Category: Research Methods and Statistics


Linear and Nonlinear Multilevel Models in CBT Research: Separating Truth From Fiction

Sunday, November 19
8:30 AM - 10:00 AM
Location: Aqua 300 A & B, Level 3, Aqua Level

Keywords: Research Methods | Statistics | Longitudinal
Presentation Type: Symposium

Multilevel models have become a mainstay in longitudinal CBT research (Baldwin et al., 2014). These models can address unique clinical questions that were previously inaccessible, such as how treatment effects differ over time and how diversity in patient backgrounds and symptom presentations can affect treatment outcome (Stegmueller, 2013). Thus, multilevel models provide the ability to make new breakthroughs, and they are also much more easily implemented than they were even five years ago.

However, there are also a number of myths that are perpetuated about these models. For instance, they are not always needed for longitudinal or nested data, and they often address missing data for only one variable in the model (the dependent variable) as opposed to all variables in the model (as is widely reported). For these reasons, misunderstanding of the role of multilevel models can lead to unnecessary additional work and unexpected results (when missing data affects research in a much different way than was originally anticipated). Additionally, there is frequent confusion about when nonlinear multilevel models are beneficial, and historically there has been a lack of consensus on which effect sizes to use with linear and nonlinear multilevel models. Effect sizes in particular are essential, as p-values are insufficient as a sole indicator of model quality (American Statistical Association, 2016).

In this presentation, I will clarify when multilevel models are beneficial and how to know if nonlinear specifications are necessary. Additionally, although there has been a longstanding debate on which effect sizes are appropriate for multilevel models (Johnson, 2014), major breakthroughs have been made in the past several years that allow for easily interpretable effect sizes (Nakagawa, 2017). To illustrate these concepts, I will provide an applied example that focuses on multilevel models in behavioral intervention research (Smith, De Nadai, Storch, & Petrila, in press). Through this presentation, audience members will be able to avoid pitfalls that often occur with linear and nonlinear multilevel models and implement nonlinear models with their own data.

Alessandro S. De Nadai

PhD candidate
University of South Florida, University of Mississippi Medical Center, Texas State University


Send Email for Alessandro De Nadai


Linear and Nonlinear Multilevel Models in CBT Research: Separating Truth From Fiction

Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Linear and Nonlinear Multilevel Models in CBT Research: Separating Truth From Fiction