UROP Project
Using AI to Predict Relationship Outcomes from Coded Couple Communication
Psychology, Relationships, Communication, AI
Research Mentor: Dr. James McNulty, He/Him
Department, College, Affiliation: Psychology, Arts and Sciences
Contact Email: mcnulty@psy.fsu.edu
Research Assistant Supervisor (if different from mentor): Megan Marowski She/Her
Research Assistant Supervisor Email: marowski@psy.fsu.edu
Faculty Collaborators:
Faculty Collaborators Email:
Department, College, Affiliation: Psychology, Arts and Sciences
Contact Email: mcnulty@psy.fsu.edu
Research Assistant Supervisor (if different from mentor): Megan Marowski She/Her
Research Assistant Supervisor Email: marowski@psy.fsu.edu
Faculty Collaborators:
Faculty Collaborators Email:
Looking for Research Assistants: Yes
Number of Research Assistants: 1
Relevant Majors: Open to all majors
Project Location: On FSU Main Campus
Research Assistant Transportation Required: Remote or In-person: In-person
Approximate Weekly Hours: 5-10, Flexible schedule (Combination of business and outside of business. TBD between student and research mentor.)
Roundtable Times and Zoom Link:
Not participating in the roundtable
Number of Research Assistants: 1
Relevant Majors: Open to all majors
Project Location: On FSU Main Campus
Research Assistant Transportation Required: Remote or In-person: In-person
Approximate Weekly Hours: 5-10, Flexible schedule (Combination of business and outside of business. TBD between student and research mentor.)
Roundtable Times and Zoom Link:
Not participating in the roundtable
Project Description
BackgroundBehavioral observation of couple interactions provides exceptional predictive validity for relationship outcomes. Gottman (1994) demonstrated that observational coding of 15-minute conflict discussions could predict divorce with 90% accuracy, and this predictive power has been validated longitudinally, with observational data from early marriage predicting relationship dissolution 6 years later (Gottman & Levenson, 1992). Despite this evidence for observation’s superior validity, most relationship research relies on self-report measures, which are subject to sentiment override bias — the tendency for global relationship satisfaction to bias reports of specific behaviors (Weiss, 1980) — and demonstrate limited predictive validity for relationship outcomes (Joel et al., 2025).
Work in our laboratory (McNulty et al., 2021) and elsewhere has demonstrated observation’s explanatory power for understanding relationship processes, yet this work required substantial resources, illustrating a key limitation of current observational methods. Recent studies have identified distinct relationship satisfaction trajectories (Joiner et al., 2023) and documented that distressed couples show more rigid communication patterns (Eldridge et al., 2007), yet these studies examine either satisfaction without behavioral observation or behavior at only a single timepoint.
What remains unexplored is whether observational coding of couple communication at multiple timepoints can predict longitudinal changes in relationship outcomes. Do couples whose communication patterns change over time show corresponding changes in satisfaction? Which communication properties best predict relationship change? Research has established that observed communication patterns predict relationship satisfaction over time (Cohan & Bradbury, 1997), yet we lack understanding of whether changes in communication drive changes in outcomes. This research has been limited by practical constraints, as human-coding multiple interactions from hundreds of couples across years requires substantial time and resources, making large-scale longitudinal observational studies economically infeasible for most researchers.
Research Question
The central research question is: Do AI-generated behavioral codes from couple conflict transcripts predict longitudinal changes in relationship satisfaction and dissolution?
Specifically, we examine whether changes in communication patterns over time — such as reductions in opposition, increases in integration, or changes in interaction rigidity — predict whether couples maintain, improve, or decline in relationship satisfaction, or ultimately dissolve their relationships.
Methodology
This study will pool couple interaction transcripts from multiple longitudinal studies with assessments at baseline, 6 months, and 12 months. AI-generated behavioral codes will be created using the Verbal Tactics Coding System (VTCS; Sillars et al., 1982), which codes each speaking turn as expressing opposition (critical, blaming, attacking behavior) or integration (cooperative, solution-focused behavior). Using these coded data, we will employ multilevel modeling to examine how communication patterns change over time within couples and whether these changes predict relationship outcomes. This approach identifies which specific communication properties best predict downstream relationship change.
Significance
This research demonstrates the potential of AI-assisted behavioral coding to enable large-scale longitudinal observational research that addresses fundamental questions in relationship science. By reducing the resource demands of behavioral coding, AI makes it feasible to conduct studies examining how couples’ communication patterns change over time and predict relationship outcomes. This work establishes AI-generated codes as a valid tool for predicting meaningful relationship outcomes, validating its utility for future large-scale longitudinal investigations and potentially expanding observational research accessibility beyond well-resourced laboratories.
Research Tasks: Literature review, data cleaning, data analysis, presentation
Skills that research assistant(s) may need: Statistical analysis (recommended)
Mentoring Philosophy
Our mentoring philosophy is student centered, beginning by understanding their goals and strengths, which we intentionally build upon throughout the project. By recognizing what motivates each student, we create a learning environment tailored to their needs and interests.We want our students to genuinely understand the research they’re doing, so we pair hands-on work with readings from psychological journals that give context for each phase of the research project. When we do statistical analyses, we walk through the logic of what we’re doing and why, rather than treating it as a black box. This matters because we want students to feel confident explaining their work, as well as be prepared for future research projects or graduate school, where analytical thinking is essential.
We also recognize the importance of mistakes and challenges. We create an environment where it’s encouraged to ask questions, work through feedback, and learn from setbacks.
Finally, we give students real ownership over their work. They’re involved in all aspects of the project, from the initial hypothesis forming all the way to presenting the findings. We also share our own experiences as researchers, showing students that science is messy and iterative, and that motivation and persistence are of utmost importance.
We’re aiming to develop researchers who are thoughtful, curious, and confident in their ability to contribute to the field.