With the increasing availability of ecological momentary assessments (EMA) such as daily dairy and experience sampling measurements, behavioral scientists are better able to investigate the within-person dynamic patterns (i.e., relations among variables across time) underlying symptoms, behaviors, and life events. One prominent challenge in this endeavor is the inherent heterogeneity in individual mental health processes. Others have demonstrated that this heterogeneity requires personalized measurement models to accurately assess constructs of interest. By measurement model, we mean the pattern of how observed variables relate to a latent construct. As an example, depression can be thought of as a latent construct that psychologists often seek to measure. Individuals may differ with regards to which (observed) symptoms relate to their overall (latent) depression levels at a given time point. For one person, the symptoms of sadness, feelings of hopelessness, and irritability may be the best measures of depression over time whereas for another, perhaps sadness, anhedonia, and fatigue are the symptoms that indicate depression.Allowing individuals to have personalized assessments will enable the field to get even closer to personalized treatment plans by better quantifying these somewhat abstract constructs.The current standard is to force all individuals to have the same measurement model, but the field is quickly moving towards adopting personalized measurement models for assessments. Critically, the available methods have a number of issues that prevent reliable personalized measurement models. First, some approaches (such as simply using observed variables) ignore the reality of measurement errors. This causes bias in the effects among latent constructs of interest and can lead to inaccurate inferences regarding anindividuals' process. Second, the number of observations obtained for a given individual is often too small to arrive at person-specific measurement models. Third, the current methods require the assumption of multivariate normality to be met; this is typically not seen in many forms of ecological momentary assessment data. Fourth, many available approaches for arriving at individual- level models do not perform well when the model is misspecified (i.e., the pattern of relations among observed symptoms and latent constructs is incorrect). This prevents a considerable hurdle when attempting to arrive at model structures in an exploratory manner where by definition the correct model is unknown in the beginning.Our project, if funded, would provide researchers with an easy-to-use tool for arriving at personalized measurement models. This can be achieved by building an exploratory approach within a well- understood estimation approach that has a number of desirable properties. Measurement errors would be accounted for, the method will work well even when the number of time points (observations) is less than the number of variables, multivariate normality will not be a required assumption, and misspecifications will not influence the identification of a reliable personalized measurement model.
The present project builds an analytic toolbox that attends to emerging needs for analyzing ecological momentary assessment data. Namely, that individuals are heterogeneous in their psychological physical processes. For this reason, personalized analytic models are needed in order to obtain precise and accurate measurement of abstract constructs (e.g., depression, happiness), and the proposed toolbox would enable this.
