PUB - Publikationen an der Universität Bielefeld

This thesis aims to make latent state-trait (LST) models for experience sampling data available to a broader research population, and demonstrate their potential for theory building in psychological sciences and beyond. The developments outlined in this thesis should thus facilitate LST analyses of experience sampling data where these models can suitably address substantial research questions. LST models mainly differentiate between person- and situation-specific influences on measurements. Additionally, they can account for carry-over effects between measurement occasions. In experience sampling studies, participants regularly respond to questions about their current experience, usually several times a day for multiple days. Due to this high frequency of data collection, experience sampling LST models have a large number of measured variables. This complicates the implementation of these models for researchers, mainly because (1) the specification of LST models for experience sampling data in statistical software is complex and error-prone and (2) it is difficult to judge the model fit of structural equation models with many measured variables. The first challenge -- model specification -- is addressed with a new user-friendly browser app and R-function. Both tools are specifically designed for the specification and estimation of LST models suitable for experience sampling data, and are included in the R-package _lsttheory_. Their use is outlined in a tutorial, which first provides a detailed introduction to LST theory and models suitable for experience sampling data. The software and interpretation of results are illustrated with an empirical application about well-being in everyday life. The software includes 9 pre-defined models. Invariance assumptions of all parameters can be adjusted freely, allowing for more than 100.000 possible LST models. Additionally, covariates can be added to further explain variance in the traits. The second challenge -- model fit evaluation -- is addressed with a new block-wise evaluation approach. In this approach, fit indices are computed for each period (e.g., days in an experience sampling study) from the empirical and model-implied (co)variances of the variables uniquely associated with each period. Degrees of freedom are simulated for each block. Two simulation studies show that block-wise fit evaluation is a suitable alternative to typical global evaluation in the most typical experience sampling data scenario. Here, global evaluation shows unreasonably high model rejection rates, while block-wise evaluation can (1) more reliably identify correctly specified models and (2) correctly reject most misspecified models. Block-wise indices also provide additional information about the fit within each period, thus revealing if some parts of the model fit better than others. Finally, two empirical research applications demonstrate the use of LST models for theory building in the context of experience sampling studies. First, research questions about well-being in everyday life are addressed with a series of LST models. The results show that well-being is highly consistent across five days and only a small portion of trait variance is due to differences in BIG 5 personality dimensions. The second application explores perceived interdependence in everyday social interactions. LST analyses of three dimensions of situational interdependence -- conflict of interest, asymmetry of power and mutual dependence -- support a dynamic model of situation perception. In this model, perceivers have a past which shapes their future selection into situations and their experience of situational interdependence. Overall, the research presented in this thesis broadens the applicability of LST models in experience sampling settings. There are many possible research questions in psychology and beyond that could suitably be addressed with LST models. The methodological developments and empirical demonstrations aim to facilitate the application of LST theory in experience sampling studies.