ways to aggregate data:

1. Over persons. This is what any group study does, it averages data for a group of people.

2. Over trials/occasions. This involves the repeated measurement of the same response, in the same situation. Only time changes. For example, if we are aggregating punctuality to class, we would measure how early/late you are for each 673 class. This is also known as TEMPORAL STABILITY.

3. Over stimuli/situations. This involves the repeated measurement of the same response, but in

different situations. For example, on the average, how early or late are you for 673 class, a different class, doctor’s appointments, meals, work, or racquetball.

4. Over different measures. This would involve averaging over different measures of a trait. For example, a researcher could average over different measures of conscientiousness (e.g., punctuality to class, neatness, probability of someone keeping promises) that could be recorded in different situations (for example, at school, at home, at work, at recreation, etc.)

Epstien’s point is that aggregating in these ways provides support for the importance of personality traits.

In understanding Mischel & Peake’s critique of the “Reliability Solution” it is important to understand the following terms and how they are defined:

1. TEMPORAL STABILITY, refers to the correlation of the same behavior with itself in the same situation. Only time changes. This would be the correlation of how early/late you arrive to 673 class on all Mondays correlated with punctuality to class on all Wed.

2. CROSS-SITUATIONAL, UNI-MODAL CONSISTENCY, refers to the correlation of the same behavior with itself (same response forms) across different situations. How early or late are you to 673 class correlated with how early or late you are to work. A high correlation here would be evidence for the trait of conscientiousness having an impact in two different situations.

3. CROSS-SITUATIONAL, CROSS-MODAL CONSISTENCY, refers to correlations among different behavioral measures (which have psychological similarity) over multiple situations and time. For example, consider the possible correlation of doing readings for Research Methods within 24 hours after they are assigned with showing up how early or late to 673. The two different measures are psychological similar (i.e., measures of punctuality/conscientiousness), but they are obtained in different situations.

Mischel and Peake (M&P) argue that the true test for the importance of personality variables are obtained from #2 and #3 above, the cross-situational evaluations. Mischel & Peake report data from the Carleton Behavior Study, which was an attempted replication of BEM & Allen using aggregated data. M&P agree that aggregating data does produce greater reliability, but even studies from the 40s through the 60s that used reliable data were still not able to find evidence for the two forms of CROSS-SITUATIONAL CONSISTENCY. Aggregating the Carleton data for conscientiousness did show impressive increases in TEMPORAL STABILITY (rs increase from .29 for single behaviors to .65 for aggregated data). But evidence for CROSS-SITUATIONAL, CROSS-MODAL CONSISTENCY using aggregated data was only mean r = .13 (.20 when corrected for attenuation), an increase from mean r = .08 when using correlations of single behaviors. Aggregated data from CROSS-SITUATIONAL, UNI-MODAL correlations represented a mean r = .28.

Although average cross-situational consistency was low, there were still a few striking patterns of consistency (and inconsistency) in the Carleton data. For example, consider the following correlations with attendance in class (using all subjects, not just low variability ones):

r

.67--appointment attendance

.53--assignment punctuality

.58--completion of class readings

.31--time studying, and these low correlations

.14--class note thoroughness

-.03--punctuality to lectures

-.04--assignment neatness

In your teacher’s opinion, M&P ignore some of Epstein’s more interesting data for reports of physical symptoms, which potentially may reflect a personality dimension of somatization/hypochondriasis. We will go into more details in class.

REGARDING BEM & ALLEN’S (B&A) “IDEOGRAPHIC SOLUTION”, M&P first state that B&A misuse the term “ideographic”. M&P are likely correct in this regard, but this is not the major point. M&P agree that there are impressive levels of agreement among raters (Self-report and the reports of mom, dad, and peers) in the B&A data for friendliness and conscientiousness. M&P do not believe that this is good evidence for the importance of traits, however. M&P note that there is only one very strong correlation in the B&A data supporting CROSS-SITUATIONAL CONSISTENCY, the correlation of .73 between SPONTANEOUS FRIENDLINESS and GROUP DISCUSSION FRIENDLINESS for the low variability subjects. They again point out that the average CROSS-SITUATIONAL CORRELATION in the Carleton data was only r = .13, poor evidence for the importance of traits.

M&P note a paradox to be explained: Subjects classified as consistent (low variability from situation) tend to show high levels of inter-rater agreement when rated on these personality dimensions by relevant others (intuitively implying some consistency), yet these subjects do not show appreciably high levels of CROSS-SITUATIONAL CONSISTENCY IN THEIR BEHAVIORS. M&P try to answer the paradox by suggesting that raters judge the degree of consistency from observing the TEMPORAL STABILITY of a few relevant prototypical behaviors, rather than from observations of cross-situational consistency.

How far have we come in answering the question of how important is the consideration of personality variables in trying to explain why a person does what she does. We obviously can NOT predict the occurrence of any one particular behavior for a person in any one situation with much accuracy. With a group of similar people (e.g., all consistent--low variability-- in their expressions of any one trait), we might have some accuracy in predicting the average performance in the average situation. For example, for a group of low variability, conscientious people; we might be able to predict that most of them would show up for most appointments they would schedule.