New insights into cognitive processes in pseudocontingency inference by means of experimental methods and statistical modeling

When estimating the contingency between two variables, individuals often show biases in the association they infer: Sometimes they infer an association where there is none, other times they infer an association that is opposite to the one that is actually observed. These biases have been shown to occur when individuals attempt to infer a contingency based on skewed samples of the variables: frequent categories are assumed to be associated with each other as well as infrequent categories; a phenomenon called pseudocontingency (e.g., Fiedler et al., 2009). The present thesis aimed to deepen the understanding of pseudocontingencies through the combination of experimental methods and statistical modeling. Empirical research demonstrates that pseudocontingencies are relied on for probability judgments and choices (e.g., Meiser et al., 2018). They are also reflected in reconstructive guessing processes, when memory for individual observations fails (e.g., Klauer & Meiser, 2000). In the present thesis, I corroborate this result by analyzing guessing processes based on pseudocontingencies using hierarchical multinomial processing tree models and by validating the model parameters’ substantive interpretation (see Manuscript I, Bott et al., in press). In this context, analyses additionally revealed that interindividual differences in cognitive performance as measured by the INSBAT test battery (Arendasy et al., 2009) do not seem to predict interindividual differences in relying on pseudocontingencies (Bott et al., in press). While early proposals assume differential processing of frequent versus infrequent events (e.g., Hamilton & Gifford, 1976), more recent research suggests that pseudocontingencies are the result of utilizing the marginal frequencies of variables, instead of their joint frequencies, when inferring a contingency (e.g., Fiedler et al., 2009). In line with this notion, in this thesis, I discuss a computational model, the Bayesian Marginal Model (see Manuscript II, Bott et al., 2020; Klauer, 2015), as a normative reconstruction of the pseudocontingency heuristic. The model succeeds in capturing effects found in the literature on pseudocontingency inference by assuming that beliefs about joint frequencies and thus contingencies are updated by observed marginal frequencies. Most research as well as the Bayesian Marginal Model put individuals in the role of a passive observer of predetermined information. Thus the present thesis furthermore extends empirical research by investigating the role of self-determined information sampling in pseudocontingency inference (see Manuscript III, Bott & Meiser, 2020). The results indicate that pseudocontingencies may result in wrong judgments and sub-optimal choices. However, the probability of misguidance is largely reduced when individuals actively search for information themselves. Taken together, the experiments and statistical analyses discussed in the present thesis corroborate pseudocontingencies as inferences based on marginal frequencies. Moreover, I provide original evidence that pseudocontingency effects can be captured by a normative model following the norms of Bayesian belief updating, thereby rendering pseudocontingencies as not "illogical" as described by some of its proponents. The results additionally highlight that pseudocontingencies are not necessarily wrong contingency inferences, especially when individuals actively search for information