Paper Summary
Share...

Direct link:

Latent Transition Analysis for Intervention Students

Tue, April 26, 11:30am to 1:00pm PDT (11:30am to 1:00pm PDT), Marriott Marquis San Diego Marina, Floor: South Building, Level 3, Balboa

Abstract

Understanding heterogeneity in treatment effects has gained much attention in intervention research (Supplee et al., 2013) and is integral in understanding the efficacy of educational programs and interventions. While there are a range of approaches to understand treatment effects, latent transition analysis (LTA) is naturally suited to exploring theories about heterogeneous treatment effects. The LTA model is distinguished by modeling transitions between discrete time points with variation captured by a series of repeated latent categorical variables. Theories with salient developmental periods, such as early childhood where transitions between life phases are present, fit well with the assumptions of this model. At each timepoint the mixture models that compose an LTA capture variation by sorting individuals into classes and modeling transitions among the classes.

In this study we evaluate a behavioral intervention administered to 4th graders measured across three time points. Previous research (e.g., Duchnowski et al. 2002) has shown up to 25% of students may be at risk of developing a behavioral difficulty and may need intervention. Behavioral profiles were measured by four standardized indicators capturing behavioral challenges and social competence. Four profiles emerged at pre-test, which were similar for control and intervention groups. Profile 1 was labeled At-Risk and was characterized by elevated levels of behavioral challenges and depressed levels of social competence while profiles 2-4 were marked by differing levels of typical development. Preliminary analyses indicate, at post-test, the At-Risk profile in the intervention group exhibited fewer behavioral challenges compared to the analogous control group profile. The typically developing profiles remained relatively constant in both intervention and control groups.

There are two common approaches to modeling treatment effects in LTA: 1) through the multiple group framework, where treatment status is the grouping variable, and 2) by specifying the treatment variable as a covariate. In this study we will illustrate both methods and demonstrate the strengths and limitations of each approach. Model specification in LTA requires a stepwise procedure. Enumeration was conducted independently at each time point (Nylund, 2007). Subsequently, the full LTA was specified using both multigroup and auxiliary treatment approaches. For auxiliary variable integration the 3-step ML approach (Vermunt, 2010) and the newer 2-step approach (Bakk & Kuha, 2018) were used.

Recently there has been attention to models that highlight variation in treatment effects with the goal of targeted or tailored interventions which maximize efficacy and acknowledge individual differences (Kent & Hayward, 2007). Using this framework, we demonstrate how the LTA model can be used to identify for whom a treatment is most effective and whether there are individuals who may or may not benefit from a given intervention. That is, LTA can accommodate complex theories of heterogeneous treatment expressions. Alternative longitudinal modeling approaches may mask effective treatments by averaging over variation in treatment magnitude or direction across groups. By modeling the differences in populations interventions may be tailored specifically to groups with different behavioral profiles increasing the efficacy and utility of a treatment across diverse populations.

Authors