To put asunder: Are there perils of partialing in actor–partner interdependence models?

Abstract

In research on couples, statistical adjustment (i.e., partialing) for correlations between partners’ parallel scores is common and useful, as in the actor–partner interdependence model. Original and partialed scores are typically interpreted as assessing the same construct, but this may not be a valid assumption. Other approaches to nonindependence—such as common fate modeling—may better represent some couple constructs. This study of 300 couples utilized participants’ interpersonal circumplex ratings of partners’ typical behavior during marital interactions to evaluate the interpersonal meaning of unadjusted and partialed forms of the Marital Adjustment Test (MAT), a measure of overall relationship quality, and the Quality of Relationships Inventory-Support (QRI-S) and Conflict (QRI-C) scales, which measured perceived support from and conflict with the partner. After partialing partners’ scores, MAT and QRI-S scores were substantially less closely associated with ratings of partners’ warmth, their primary expected interpersonal correlates. Partner-partialed QRI-C scores were substantially less closely correlated with ratings of partners’ hostility and were associated with a somewhat more controlling form of hostility. In contrast, partialing partners’ trait optimism scores resulted in minimal changes in interpersonal correlates of this personality characteristic. Couple-level MAT, QRI-S, and QRI-C variables representing overlapping variance across partners while partialing unshared variance in spouses’ scores (i.e., common fate scores) had highly similar interpersonal correlates when compared to unadjusted versions. Potential alterations in construct validity resulting from partialing partners’ scores warrant interpretive caution, and alternative analytic frameworks (e.g., the common fate model) may better maintain the construct validity of some dyadic measures.

Good, bad, and beyond: Perils of partialing positive and negative couple variables within individuals.

“…Figure 2 depicts point estimates of IPC locations for unadjusted and partialed QRI scales, with 95% CIs for SSM amplitudes (radial CI) and angular displacements (angular CI) as boxes around each point. Wives’ and husbands’ QRI-support scores were associated with IMI ratings of partners as clearly warm and somewhat submissive (see Table 1 and Figure 2), as in prior results from this sample (Smith et al, 2010, 2022). Compared to original scales, partialed QRI-support scores (i.e., controlling QRI conflict) had substantially smaller amplitude values, indicating a large decrease in specific interpersonal content: wives’ difference = −.27 (CI = −.33, −.21); husbands’ difference = −.22 (CI = −.27, −.18).…”

“…As in the prior reports (Smith et al, 2010, 2022), wives’ and husbands’ QRI-conflict scores were associated with IMI ratings of partners as clearly hostile and somewhat dominant. Within-partner partialed scales (i.e., controlling QRI support) were also associated with ratings of hostile dominance, but compared to the original scales, partialed versions had substantially smaller amplitudes: wives’ difference = −.29 (CI = −.37, −.23); husbands’ difference = −.25 (CI = −.31, −.20).…”

“…In a prior report (Smith et al, 2022), we demonstrated that partialing partners’ parallel scores on relationship quality measures, akin to adjustments in actor–partner interdependence models (APIM; Kenny & Cook, 1999), reduced associations with their primary interpersonal correlates. Partner-partialed measures (i.e., adjusting for partners’ scores) of marital adjustment and support from the spouse became less closely correlated with ratings of spouses’ warmth, and marital conflict became less closely correlated with ratings of the spouses’ hostility.…”

“…In prior analyses (Smith et al, 2010) using a different method, support from the partner was associated with participants’ IPC-based ratings of partners’ warmth; conflict was associated with ratings of partners’ hostility and dominance. As described above, we previously reported effects of partialing partners’ parallel scores (Smith et al, 2022). In the present analyses, we compared unadjusted and within-person partialed conflict and support scores.…”

“…Elsewhere (Smith et al, 2022) we discuss considerations for between-person partialing of partners’ parallel scores as in the APIM. In limiting problematic effects of within-person partialing as examined here, Lynam et al (2006) suggest avoiding closely correlated multifaceted variables, as substantial shifts in construct validity with partialing are more likely.…”