(PS8-17) Replicating Couple Profiles in a Low-income Help-seeking Sample Using a Couple-centered Approach: A Cautionary Tale

A person-centered approach has been widely used to differentiate distinct groups of individuals. This method has gained popularity in recent years for classifying couples when applied to couple data, also known as a couple-centered approach. Considering such modeling techniques are data driven by nature, it is difficult to ascertain whether the identified groups truly represent subgroups existing in the population or if they are sample specific. This study seeks to explore this replicability issue by analyzing an independent sample that closely resembles the sample in a previous study (Le et al., 2022) and investigating whether the five previously identified couple profiles can be replicated.

This independent sample of mixed-gender help-seeking low-income couples (N = 1126) closely mirrors the original sample with respect to demographics, recruitment strategies, relationship functioning, and assessment tools. On average, participants were 33.35 years old (SD = 7.83), 66% identified as non-Hispanic White and 19% as African American. Most (75%) were either married or engaged, fell below the clinical cut-off for relationship distress (77%) and reported an average relationship length of 5.8 years (SD = 5.15). Latent profile analyses were conducted with indicators from four relationship domains, namely communication, commitment, emotional support, and sexual satisfaction, that were identical to those used in Le et al. (2022) study; both partners reported on the exact same measures across both studies. Model selection was based on the comparison of (a) model fit indices, (b) theoretical interpretability, and (c) profile stability.

              When using all four relationship indicators, the optimal model was not identifiable due to the relatively low prevalence of male partners reporting low levels of commitment (7.7% in this sample compared to 12% in Le et al., 2022). Post hoc exploratory analyses were conducted, using the three remaining indicators either with commitment excluded or as dichotomized indicators. In both cases, three of the five couple profiles identified in Le et al. (2022) that were not differentiated based on commitment levels were replicated, including: (1) conflictual passionate couples, characterized by higher-than-sample-average levels of sexual satisfaction, lower-than-sample-average levels of communication, and sample-average levels of emotional support; (2) companionate couples, characterized by lower-than-sample-average levels of sexual satisfaction and sample-average to higher-than-sample-average levels of emotional support and communication, and (3) satisfied couples, characterized by higher-than-sample-average levels in all three domains. The two couple profiles characterized by asymmetrically committed relationships were not replicated. The three replicated profiles observed across two independent samples support the reliabilities of these profiles among low-income help-seeking couples. However, questions about the reliability of classes derived from this approach remain, considering that we were not able to identify an optimal solution with all four indicators. Taken together, these findings underscore the importance of replication work with similar samples.