Chapter 6: Correlating Personality & Aggregated Behavior Scales
This exercise illustrates correlation and content aggregation. It uses the PERS dataset, consisting of 90 cases and 968 variables. The variables represent measures of traits and relevant behaviors for the dimensions of extraversion (outgoingness) and conscientiousness, reported each week for three weeks by a group of undergraduate psychology students.
How well can a score on a personality test predict behavior, if several different behaviors are added together into a single, aggregated "scale."
This exercise will examine the prediction of behavior from personality. This exercise continues the illustration from Chapter 5, which predicted various single behaviors which had been aggregated over time. This exercise uses the same personality trait measures and behaviors, but aggregates the various behaviors into a single "scale" score, to increase the prediction of behavior by personality traits. This is desirable because personality trait measures are more broad and general than specific behaviors. See Epstein (1984) and Buss and Craik (1984) for further explanation. Aggregating a range of similar, relevant behaviors into a single index or "behavioral scale" should make the behavioral measure comparable in level of generality to the trait measure. Behaviors are unreliable when measured at one point in time, so this exercise assumes that the behavioral measures involved have first been averaged over several occasions (temporal aggregation); see Chapter 5 for a demonstration of using personality traits to predict behaviors which have been measured only once, versus those that have been measured repeatedly and averaged over time.
Generality of personality traits and specificity of individual behaviors.
How well can personality traits predict behavior? The empirical counterpart to this question is "How strongly do personality trait measures correlate with behavioral measures?" In addition to being affected by the reliability over time of single behavioral measures (see Chapter 5), personality traits are general tendencies or dispositions to display a variety of behaviors or styles in behavior, and are not necessarily prescriptions to enact any particular, specific behavior. Likewise, the inventories used to measure personality traits ask general questions about likes and dislikes, "usual" patterns of behavior, and so forth. A behavioral measure may ask "How many nights last week did you meet with friends to socialize?" or "How many different people did you socialize with today?" A personality test item may ask "Do you like to socialize with friends?" Thus, it may be that any one specific behavior, even those that are very relevant to the personality trait in question, is just too narrow to correlate very strongly with the general trait measure. In addition to measuring behaviors repeatedly over time and averaging them to increase reliability, we may also need to average a set of relevant behaviors, that is to aggregate across "content," or specific behaviors, in addition to aggregating across time. This means of course that we are no longer predicting one identifiable behavior, but rather we are predicting a set of behaviors, with our personality trait measure.
If you completed Chapter 5, you have already seen the personality inventories for outgoingness (JPISPT and CPISY) predicted outgoingness behaviors better after they had been averaged over the three occasions of measurement, compared to one single week’s measurement. This exercise will use the same 10 weekly behaviors for outgoingness, aggregated across time (wbo1 to wbo10), and the JPI and CPI outgoingness scales.
Check your results from Chapter 5. You should have found that for predicting single-occasion measures (Week#1), the correlations were as follows:
Scale lowest highest median
JPI .082 .351 .228
CPI .040 .335 .266
And for the 3-week average measures of behavior, the correlations were:
Scale lowest highest median
JPI .120 .510 .250
CPI .020 .382 .270
Aggregating the behaviors over time improved their correlation with personality traits to a modest extent. Applying the reasoning explained above about the specificity of individual behaviors, we should expect a stronger correlation between the trait measures and an average of the ten behaviors. Start by creating an aggregated variable from the 10 weekly behaviors used above. Use Transform>Compute, just as you did in Exercise 2 to create an average of the same behavior over time. The difference here is that we are using different behaviors to create our aggregate (which have already been averaged over time). For the Target variable, type a name like wbo10tot (to signify that it is a total across 10 weekly behaviors). For the Numeric Expression, type: wbo1+wbo2+wbo3+wbo4+wbo5+wbo6+wbo7+wbo8+wbo9+wbo10
(you can leave this as a sum, or get an average by enclosing the entire expression in parentheses and dividing by 10—it won’t affect the correlations with anything else; you can also use the Compute Functions, such as "Sum," if you know how to do this)
Now, correlate this new aggregated variable, wbo10tot, with JPISPT and CPISY. You should get r=.409 with JPISPT and r=.410 with CPISY. Compared to the lowest or the median correlations (see above) with the individual behaviors, this represents a considerable improvement. It is comparable to the highest correlations obtained when individual, specific behaviors were used, and in fact exceeds the highest correlation for the CPI personality inventory.
Buss, D. M. & Craik, K. H. (1984). Acts, dispositions, and personality. Progress in Experimental Personality Research, 13, 241-301.
Epstein, S. (1984). The stability of behavior across time and situations. In R. A. Zucker, J. Aronoff, & A. I. Rabin (Eds.), Personality and the prediction of behavior. Orlando, FL: Academic Press.