Teaching Applied Multilevel Modeling

Social work researchers and those in related areas have struggled with analyzing data from studies in which individuals are nested within environments in a multilevel hierarchical structure (e.g., clients nested within agencies). Similar problems arise in analyzing data from repeated measures longitudinal designs and interest lies in understanding factors that influence participants’ trajectories of change over time. Statistical models have been developed to better analyze these types of hierarchically structured data. Such models are typically referred to as multilevel models (MLMs), hierarchical linear models, random-effects models, random-coefficient models, or mixed effects models. These models allow investigators to specify and test meaningful hypotheses regarding relationships between variables observed at different levels (e.g., contextual effects) or observed over time (e.g., trajectories of change).

As with all complex statistical models, multilevel modeling poses challenges to social work educators because students have difficulty applying complex mathematical models to meaningful real-world problems. We recently designed and co-taught an applied MLM course open to graduate students in social work and related areas. In this workshop we will discuss the content of our course, present key “lessons learned,” and suggest strategies for effectively designing and implementing an applied MLM course.

Our MLM course covers two-level cross-sectional and longitudinal MLMs in which there is a single, continuous, normally distributed dependent variable. We designed our course based on the teaching philosophy that MLMs are an extension of linear regression in that so much of what students know about linear regression is applicable, and a prerequisite. Our course is designed to give students a conceptual and practical understanding of linear MLMs and their assumptions, and the ability to specify, test, interpret, and present results of these models. More specifically, students learn how to analyze data in which (a) individuals are nested within social contexts (e.g., social workers nested within agencies) or (b) observations are nested within individuals over time. In the former application, individual outcomes (e.g., social worker burnout) are examined as a function of social context (e.g., agency size), individual characteristics (e.g., social worker education), and/or interactions among social context and individual characteristics (e.g., interaction between agency size and social worker education). In the latter application, students learn how to examine patterns of change in individuals over time (e.g., increasing or decreasing social worker burnout), differences among individuals in patterns of change over time (e.g., increasing burnout over time for some, decreasing burnout for others), and/or effects of individual characteristics (e.g., education) on patterns of change over time (e.g., different patterns of change in burnout over time for social workers with different educational levels).

Designing an applied multilevel modeling course and developing resources for it is challenging and time-consuming. Therefore, we will give participants all of the materials we developed for our course. These include: (1) our course syllabus; (2) MLM course syllabi collected from the web; (3) Power Point presentations; (4) datasets in SPSS and HLM7 format; (5) in- and out-of-class exercises and assignments; (6) MLM websites; and (7) bibliography of MLM texts and articles.