Our general research approach to study decision making is to combine behavioral experiments with simulations and behavioral modeling and complement these with questionnaire data, and, if applicable, with neuroscientific measures (e.g. functional magnetic resonance imaging data).
Compared to empirical data experiments offer the possibility to control important aspects of the choice situation. Keeping all important factors constant (or controlling them via complementary measures) and varying only a few target factors allows the investigator to establish a clear causal relationship between the observed effect and the varied target factors. This approach has the possibility to detect important effects about which one could only speculate based on empirical data.
In many cases, however, the detection of a causal relationship offers only limited evidence on the question how specific factors influence behavior. Behavioral modeling of choice behavior offers the possibility to distinguish different possible explanations regarding this question and compare competing theories and models.
Ideally, the process of conducting experiments and modeling the observed behavior is complemented by simulations. Modeling data from pilot experiments allows the investigator to determine a reasonable range for the parameters of the competing models. Simulating the behavior of decision makers following specific choice models over a wide range of possible choice situations allows the investigator to pick specific situations for the main experiment that best discriminate between the competing models. Including simulations in the research process therefore significantly increases the ability of experimental data to discriminate between competing models. This ability can be even further strengthened by means of using adaptive experiments, i.e. experiments in which new choice situations are automatically chosen to improve the discriminability between competing models.
One important criticism of experiments in general is the usually small sample size (around 40-60 participants), raising doubts on the representativeness of the data. This is especially problematic in experiments involving older adults, as older samples usually face a selection bias.
To ensure the representativeness of samples we complement the experiments with questionnaires and psychological tests. Questions from the Swiss Household Panel (SHP) or the German Socio-Economic Panel (SOEP), for example, offer the opportunity to compare the experimental sample with the large representative panel-sample in key questions related to the research question (e.g., willingness to take risks). Similarly offer psychological tests (e.g., w.r.t. working memory capacity or processing speed) the opportunity to judge the representativeness of the specific samples in important psychological traits and abilities.
Neuroscientific measures like functional magnetic resonance imaging (fMRI) can serve two purposes. If competing models explain the data (nearly) equally well, neuroscientific data can work as a “tie breaker”, providing evidence in favor or against one of the tested models. In general, neuroscientific data can be used to judge the biological plausibility of models, that is, answering the question whether the brain is able to perform the assumed decision process. Neuroscientific data can further guide the generation of new hypotheses. Knowledge about the function of brain areas implicated in specific decisions can be used to formulate new hypotheses and increase the efficiency of the research process.