Characterizing heterogeneous treatment effects of couple relationship education : a machine learning approach

Intimate relationships form the cornerstone of personal well-being and societal stability, yet maintaining a stable and satisfying relationship is challenging for many couples. To enhance relationship functioning as well as prevent deterioration of relationship quality over time, Couple Relationship Education (CRE) was developed as a type of preventive intervention that provides intimate partners with knowledge and communication skills (Halford et al., 2001). Although CRE has been shown to be effective in some circumstances, a large body of research indicates that treatment effects of CRE exhibit a great deal of variability depending on the pretreatment conditions and the characteristics of the program attendees (Wadsworth & Markman, 2012). Given that significant federal expenditures have been invested to disseminate CRE to diverse populations of couples (via the Healthy Marriage and Relationship Education initiative), a better understanding of the heterogeneity of CRE treatment effects is needed to ensure that couples receive an intervention that is effective for them. Unfortunately, the existing literature has failed to account for the complex and intertwined nature of pretreatment risk factors, leading to inconsistent and inconclusive results. The current study addresses the complex statistical challenges by using machine learning techniques, providing a granular analysis of how each risk factor contributes to heterogeneous treatment effects of CRE. Study 1 employed causal forests (Athey et al., 2019) to investigate the extent to which pretreatment risk factors contribute to heterogeneity in treatment outcomes using data from a large-scale randomized controlled trial (RCT) of couple relationship education (N = 6,298 couples). Findings reveal heterogeneous treatment effects on relationship happiness and negative emotions and behaviors at 12-month follow-up. Participants with higher psychological distress and lower baseline relationship happiness experienced greater improvements in relationship happiness, while those with higher psychological distress and perceived stress showed more significant reductions in negative emotions and behaviors. Study 2 cross-validated these findings by applying the trained machine learning model from Study 1 to another large-scale RCT of CRE (N = 1,595 couples). Results confirmed the accuracy of the estimations and revealed similar heterogeneity in treatment outcomes, underscoring the generalizability and robustness of the Study 1 findings.

Name thesis:degree_name

Doctor of Philosophy

Discipline thesis:degree_discipline

Human Development and Family Sciences

Grantor

The University of Texas at Austin

Year dc:date.issued

2025

Author dc:creator

• Chen, Po-Heng

Advisor dc:contributor.advisor

• Williamson, Hannah C.

Committee members dc:contributor.committeemember

• Neff, Lisa
• Timmons, Adela C
• Gleason, Marci E.J.

• Causal forests
• Supporting healthy marriages
• Close relationships
• Intervention

Language dc:language.iso

English

Identifier URI

https://doi.org/10.26153/tsw/62025

OAI identifier oai:identifier

oai:repositories.lib.utexas.edu:2152/134703