4 - (SYM 30) Making Relationship Education Shiny: A Recommendation Algorithm for Varying Levels of Coaching

In-person relationship education classes funded by the federal government have problematic attrition rates and do not have the intended effects. Recently, investigators in this area have innovated the curricula to make these programs briefer and web-based, increasing completion rates and improving relationship distress. One area that requires additional attention is whether baseline demographic characteristics (e.g., race, ethnicity, age), measures of individual distress (e.g., psychological distress, perceived stress), measures of relationship distress (e.g., relationship satisfaction, communication), and motivation to improve relationship functioning (e.g., grit, motivation to change), can be leveraged to create a web-based application that can recommend (i.e., predict) what method and intensity of practitioner contact is needed for a given couple to see the largest improvements in relationship satisfaction and program completion during the online relationship education. Data for this study come from a large Sequential and Multiple Randomized Trial (N = 1,246 couples) where varying levels of coaching (e.g., full coach, contingent coach, email only, or a waitlist control) was the primary variable of interest. Using machine learning, random forests, precision medicine, and counterfactual modeling techniques, nearly all variables reliably predicted improvements in relationship satisfaction and program completion. Baseline variables accounted for 41.6-63.1% of the variance in likelihood of program completion and 51.0-84.9% of the variance in change in relationship satisfaction between coaching conditions. In reaction to the amount of variance explained, using Shiny in R, a web-based application was created to allow practitioners to recommend unique couples to a level of coach contact that best meets their unique needs (https://gabe-hatch.shinyapps.io/hatchdamh/; this can take a few minutes to load given the computational intensity). Future studies could explore if the recommendation algorithm leads to improved effect sizes and completion rates.