Exploring Dynamics in Mood Regulation—Mixture Latent Markov Modeling of Ambulatory Assessment Data

Objective To illustrate how fluctuation patterns in ambulatory assessment data with features such as few categorical items, measurement error, and heterogeneity in the change pattern can adequately be analyzed with mixture latent Markov models. The identification of fluctuation patterns can be of great value to psychosomatic research concerned with dysfunctional behavior or cognitions, such as addictive behavior or noncompliance. In our application, unobserved subgroups of individuals who differ with regard to their mood regulation processes, such as mood maintenance and mood repair, are identified. Methods In an ambulatory assessment study, mood ratings were collected 56 times during 1 week from 164 students. The pleasant-unpleasant mood dimension was assessed by the two ordered categorical items unwell-well and bad-good. Mixture latent Markov models with different number of states, classes, and degrees of invariance were tested, and the best model according to information criteria was interpreted. Results Two latent classes that differed in their mood regulation pattern during the day were identified. Mean classification probabilities were high (>0.88) for this model. The larger class showed a tendency to stay in and return to a moderately pleasant mood state, whereas the smaller class was more likely to move to a very pleasant mood state and to stay there with a higher probability. Conclusions Mixture latent Markov models are suitable to obtain information about interindividual differences in stability and change in ambulatory assessment data. Identified mood regulation patterns can serve as reference for typical mood fluctuation in healthy young adults. Abbreviations AA = ambulatory assessment MLM = mixture latent Markov MMQ = Multidimensional Mood Questionnaire BIC = Bayesian Information Criterion AIC3 = modified Akaike Information Criterion