Multilevel Conference

Keynote 2: Dan McNeish

Dan McNeish

Dan McNeish is a full professor in the Psychology Department at Arizona State University, where he is the director of the Mixed Effects Modeling Lab. He has numerous publications on his name in the area of clustered, longitudinal, and time-series data and structural equation and measurement models, many of which are first author publications in esteemed journals such as Psychological Methods. He is an elected member of the Society of Multivariate Experimental Psychology and his research has been acknowledged with many prizes, among which the Distinguished Scientific Award for Early Career Contributions of the American Psychological Association (APA).

Title:  Measurement in Intensive Longitudinal Data


Intensive longitudinal data – where participants are measured many times over a short duration – have recently increased in popularity due to technological advances like wearables and smartphones. Ecological momentary assessment (EMA) is one such study design that has been particularly common in behavioral research when studying mood or affect. Models in EMA studies often feature outcomes and predictors that are created from sum scoring Likert-type or binary item responses at each time point. However, the behavioral literature has recently emphasized the importance of psychometrics and potential benefits of more modern psychometric approaches such as factor analysis and item response theory, especially relating to assessing measurement invariance across time and people. This talk discusses how to combine psychometric approaches with multilevel models for EMA designs to increase measurement precision and more accurately reflect the construct being studied. The proposed model is applied to motivating EMA data from a study on people with binge eating disorder to demonstrate the importance of psychometrics in intensive longitudinal designs. Specifically, statistical models are only as good as the data and variables to which they are applied – if scores on behavioral variables are imprecisely created, conclusions could be driven by inadequate measurement practices rather than the underlying dynamics of the construct of interest.