Skip to main content

Aim 3

Goal – To investigate the effects on student outcomes of variations in:

  • program implementation at the school (e.g., counselor) and classroom levels (i.e., fidelity) and
  • student-level (e.g., student participation, family engagement, etc.) exposure (i.e., dosage)

We hypothesize that:

  • variations in PA implementation fidelity will help explain PA effects on student outcomes
  • measures of dosage mediate PA effects on student outcomes

The interest is in testing whether implementation mediates PA effects, moderates PA effects, does both or neither of the two. With only seven schools implementing PA out the 14 schools in the study, and only about twice as many classrooms for our cohort of students, there are not enough schools or classrooms to model school/classroom-level implementation measures. Therefore, the following will be conducted: propensity score analysis (Foster, 2003; Rosenbaum & Rubin, 1983), descriptive and correlation analyses, and percentage and graphical comparisons to test for possible relationships between distal student outcomes and implementation fidelity.

In addition to implementation, it is believed that PA exposure (i.e., dosage and fidelity) measured at the family and student levels will help explain PA effects. The hypothesis is that measures of dosage mediate PA effects on student outcomes. To test this hypothesis, mediation analysis will be conducted using models similar to Figures 3-5, depending on the type of mediation variable tested (i.e., time variant, time-invariant, or latent construct). If it is discovered that dosage does not mediate PA effects,  dosage will be tested as a possible moderator following the procedures described earlier for testing moderation.

In addition, it is believed that the extended regression models (ERMs) in Stata will provide additional opportunities to investigate data validity (e.g., teacher/parent outcome measures, scale scores, etc.) and moderators of student outcomes. ERMs address multiple problems when modeling data (StataCorp, 2017, pg. 5).