Section outline

  • Lecture:  Multilevel modeling

    Approaches to analyzing complex data to investigate influence across multiple levels

    Faculty:  Kelsey Holt

    Location:       Mission Hall 1407

    • Session Slides:

    • Multi level slides File

    • Session Audio/Video Recording (Access restricted to registered students):

    • Watch URL
      Not available unless: You belong to a group in Registered Students Only

    • Required Reading:

    • Diez-Roux multilevel File
      Not available unless: Your ID number contains 02
    • Subramanian et al. File
      Not available unless: Your ID number contains 02
    • Gausman et al. File
      Not available unless: Your ID number contains 02

    • Assignment: Please consider these questions below when reading each article. If you are unable to attend the class session, please post answers to these questions to the forum by midnight.

      Diez-Roux article:

      1. Multi-level modeling approaches aim to analyze contextual effects beyond individual characteristics. Briefly describe the benefits of conducting a multilevel analysis for health disparities research areas. 

      2. Draw from your own research or interest area to give an example of macro-level factors or group-level properties that influence individual health.

      3. Define the ecological and atomistic fallacies and explain how these fallacies differ from psychologistic and sociologistic fallacies. 

      Subramanian:

      1. Drawing on your area of work, what is an example of a potential individualistic fallacy and how could a multi-level modeling approach help lead to more meaningful analysis of individual level relationships?

      2. Briefly describe the difference between fixed effects and random effects in the models in this paper. In broad stokes, in what circumstances would you use one or the other to examine a higher level variable?

      3. What is a main takeaway you got from reading the authors’ analysis of Robinson’s original paper in historical context?

      Gausman:
      (note: for this article, please try to glean the big picture approach without getting bogged down in the technical modeling details; we can discuss the models more in class if there is interest)

      1. On page 3, discussing Table 2, the authors point out that “the fact that the variance parameters remain significantly different from zero indicates that the geographic variation at these levels is not entirely determined by differences in population composition.” Describe what the authors would have missed if they had not conducted a multi level model with random effects for community and country.

      2. Briefly describe the main findings and their implications for interventions to address childbearing before age 16 in low and middle income countries. How does the examination by country in Figure 1 change your answer, if at all?