Do indirect measures of attitudes improve our predictions of behavior? Evaluating and explaining the predictive validity of GATA

The generalization of the results accumulated to date has shown that the implicit measures of attitudes (some even suggest defining them with a less pretentious term “indirect”) show a disappointingly weak predictive potential in relation to real behavior. Thus, the predictive validity of the Graphical Association Test of Attitude (GATA), which also claims to be an indirect method, has been questioned. To check this assumption, we analyzed the results obtained with GATA in 64 predictions provided that the predicted outcome could be verified by real action. Such forecasts cover the domains of electoral, consumer and communicative behavior. In some cases, the prediction based on the data from a representative sample was checked referring to the actual behavior of the group represented by the sample, e.g., the electorate, or the consumers of a certain category of goods, etc. In other cases, the accuracy of the forecast was checked for each respondent. This allows to avoid the effect of “mutual compensation” of erroneous forecasts with opposite valence. The test method consisted of a comparison of the prediction accuracy of pairs of “control” and “experimental” prediction models: the only difference identified was that the latter used the data from indirect measurements of GATA as an additional factor of action. In the article, all models are presented in their simplest and most transparent versions. The results of the conducted meta-analysis do not fully correspond to the general trend: the use of the GATA data significantly and continuously improves the accuracy of predicting behavior. In addition, the incremental effect on the accuracy of individual forecasts (for each respondent) turned out to be higher than that of the sample-based group forecasts.