Aim 2b

Goal: To test the relative effectiveness of PA for groups at higher and lower risk as suggested by baseline levels of academic and emotional measures, and perceptions of peer group behavior

Hypothesis: Students who are at higher risk of negative student outcomes will experience larger PA effects than students who are less at risk.

In Aim 2b, risk is defined as issues related to life circumstances (e.g., residential mobility, SES, family engagement), environment (e.g., neighborhood disorder, school climate), and emotional well-being (e.g., baseline stress, anxiety, self-esteem) rather than risks associated with baseline levels of behavioral dependent variables (e.g., substance use, bullying). For individual risk variables, moderation analyses will be conducted, as represented in Figure 6 & 7 for time-invariant and time variant moderators, respectively.

For risk constructs in which the moderator represents a collection of risk variables, the risk moderator will be tested by first investigating data-reduction procedures for the risk construct, and then analyzing a model similar to the one presented for a mediation model in Figure 5, except altered to reflect a test of moderation as in Figure 7.

The findings will be examined and conclusions drawn in the same fashion described in Aim 1a. One issue that may be encountered is that the complexity of the hypothesized risk moderator may exceed what can be established within a moderation model with the available data (e.g., because of small group sizes, limited variation).

In these instances, a 3-step approach will be taken that will group students per their risk level and then test to determine if group membership helps to explain PA effect differences.

The first step will be to use theory to identify the set of risk variables that represent one risk latent construct, and use factor analysis to support the selection of variables identified by theory. The second step will be to conduct a Latent Class Analysis (LCA) using the set of variables identified in step one (Collins & Lanza, 2013; Lanza & Cooper, 2016). The LCA will place the students into the optimal number of groups that represent unique risk levels. In the final step, a moderation model will be analyzed, similar to Figure 6 using the new group membership for the moderator. Using the 3-step approach will extend the ability to test additional risk factors and to identify moderators associated with PA effects on student outcomes.