Between- and within-person differences using mixture factor analysis
In their enthusiasm about the new technological possibilities, researchers have been eagerly gathering time-intensive longitudinal data (e.g., with experience sampling methodology) without much regard of crucial measurement issues, partly because proper data-analytical tools were lacking. To close this gap, in her PhD project, Leonie V.D.E. Vogelsmeier develops a new approach for evaluating within-person changes and between-person differences in measurement models underlying participants’ answers in (time-intensive) longitudinal data.
Background: Differences and changes in measurement models underlying participants’ answers
Time-intensive longitudinal studies such as the currently booming ‘Experience Sampling Methodology’ data allow one to investigate daily-life dynamics of psychological ‘constructs’ (or ‘factors’) such as well-being or depression within persons over time. The validity of these studies, for example in terms of the allocation and adaptation of treatments over time, may be impaired by distortions of the measurements of the relevant constructs, such as response styles or differing interpretations of questionnaire items. Such response styles and interpretations may differ between persons but also change within a person over time. For example, participants may lose their motivation to fill in the questionnaire at some point in time and start to give extreme responses to all the items without much regard to the content. Or, participants may learn about a construct such as depression in therapy and therefore change certain item interpretations.
The PhD project
In her PhD project, Leonie develops a new approach to evaluate person- and time-point-specific distortions of the construct measurements, while taking into account the specific characteristics of (time-intensive) longitudinal data. More specifically, the new approach, called latent Markov factor analysis (LMFA), extends mixture factor analysis and clusters subjects and time-points within subjects according to their factor model. The ‘factors’ correspond to measurement model dimensions. This way, researchers can explore which measurement models are underlying the data, for which time-points they apply and thus, which observations are validly comparable. Furthermore, such insights may be substantially interesting. For instance, in personalized medicine, detecting the onset of response styles is crucial for valid decisions about treatment allocation over time as response styles may be related to the onset of depressive episodes.
The project team
The project started in July 2017 and was funded by a Veni grant and a Research Talent grant from the Netherlands Organization for Scientific Research (NWO). Next to the PhD-candidate Leonie V.D.E. Vogelsmeier, the core team consists of her supervisor Dr. Kim De Roover and promotor Prof. Dr. Jeroen K. Vermunt (all employed at the Department of Methodology and Statistics at Tilburg University). They furthermore seek for collaborations with applied researchers to apply the new methods to empirical data. If you want to know more about the research project or if you think that you have experience sampling methodology data that may be interesting to analyze with LMFA, please contact Leonie:
Detailed information about the new approach can also be found in the first two PhD project articles:
Vogelsmeier, L. V. D. E., Vermunt, J. K., Böing-Messing, F., & De Roover, K.
Continuous-time latent Markov factor analysis for exploring measurement model changes across time. Methodology. Read the paper…
Vogelsmeier, L. V. D. E., Vermunt, J. K., van Roekel, E., & De Roover, K. (2019). Latent Markov
factor analysis for exploring measurement model changes in time-intensive longitudinal studies. Structural Equation Modeling: A Multidisciplinary Journal, 26, 557–575. doi:10.1080/10705511.2018.1554445 Read the paper…
Written by: Leonie V.D.E. Vogelsmeier