Selecting predictors in regression models: What can regularized regression models do for process-outcome research?

Activity: Talk or presentation typesOral presentation

Description

[Half-day pre-conference workshop, delivered online to hybrid audience.]
It is a commonly encountered situation when planning studies or analyzing data that there is a substantial number of variables that are potentially predictive of a target outcome. Even in the presence of strong theoretical models and previous empirical results, it is often difficult to decide on an adequate approach to model process and outcome variables. In situations where the focus is on prediction and exploration (instead of e.g., causal modelling), 'regularization methods' provide an appropriate and flexible alternative to typical regression models. This workshop will first re-iterate how to test a regression model; and then mainly focus on the second part, exploring how to fit regularized regression models. Fitting regularized regression models (e.g., LASSO or ridge regressions) entails two processes (cross-validation and the determination of the 'penalty') which align naturally with the goals of researchers in many-predictor situations: they identify variables that are promising owing to their predictive value beyond a particular sample. Since the rationale of regularization extends to other model types, the third part of the workshop will point attendees briefly to testing many mediators in regression models; applications in structural equation models (mediators, many outcomes) and confirmatory factor analysis; and applications in item response models. Attendees will need a basic understanding of ordinary least squares regression for the first part of the workshop, but neither experience with regularized regression models nor the specific software packages (R, Stata) is required. The workshop will be based on published papers and worked examples(including syntax) will be made available.
Period6 Jul 2022
Event title53rd International Annual Meeting of the Society for Psychotherapy Research: How, when, and why do people change: The rise of data-informed psychotherapy
Event typeConference