Breakout R – May 17, 2013
Design & Analysis Challenges with Multilevel Implementation Data
Presentation Slides Video of Presentation
Presenters: David C. Atkins, PhD1 & Scott A. Baldwin, PhD
Authors: David C. Atkins, PhD,1 & Scott A. Baldwin, PhD2
1University of Washington; 2Brigham Young University
Abstract: Implementation research often involves multilevel data (sometimes called hierarchical, clustered, or nested data). Examples of such data include patients clustered within providers, providers clustered within sites, and therapist fidelity items clustered within therapists. Such multilevel data present a number of design and analysis challenges. Our presentation will provide a brief, general overview of multilevel models and then focus on specific challenges related to implementation research. Topics will include:
- How sample sizes at provider and site levels affect the use of multilevel models
- Advantages and disadvantages of randomizing between or within clusters (e.g., randomizing therapists within a site or randomizing sites to treatment condition)
- Power and sample size calculations, including costs (e.g., costs of adding providers vs. sites) and attrition
- How multilevel designs influence the assessment of therapist fidelity, including reliability and psychometric considerations
Abstract: Our goal is to provide a non-technical introduction to these topics, emphasizing concepts as opposed to statistics. Moreover, we hope to have a highly interactive session with input and questions from the audience. Finally, there will be time for general questions on implementation designs at the end of the session, not necessarily specific to multilevel data.