Modeling inconsistencies in people's preferential choices with sequential sampling models

Abstract

People are faced with hundreds of decisions every day, of which many involve choices between multi-attribute options. To study how people evaluate these options, different approaches have been suggested. One promising approach is to apply sequential sampling models (SSMs), which make predictions about the underlying cognitive processes involved in decision making. Multialternative Decision Field Theory (MDFT), a powerful SSM, assumes that people accumulate evidence for each option over time by continuously comparing the options’ attribute values. A decision is made once the evidence of one option first passes a predefined threshold. Although MDFT has been widely used to predict people’s choice behavior, an empirical model comparison is missing. This is mainly due to certain model restrictions. One such restriction is that the function to measure the distance between the options, is not fully specified. We addressed this challenge in the first manuscript and suggested a generalized distance function for multi-attribute choice options. In the second manuscript, we provided a testable version of MDFT and empirically compared the model to two of the most frequently used random utility models in the choice literature, the logit and probit models. We illustrated that when violations of consistency principles are likely to occur, MDFT described and predicted people’s choice most accurately. In the third manuscript, we showed in an empirical experiment that previous choices can influence people’s subsequent choice, even when independent sequential choices are stressed, suggesting another violation of consistency principles. Current SSMs, such as MDFT, cannot account for the observed compensating choice behavior, as shown in the manuscript. To overcome this limitation, we suggested a dependent sequential sampling model (DSSM), which assumes that the initial preferences for the choice options are updated after every choice. Using the preferential choice options of the experiment, we illustrated how DSSM can model the observed compensating choice behavior, as well as consistent choice behavior.