Research

Self ratings in questionnaires with ordinal response formats are widely used to assess individual differences, like in research on personality, social cognition, attitudes or in the clinical domain. It is well known, however, that observed rating responses are affected by multiple response processes, including general response styles, social desirability, and interactions between item wording and person characteristics, over and above the to-be-measured traits. To separate the intended traits from the potentially biasing effects of further response processes, latent variable models can be chosen from item response theory or structural equation modeling. In our research, we develop such psychometric models further, examine the statistical properties and psychological validity of the model components and parameters, and we apply the models to data from different areas of psychological research.

In the measurement of person attributes by means of questionnaires or performance tests, it is commonly assumed that the structural features of items and the parameters of psychometric models are constant across persons. This assumption includes, for example, that the items are of equal difficulty across all testees and that the relative weighting of response processes is identical for all persons. If the assumption is violated, the validity of the measurement is at risk and the fairness of person comparisons is jeopardized. Our research extends psychometric models, such that heterogeneity of item characteristics and response processes across persons can be identified and accounted for in the data analysis. For this purpose, we use mixture-distribution models and model-based partitioning methods for multidimensional psychometric decision trees, in order to examine sources of heterogeneity and to optimize the assessment of psychological constructs in a tailored way.

Human episodic memory comprises the encoding, storage and retrieval of events and related context information, such as information on the temporal and physical context of an event or information on the subjective state during the event experience. The joint encoding and storage of event and context information allows a bound memory representation and thereby enables episodic recollection. For the retrieval and report of episodic recollections, metacognitive processes and reconstructive inferences are relevant beyond genuine memory processes. In our research, we analyze episodic memory representations as well as retrieval and judgment processes by means of experimental designs and statistical models. The models include multinomial processing models for the separation of memory processes and reconstructive inferences, and generalized multilevel models for the analysis of dynamic cognitive processes on the level of individual events.

Learned contingencies are essential for judgment formation and decision making, such as contingencies between social groups and behaviors in stereotyping or contingencies between options and resulting outcomes in decision making. Contingencies that are inductively inferred from information in the social environment, however, can be biased or wrong. For example, information can originate from selective samples or be of limited validity for the judgment to be made, and individuals can overlook relevant moderating variables in their judgments and decisions. Heuristics can exert additional influences on judgments and decisions, like the use of perceived processing fluency in truth judgments of statements or in familiarity statements of events. Our research utilizes different experimental paradigms to investigate potential biases in judgments and decisions due to inductive contingencies and heuristics. The experimental paradigms together with statistical models of data analysis allow us to study the underlying processes as well as individual differences in the judgment and decision effects.