Multi-stage Decision Model

Project Title: Multi-stage decision model: Further developments and empirical tests (DFG: DI 506/15-1)

Dynamic-stochastic models, based on the notion of sequential sampling, are commonly used to predict choice behavior and response times in many areas of psychology, from basic perceptual tasks to multidimensional preference and decision making. Typically, these models can only account for binary decisions and unidimensional stimuli. However, many applications require an extension of the sequential sampling mechanism to multi-attribute choice options with multiple alternatives. Formal derivation of model predictions for more complex situations is often difficult or impossible, severely limiting the psychological interpretability and experimental tests of these models. An exception is Multiattribute Decision Field Theory (MDFT), developed in Diederich (1997), and its extension, the multi-stage model (Diederich & Oswald, 2014), which is based on a matrix approximation of continuous processes.

The goal of this project is to advance the sequential sampling approach within the class of dynamic-stochastic decision models. To this end, we (1) extend the multi-stage model to include a variety of mechanisms for attribute handling (fixed and random processing order, switching times, and process duration) that constitute different hypotheses about the distribution of attention in stimulus processing. Then follows (2) the development of a mechanisms that also accounts for non-responses (e.g. missed deadline). For that, different functions of decision criteria are considered (i.e. absorbing boundaries of the stochastic process). We will strive for analytical solutions for all models.

The empirical part of this project will experimentally probe the multi-stage model. We will explore different aspects of attribute processing in discrimination and choice paradigms. The first series of experiments will test hypotheses about the effect of payoffs, considered as stimulus attribute, on response frequencies in two different perceptual discrimination tasks. The effect of temporal distribution of information on task performance will be studied in a second series of experiments. Finally, the effect of the number of attributes on preference under risk will be investigated in a third experimental series.