Because it combines multiple regression with the estimation of latent, or unmeasured, variables, Structural Equation Modeling (SEM) has become the dominant analytical paradigm for most social science research. The purpose of this AMASS is to relay and review the critical SEM concepts and procedures necessary for the informed, applied clinical researcher. This AMASS is intended for novice users with some SEM experience who are looking to brush up and expand their skill set. Specifically, I will briefly review fundamental SEM concepts and definitions (e.g., model-fit), and discuss and illustrate how SEM can be used to address common clinical research questions using current recommendations. Further, I will focus on interpreting and reporting results, as well as on common issues that may obscure results. Topics will be limited to continuous outcomes, and will include assessing group differences, mediation, moderation, and bootstrapping; advanced topics may be included if time permits (i.e., measurement invariance, power analysis). Examples and references will be provided at the workshop and be available online. Although there will not be enough time to demonstrate navigation through SEM programs, all examples will include instructions for each of the four major SEM packages: Lisrel, EQS, AMOS, and MPlus.

You will learn:

• SEM terminology and "rules of thumb"
• How to conduct and interpret the appropriate SEM model for common research questions using current recommendations key issues and assumptions for each analysis and how to address them

Recommended Reading: Users should have some experience with SEM and should be comfortable with the fundamental concepts of regression. A good resource to review basic regression issues and interpretation includes

Lewis-Beck, M.S. (1980). Applied regression: An introduction. Sage University Paper series on Quantitative Applications in the Social Sciences, 07-022. Thousand Oaks, CA: Sage.

Clinical research is often focused on longitudinal questions and data: Is CBT more effective than treatment as usual at reducing panic attacks? What is the daily association between drinking urges and PTSD symptoms across 30 days? Classical approaches to longitudinal data, such as repeated measures ANOVA, can be quite limiting and are largely being replaced by newer methods. This AMASS will focus on using hierarchical linear modeling (HLM; also called multilevel or mixed-effects modeling) to analyze longitudinal data. I will present an applied introduction to HLM that focuses on practical skill-building in using HLM to analyze data. We will discuss motivations for using HLM over classical approaches, level-1 and level-2 equations, fixed and random effects, and how to interpret output. Two examples will be used throughout the presentation: a randomized clinical trial of couple therapy and a daily diary study, which will be compared and contrasted in terms of how HLM is applied to these different designs. We will also discuss missing data in HLM, including assumptions about missing data and what to do when these are violated. Both datasets will be made available to participants along with computer code for HLM, SPSS, R, Stata, and SAS to fit the models presented.

You will learn:

• Advantages of HLM over repeated measures ANOVA for analyzing longitudinal data
• How to interpret HLM output including fixed and random effects
• How to Fit basic HLM models to longitudinal data using one of the following statistical packages: HLM, SPSS, R, Stata, SAS

Recommended Readings:

Atkins, D. C. (2005). Using multilevel modeling to analyze couple and family treatment data: Basic and advanced issues. Journal of Family Psychology, 19, 98-110.

Nezlek, J. B. (2001). Multilevel random coefficient analyses of event- and interval-contingent data in social and personality psychology research. Personality and Social Psychology Bulletin, 27, 771-785.