Using the implicit association test to measure age differences in implicit social cognitions.

Summary of Study 1 Table 3 provides an overview of the hypotheses and results forStudy 1. As this table reveals, the IAT results of our secondanalysis provided support for our basic hypotheses regarding thedirection of implicit associations (i.e., more positive implicit atti-tudes toward the young than the old, a young implicit age identity,and positive implicit self-esteem), as had our first analysis but,unlike the first analysis, also supported the majority of our expec-tations regarding similarities and differences in the strength ofimplicit associations across age groups. As expected age differ-ences were found only on the Age Identity IAT, showing that theoldest participants identified less with youth than did the youngestparticipants (although the young-old group did not differ in ageidentity from the other two age groups). Also as predicted, AgeAttitude and Self-Esteem IATs indicated equivalent implicit atti-tudes and self-esteem across the three age groups. With thez-score 4 As a further check on our choice of transformation, we evaluatedproportion and task-levelz scores in comparison to the trial-levelz scores(Faust et al., 1999). The proportion scores were calculated by dividingparticipants’ means for each of the six IAT blocks by their grand meanacross blocks. The task-levelz-score transformation converted participants’means for each of the six IAT blocks to z scores on the basis of their grandmean and standard deviation across blocks. Brinley plots (Brinley, 1965)showed that in comparison to trial-levelz scores (slopes in Graph C ofFigure 2), proportions undercorrected for age-related slowing (slopesof 1.23 for both young-old and old-old) and the task levelz scoresovercorrected for age-related slowing (slopes of 0.87 for young-oldand 0.85 for old-old). Note that each of these transformations was appliedto data from young and older participants alike. Readers may speculate thattransforming only latencies from older adults by using the Brinley plotslope and intercept for raw data would be a simpler strategy. However, thisstrategy would not take into account the individual differences in RT thatunderlie group differences (Faust et al., 1999). Further, Ratcliff, Spieler,and McKoon (2000) caution that a Brinley plot slope may signal agedifferences in RT but it should not be interpreted as a measure of age-related slowing.