To understand the underlying factors that are common in mood transitions across different cohorts, we use Bayesian nonparametric factor analysis. When there are more than one data cohort and they are related, it is useful to perform joint factoranalysisthatexploitsharedstatisticalproperties.Anaive way to perform joint factor analysis is to augment the data from each cohort into a single dataset as if they come from the samesourceandperformfactoranalysis.However,thisisoften suboptimal because each data cohort has its own variations, differingindistribution.Sharedsubspacelearningmodels[11], [12]–[14], [16] exploit the sharing strengths, whilst preserving the individual variations. They learn common factors across cohorts but also ones specific to each cohort. We show how to applyourrecentmodel[11],[16]tothisapplication,extracting shared and individual factors across the cohorts being considered. This model has two main advantages: it uses Bayesian nonparametric prior, thus not requiring a priori parameters such as numbers of factors, and it allows us to analyze the individual and shared aspects across 18 cohorts in a rigorous Bayesian framework. This allows us to understand the mood transitionpatterns.Weshowquantitativelythatmoodtransition to negative moods are commonly seen in cohorts with low social capital; cohorts with high social capital mostly have positive mood transitions.