Toward a science of relationship forecasting: Considerations and recommendations

Given the importance of high-quality romantic relationships for health and well-being, researchers have spent decades examining factors that promote them. In doing so, they have identified factors that influence the formation and maintenance of relationships on average, in the moment, and for certain populations. However, making more targeted predictions for particular groups of people or even specific couples-with a meaningful degree of accuracy-remains challenging. We argue that the field is now at a pivotal moment, as technological advances are transforming what is possible. Drawing an analogy to weather forecasting, we suggest that relationship scientists can now gather the kind of data that would be necessary to try to truly map and model relationships as dynamic, complex systems and possibly even "forecast" relationship outcomes. In Part I, we consider the types of data needed to build these targeted predictive models, including dyadic data across timescales (micromoments to lifespans) that are embedded within contexts. We also highlight the need to incorporate heterogeneity in both data collection and analyses. In Part II, we posit that building these models will require the field to embrace interdisciplinary collaborations, leverage tools such as smartphones and artificial intelligence, and reevaluate long-standing assumptions about romantic relationships. While we cannot say with certainty whether more accurate, targeted relationship forecasts are possible, we believe that pursuing them is a worthwhile scientific venture-one that will help direct the field of relationship science regardless of what we find. (PsycInfo Database Record (c) 2025 APA, all rights reserved).