This research program investigated the reasonable possibility that differential information processing strategies can be manifestations of racist prejudgments. Our research design applies a technique often used in social cognition studies. This technique captures evidence of a rather habit-driven (though not instantaneous) decision made by subjects rapidly presented with information about (typically fictional) characters. These target characters are associated (in the context of the experiment) with some form of generalized expectancy (i.e., they are typically presented as a likable or unlikable person). This is accomplished either by creating the expectancy artificially, or by using targets that are members of a conspicuously or notably-stereotyped group. The rather non-conscious decision involved is one of either bolstering one's pre-conceived notions or engaging in inconsistency resolution (e.g., either marshaling evidence to bolster your prejudicial expectation or pondering more earnestly those pieces of information which are inconsistent with your expectancy or well-known prejudicial stereotypes).

Typically, the likelihood of pursuing one strategy or the other is manipulated in experimental settings by first providing an artificial expectation, then altering the structure of the person-memory task or adjusting the rate of information flow to the subjects. We hoped to reveal how a non-artificial pre-existing race-based prejudicial expectancy (of a largely non-evaluative as in non-likable/dislikable nature) might effect the pursuit of one strategy or the other. By and large, tests of our five hypotheses provide only mixed support for use of a person-memory associative network model in this context. The first and second hypotheses have some visual support (i.e., recall proportions across sentence types start out roughly equal for low prejudicial expectation - PE -- subjects then branch out; high PE subjects seem to treat sentence types differently from the start); however, these differences are not amenable to clearly interpretable statistical tests.

Analysis of the third and fourth hypotheses was confounded because the candidate contaminating covariate failed to have consistent effects. This, coupled with the floor effect of the PE scale, the unexplained (and substantial) variability in recall behavior, and some other control issues (detailed below), made the use of the continuous DVs less than fruitful. The floor effect of the PE scale was especially problematic - with many subjects compressed at this floor, relations would be difficult to see even if present. In an attempt to detect weak effects of prejudice, we aggregated subjects by PE (as in high and low prejudice). Aggregation probably made the floor effect-driven range restriction less problematic (the subjects lumped together on PE's floor are probably less-afflicted with well-practiced prejudicial expectations than the high half of PE scorers). This exercise generated weak support for the third hypothesis: the time interval data feebly indicates that high PE subjects manifest a negative impression-centered person-memory schema in their storage of sentences about a Black target - and, unlike the low PE subjects, they apparently do this starting with the earliest blocks of sentences.

The median split approach failed to generate support for the fourth hypothesis - where we expected to see bolstering replace inconsistency resolution (in the slow condition) since subjects were afforded the time. There was weak evidence, however, that more inconsistency resolution was occurring in the fast condition (as the proposed model had predicted). This evidence was in the form of greater recall time interval differences seen when comparing high PE subjects and their schema-speeded versus non-speeded intervals. The bottom line for the first four hypotheses is still this: we failed to create a condition where prejudice would paradoxically favor recall of laudable or admirable inconsistencies associated with a fictitious Black target.

The fifth hypothesis was just intended to verify that racial prejudice does not predict recall behavior when the target is White and so are the subjects. So using a White target, we performed the same sort of tests seen above. Fortunately, relations with PE ranged from weak to very weak - and, of course, were non-significant. In sum, these outcomes suggest that Hastie-Srull associative network (H-SAN) processing effects may not reliably or consistently manifest themselves in the prejudiced rater/performance appraisal arena -- at least not in designs similar to those used previously to illustrate H-SAN effects. There were some clear exceptions, however, in our data. Taken together, our results suggest that H-SAN mechanisms may apply when appraising performance potential, but have a difficult time manifesting themselves in substantial ways.