Advanced Statistical Modeling of Ecological Constraints in Information Sampling and Utilization

Abstract

Information sampling and utilization are ubiquitous in daily life. Accordingly, both processes are affected by a variety of environmental factors. This dissertation is primarily concerned with ecological constraints that are implemented by the social context. In particular, people often consider the opinions and beliefs of others in their judgments and decisions. Research on advice taking and related cognitive phenomena such as anchoring, hindsight, or attitude change traditionally relies on ratio-of-differences-type formulas to determine informational influences. In this dissertation, two alternative modeling frameworks are presented for specifying how strongly peoples’ judgments are influenced by external information. In contrast to the traditional approach, the proposed methods are consistent with the dependency of endogenous judgments (i.e., potentially updated beliefs) on exogenous sources of information (e.g., advice, base rates, anchors). Corresponding statistical modeling has the advantage of avoiding critical measurement problems of the traditional approach and is shown to enable new substantive research. A Bayesian account provides the opportunity to test for adaptive strategy selection in sequential advice seeking by explicitly distinguishing Thurstonian and Brunswikian sampling. Moreover, mixed-effects regression of final judgment on any exogenous sources of information resolves further paradigmatic peculiarities of the classic experimental procedure. For instance, the traditional modeling approach requires independent initial judgments as well as observable intermediate judgments, or presupposes equal weighting of sequentially sampled advice, respectively. Empirical investigations of advice expectation and sequential advice seeking highlight two particularly relevant and novel ecological constraints of social information acquisition. First, traditional modeling reveals a positive effect of advice expectation on weighting for a trial-by-trial contrast of low versus high expectation to receive advice. The proposed regression-based approach validates this finding by means of processconsistent statistical modeling. Second, final judgment correspondence is taken as evidence for Bayesian advice taking in sampling extensions of the classic experimental paradigm. Indeed, empirical mixed-effects regression weights of sequentially sampled advice are moderately to strongly correlated with Bayesian weights constituting the normative benchmark. Moreover, both more advanced modeling approaches provide first evidence for nonlinear serial weighting of sequentially sampled advice. In summary, the process-consistent statistical modeling proposed in this dissertation facilitates and extends substantive research on important ecological constraints of (social) information acquisition, such as the expectation of external influences and the sequential sampling of information.