The Couple Relationship Skills Inventory: Is There Added Benefit of Partner's Report?

ABSTRACT

Although many established measures assess individuals' relationship skills, relatively few incorporate reports from both partners to predict relationship functioning at the individual and couple levels. The Couple Relationship Skills Inventory (CRSI) is a multidimensional measure of seven relationship skill domains developed using a racially and economically diverse sample. The current study extended the CRSI by incorporating partner‐reports of the focal individual's relationship skills and evaluated whether the self‐ and partner‐report version (CRSI‐SP) improved prediction of relationship quality. Using a large community sample (N = 2785), Bayesian confirmatory factor analysis supported the 45‐item, 13‐factor measure. The CRSI‐SP demonstrated good psychometric properties, including reliability, discriminant validity, and measurement invariance across gender and relationship status. The CRSI‐SP explained more variance in individual‐ and couple‐level relationship quality than the original CRSI, supporting improved predictive utility. The CRSI‐SP can provide researchers and practitioners with a practical measure for assessing relationship skills and evaluating couple‐focused interventions.

1. Introduction

Although several reliable and valid measures assess individuals' distinct skills in couple relationships (e.g., Futris et al. ²⁰¹⁰; Gottman and Silver ¹⁹⁹⁹), most existing measures include only self‐report and not a partner's perspective (i.e., a multi‐informant approach). In addition, measures are typically unidimensional and focused on a singular skill instead of a more inclusive collection of relationship skills. Furthermore, most measures of relationship behaviors or skills were developed using racially and economically homogeneous samples of White, middle‐class married couples or individuals (Funk and Rogge 2007). One recent effort addressed several of these limitations by developing the Couple Relationship Skills Inventory (CRSI), a multidimensional measure grounded in a comprehensive framework of relationship skills and validated using large economically and racially diverse samples (Adler‐Baeder et al. ²⁰²²). The CRSI also includes partner‐report items in which one partner reports on the other partner's relationship skills, but their utility was not considered in the initial validation study since it was considered a next step in the instrument development process. Therefore, the current study centered on further validating the established CRSI and explored the value of incorporating both self‐ and partner‐reports of one individual's behaviors in the couple relationship in predicting couples' reports of relationship quality. The resulting Couple Relationship Skills Inventory‐Self and Partner Report (CRSI‐SP) can improve upon studies of self‐report relational outcomes and support more comprehensive multi‐informant approaches to relationship research and evaluation of couple‐focused interventions. Developing measures that perform well across diverse populations is also essential for advancing equitable relationship research and practice.

1.1. Framework for Developing a Couple Relationship Skills Inventory

The original CRSI, and the CRSI‐SP considered in this study, were developed with a priori subscales informed by the National Extension Relationship and Marriage Education Network, a group of researchers and practitioners who compiled and summarized the existing literature on predictors of couple quality with a particular focus on identifying changeable factors. The result was the National Extension Relationship and Marriage Education Model (NERMEM; Futris and Adler‐Baeder ²⁰¹³), consisting of seven core concepts/practices: Choose–attitudes and efforts related to intentionality; Care for Self–efforts to promote individual well‐being; Care–behaviors promoting partner‐oriented positivity; Share–behaviors that promote couples' sense of shared identity; Know–efforts to promote intimate knowledge of partner; Manage–skills for managing couple stress and conflict; Connect–efforts to embed the couple in support networks. This framework has been used to design programmatic resources for supporting healthy couple relationships (e.g., ELEVATE; Futris et al. ²⁰²⁰) and the development of the validated CRSI (Adler‐Baeder et al. ²⁰²²) to enhance internal consistency in assessing the effectiveness of programming for couples.

The established and validated CRSI is a 32‐item, 9‐factor, 7‐subscale measure of relationship skills typically taught in relationship education programs. Reliabilities for the full scale including the 32 items (α = 0.92) and the subscales ranging from 3 to 5 items (α = 0.71–87) are all acceptable. Utilizing a Bayesian confirmatory factor analysis (BCFA), a good fit to the data indicated that the factor structure was validated. A second and third sample were used to establish cross‐validation of model fit and reliability; results demonstrated measurement invariance across samples. Importantly, these validation analyses were conducted using three large, economically and racially diverse community samples, improving upon the historically homogenous samples used in many relationship measure development studies. The authors also established internal discriminant validity, as well as concurrent and predictive validity, of the measure; however, the tests of reliability and validity of the CRSI did not incorporate partner‐report items for the three subscales that assess couple behaviors that are more dyadic in nature. Specifically, the three subscales that assess a person's skills for maintaining intimate knowledge of their partner (Know), their caring behaviors towards their partner (Care), and their skills for managing conflict with their partner (Manage) are observed by both the respondent and their partner; therefore, capturing the partner's perspective of the focal individual's skills in these areas could further enhance the validity of the measurement of these skills.

The CRSI‐SP's design was further informed by the assumptions outlined in interdependence theory (Kelley and Thibaut 1978), which suggests that both dyad members' perspectives and perceptions are critical to understand an individuals' experience within the dyad. This theoretical perspective provides the rationale for incorporating both an individual's self‐report and their partner's report when assessing that individual's relationship skills. The theory's principle of transformation is especially relevant in measuring couples' behaviors, as it focuses on how individuals interpret their experience based on their own perceptions. Utilizing multi‐informant data is therefore an important consideration when studying couple relationships from an interdependence theory perspective. This method can improve upon other measurement efforts that assess couple functioning by having one person in the couple report on both their own behavior and their partner's behavior on the same or similar items. Therefore, an important contribution to the measurement of couple skills is to validate the CRSI‐SP, which incorporates partner reports and assesses whether incorporating partner reports enhances the measurement of relationship skills and improves the prediction of relationship quality. In fact, most existing measures do not consider multi‐informant data, although there are a few existing multi‐informant measures focused on couple relationships.

1.2. Existing Multi‐Informant Measures of Relationship Functioning

Scholars focused on the couple relationship suggest that partners may be able to observe and report one another's behaviors more accurately than they observe their own behaviors (Sanford 2010). Although including multiple ratings on the same behavior or attitude of an individual presents several strengths, such as having more reliable data and widening the opportunities for different data analysis methods, most scholars in both basic and intervention science do not include self and partner ratings (Busby and Gardner 2008). The multi‐informant rating approach is more common in other fields of research, including parent‐child, teacher‐student, and sibling relationship research (De Los Reyes et al. 2019). Relying exclusively on self‐reports assumes these provide the most important, reliable, and valid assessments of the construct; however, there is evidence that suggests people may be the least accurate at rating themselves and often overestimate positive qualities and underestimate negative qualities when compared to reliable outside raters (Rusbult et al. 2000). For example, in one study assessing the validity of self‐report, partner‐report, and observer ratings of conflict communication, self‐reports demonstrated the lowest validity among these three types of reports (Sanford 2010). Of the current measures of couple relationship behaviors that incorporate multiple reporters, survey response measures are most common, and observational measures of relationship behaviors also exist (Weiss and Heyman 2004); however, one noticeable limitation to observational measures is that they are much more difficult to collect in large samples due to the significant time needed to individually schedule and conduct observations (Wang and Ji 2020).

1.2.1. Unidimensionality of Multi‐Informant Measures

In addition, couple relationship skill survey measures that include items for reporting on self and partner often focus on a specific skill (e.g., conflict management assessed by the Revised Conflict Tactics Scale; Straus et al. ¹⁹⁹⁶) and are not holistic, inclusive, measures of relationship skills. Furthermore, measures that include both self‐ and partner‐reports seldom use a multi‐informant approach in scoring, and instead combine each partner's self‐report to represent the couple's behaviors (e.g., Marital Satisfaction Inventory‐Revised; Snyder ¹⁹⁹⁷). While this scoring approach provides a measure of the couple's dynamic, it still reflects each person's report of their own behaviors rather than two perspectives of one individual's relationship skills. A few studies examine congruence between individuals in couples' reports of their couple dynamics (e.g., Browning and Dutton ¹⁹⁸⁶; Chapman and Gillespie ²⁰¹⁹); however, these multi‐perspectives were not incorporated into scoring the skills or evaluated for whether they enhanced predictive validity. Specifically, one of the most widely used multi‐informant measures of relationship behavior is the Revised Conflict Tactics Scale (CTS2; Straus et al. ¹⁹⁹⁶). The measure assesses self‐ and partner‐reports across five conflict‐related domains and has demonstrated strong reliability and validity. However, the CTS2 focuses exclusively on conflict and does not combine partner‐report items with an individual's self‐report in a multi‐informant scoring approach.

Beyond measures focused on a single relationship domain, the Marital Satisfaction Inventory‐Revised (MSI‐R; Snyder ¹⁹⁷⁹) improves upon the singular focus limitation as it is a multidimensional, valid, and reliable measure of functioning in a couple relationship that includes both partners' self‐reports. The measure assesses multiple domains of relationship distress, including communication, time together, finances, sexual satisfaction, family relationships, and childrearing. The MSI‐R, however, focuses on the level of relationship distress rather than the assessment of relationship skills, with lower scores indicating higher marital satisfaction. Further, although the measure itself uses a multi‐informant approach, each partner is scored separately, allowing comparisons of response patterns within the couple rather than combining self‐ and partner‐reports of the same individual.

The Spouse Observational Checklist (SOC; Jacobson et al. ¹⁹⁸¹) further expands the multi‐dimensional assessment of couple relationships by measuring multiple relationship behaviors of a partner and activities that occurred in the couple relationship within 24 h. The SOC was later expanded to include a Self‐Monitoring Checklist that assesses identical self‐report items, broadening the assessment beyond conflict to include skill‐based domains such as companionship, affection, sex, consideration, and communication (Jacobson & Moore ¹⁹⁸¹). However, similar to the MSI‐R, the SOC uses partners' reports to understand interaction patterns and agreement within the couple, rather than combining self‐ and partner‐reports of the same individual. Furthermore, the MSI‐R and the SOC are both considerably expansive measures; both measures contain more than 100 items, presenting a clear barrier for survey‐based studies with community samples.

Similar limitations characterize more general multi‐informant measures such as Achenbach and Rescorla's (2003) Adult Self‐Report (ASR) and Adult Behavior Checklist (ABCL), which combine self‐reports and other‐reports, respectively, on identical items assessing social, family, romantic, internalizing, and externalizing functioning. Although the ASR and ABCL have been used in couple relationship research to examine whether partner reports provide additional information about an individual's functioning (van Dulmen and Goncy ²⁰¹⁰), they were not developed specifically to assess relationship functioning and remain relatively lengthy (126 items). Similarly, the Negative Mood Regulation Scale (NMRS; Catanzaro and Mearns ¹⁹⁹⁰), later adapted to include a partner‐report version (Pu et al. 2019), has been used to compare self‐ and partner‐reports on emotional regulation rather than combine them in a multi‐informant scoring approach. Collectively, the ASR, ABCL, and NMRS come closest to the approach used in the present study, yet none were developed specifically to assess multidimensional couple relationship skills using a multi‐informant scoring approach. This highlights the limited availability of practical, efficient, relationship‐specific instruments that combine self‐ and partner‐reports of the same individual's relationship skills.

1.2.2. Historically Homogenous Samples of Multi‐Informant Measures

Another important consideration in the development of relationship measures is their generalizability, as many were originally developed using relatively economically and racially homogeneous samples. The CTS2 was developed and assessed with a sample of young, dating college students from well‐educated families and may not represent the experiences of racially and economically diverse individuals in relationships across the lifespan. Limited demographic information was reported for the SOC/SMC, although the authors described the sample as highly educated distressed and non‐distressed couples. Similarly, the original MSI‐R publication provided limited demographic information; however, later versions reported more racially, educationally, and economically diverse samples (Snyder 1997).

Additionally, when these multi‐informant measures were developed, the role that social addresses such as gender, sexual identity, and ethnicity/race play in how individuals use and interpret the measures were rarely discussed, even though they may influence couple relationships. For example, there are unique risk and resilience factors associated with race in the context of romantic relationships, including critical contextual factors such as racial discrimination and economic strain (Bryant et al. 2010), that are rarely taken into consideration in measurement development and testing. Gender and sexual identity may also shape how individuals experience and report relationship processes (e.g., Umberson et al. ²⁰¹⁵), underscoring the importance of evaluating whether relationship measures function similarly across diverse groups. Accordingly, measures intended for broad application should be developed and tested using diverse samples and evaluated for measurement invariance to minimize bias that may otherwise be interpreted as meaningful group differences. Consistent with these recommendations, the current study evaluates measurement invariance across gender and relationship status while utilizing a racially and economically diverse community sample.

1.2.3. Limited Guidance on Utility of Measure

Although existing measures often include both self‐ and partner‐reports, there is limited guidance regarding how these reports should be combined or whether doing so improves the prediction of important relationship outcomes. Busby and Gardner (2008) outlined several approaches for utilizing multi‐informant data from couples. The insider's perspective model combines an individual's self‐report of their own behavior and report of their partner's behavior to represent that individual's perception of the relationship. In contrast, the inter‐rater reliability model compares two reports about the same individual (i.e., self‐report and partner‐report), consistent with the multi‐informant approach evaluated in the current study. Finally, the couple‐level model combines reports from both partners (i.e., each partner's self‐report and partner‐report) to characterize relationship functioning at the couple level. These approaches address different research questions and are conceptually distinct from dyadic analytic methods (e.g., Actor‐Partner Interdependence Model or multi‐level modeling), which model interdependence between partners rather than alternative approaches to scoring or operationalizing multi‐informant data.

Studies using the CTS2 and SOC/SMC have primarily examined congruence, or agreement, between partners' reports. Using the CTS2, researchers have generally found considerable disagreement between individuals' self‐reports and their partners' reports of conflict behaviors (Browning and Dutton 1986; Chapman and Gillespie ²⁰¹⁹). Similarly, studies using the SOC/SMC have found only modest agreement between partners regarding the occurrence of relationship behaviors. In contrast, the MSI‐R can be scored separately for each partner or by combining partners' self‐reports to compute an overall couple score. Thus, the congruence, couple‐level, and separate scoring approaches are conceptually distinct from the inter‐rater reliability approach used in the current study, which combines self‐reports from the focal individual with partner‐reports of that same focal individual. However, little research has examined whether combining self‐reports from the focal individual with partner‐reports of that same focal individual improves the prediction of relationship functioning. The limited available evidence suggests that the inter‐rater reliability approach may offer greater predictive utility than congruence or individual rating approaches (Busby et al. 2019), providing the rationale for evaluating the inter‐rater reliability approach in the current study.

1.3. Current Study

Because utilizing multi‐informant data may provide a more accurate assessment of an individual's relationship skills, the current study evaluated the utility of combining self‐ and partner‐reports in the CRSI to create the CRSI‐SP measure. Specifically, we examined whether the CRSI‐SP was a reliable and valid measure of an individual's relationship skills (RQ1) and provided a better prediction of relationship quality than the original self‐report CRSI (RQ2). This study builds upon the established CSRI (Adler‐Baeder et al. ²⁰²²) and draws on Interdependence Theory, which suggests that both partner's perspectives contribute to understanding an individual's experiences within the relationship. The CRSI was initially developed using a racially and economically diverse sample, increasing confidence in its generalizability to racially and economically diverse couples. The current study similarly utilized a racially and economically diverse sample and further evaluated measurement invariance across gender and relationship status. Findings from this study will help inform best practices for researchers using the CRSI‐SP in basic relationship science and for family life educators evaluating the effectiveness of relationship education programs.

2. Methods

2.1. Procedures

A racially and economically diverse sample of couples from the southeastern United States was recruited to participate in two federally‐funded evaluation studies of couple and relationship education (CRE) programs (Adler‐Baeder et al. ²⁰²²; Futris et al. ²⁰²⁰). Recruitment targeted the general community and utilized broad outreach strategies (e.g., flyers, emails, social media, referral agencies, and word of mouth). To be included in the study, respondents had to be 19 years of age or older and in a self‐identified couple relationship. Informed consent was obtained, and individuals completed an online baseline survey assessing demographic characteristics and individual, relational, and parenting functioning before participating in the CRE program. The overall evaluation study and study procedures were approved by the Institutional Review Boards at Auburn University (Protocol # 16‐248 EP 1608) and University of Georgia (Protocol # STUDY00003068). To minimize shared response bias, each partner received a unique survey link and completed the survey independently of their partner. All respondents received a monetary incentive for completing the survey.

2.2. Participants

The analytic sample consisted of one randomly selected focal individual from each of the 2785 couples (N = 5570 individuals) in which both partners completed the baseline CRSI‐SP assessment. The complementary sample (i.e., the remaining partner from each couple) was used to cross‐validate the findings. The analytic sample consisted of 50.5% women, 49.2% men, and 0.3% participants identifying as another gender. Among the 2,785 couples, 77 were same‐gender couples (i.e., both partners identified as the same gender), and 13 were gender‐diverse couples (i.e., at least one partner did not identify as a man or woman). The average age of the participants at baseline was 38 years old (SD_age = 11.39). Approximately half (50.1%) identified as white/Caucasian, 42.7% as Black/African American, and 7.2% as another racial or ethnic identity. The majority (66.7%) were married, and the remaining were in committed relationships (21.9%), engaged (8.8%), or dating casually (2.6%). Most respondents (71.2%) were parents. Household income was diverse: 33.3% reported $39,999 or less; 23% reported between $40,000 to $74,999; 26.4% reported between $75,000 to $99,999; and 17.3% reported greater than $100,000. See Table 1 for complete demographic characteristics of the analytic and complementary samples.

TABLE 1.

Demographic characteristics of participants at baseline.

Baseline characteristics	Analytic sample (n = 2785)	Complementary sample (n = 2785)
Gender (%)
Women	50.5	51.5
Men	49.2	48.2
Other	0.3	0.3
Race (%)
Caucasian/White	50.1	50.1
African American/Black	42.7	42.9
Others	7.2	7.0
Education attainment (%)
No high school diploma	4.1	4.4
High school diploma or GED	37.0	37.1
Technical certification/Associate's degree	15.9	15.7
Bachelor's or advanced degree	43.0	42.8
Annual Household Income (%)
Under $19,999	17.3	17.4
$20,000–$39,999	16.0	16.1
$40,000–$74,999	23.0	22.2
$75,000–$99,999	26.4	27.1
Over $100,000	17.3	17.2
Relationship status
Married	66.7	67.0
Unmarried	33.3	33.0

2.3. Measures

The CRSI‐SP includes 45 self‐ and partner‐report items, adding 13 items to the original scale. Four domains (Self‐Care, Choose, Share, and Connect) were assessed using self‐report only, whereas the three dyadic domains (Know, Care, and Manage) included both self‐report and partner‐report items. Specifically, eight Self‐Care items assessed the focal individual's self‐reported ability to prioritize behaviors that enhance their well‐being; four Choose items assessed the focal individual's self‐reported level of commitment to the relationship; three Share items assessed the focal individual's self‐reported efforts to create a sense of togetherness; and four Connect items assessed the couple's connection to and engagement with supportive family, friends, and community. The Know domain included four self‐report and four partner‐report items assessing the focal individual's knowledge of their partner. The Care domain included four self‐report and four partner‐report items assessing the focal individual's positive relationship behaviors. The Manage domain included five self‐report and five partner‐report items assessing the focal individual's conflict management skills. Items were rated on a 7‐point Likert‐scale. Self‐care, Choose, Know, Manage, and Connect items ranged from 1 (very strongly disagree) to 7 (very strongly agree), whereas Care and Share items ranged from 1 (never) to 7 (more often than once a day). A total of 32 items were utilized from the focal individual (i.e., self‐report assessments), and an additional 13 items were reports from the focal individual's partner about the focal individual (i.e., partner‐report assessments). Higher scores indicate stronger relationship skills based on the individual's self‐report and their partner's report.

Couple Relationship Quality was measured using a mean score of three highly correlated (r = 0.69–0.83) measures of couple functioning used in previous studies (e.g., Adler‐Baeder et al. ²⁰²²; Wei et al. ²⁰²⁵) from both individuals in the couple. The three measures were the Quality of Marriage Index (α = 0.95; Norton ¹⁹⁸³), the Couple Satisfaction Index (α = 0.92; Funk and Rogge ²⁰⁰⁷), and the Confidence and Dedication Scale (α = 0.91; Stanley and Markman ¹⁹⁹²), which were on a scale from 1 to 7. Cronbach's alpha indicates excellent reliability at both individual and couple levels (i.e., combining both partners' responses by averaging their scores; α = 0.95 and 0.96, respectively) for the composite measures.

2.4. Analysis Plan

To address non‐independence within couples, one partner was randomly selected from each couple to serve as the focal individual in the analytic sample. Partner‐reports provided by the focal individual's partner were incorporated into the analytic dataset to evaluate the multi‐informant measurement approach. The complementary sample, in which the other partner served as the focal individual, was used to cross‐validate the measurement and predictive analysis.

Because partner‐report items were not included in the original validation of the CRSI, BCFA in R using the “blavaan” package (Merkle and Rosseel 2018) were first conducted for each of the three subscales that included partner‐report items (Know, Care, and Manage). Descriptive statistics, factor loadings, goodness‐of‐fit indices, and reliability estimates were examined for each subscale. A BCFA was then conducted to evaluate the full CRSI‐SP measurement model, including the covaried self‐ and partner‐reported latent constructs. Factor loadings were examined to evaluate the factor structure, and goodness‐of‐fit indices were used to evaluate the alignment between the observed data and the specified model. The Bayesian framework was selected because it treats model parameters as random variables and estimates a distribution of plausible parameter values (Levy & Mislevy ²⁰¹⁶). Further, the Bayesian approach provides greater flexibility for incorporating cross‐loadings and error covariances without assuming multivariate normality (Levy & Mislevy ²⁰¹⁶; Muthén and Asparouhov ²⁰¹²). Discriminant validity was assessed by reviewing the covariances among the latent constructs. Measurement invariance across gender and relationship status was evaluated sequentially by testing configural, metric, and scalar invariance models.

Finally, covariance structure analyses were conducted to compare the predictive utility of the original CRSI (self‐report only) and the CRSI‐SP (combined self‐ and partner‐reports) for relationship quality. Two regression models were estimated to examine: (1) the association between the original CRSI and relationship quality, and (2) the association between the CRSI‐SP and relationship quality. Because beta coefficients could not be directly compared across models, model fit indices, including chi‐square, AIC, and BIC, and explained variance were used to compare the predictive performance of the two measures. To evaluate the robustness of the findings, all measurements and predictive analyses were cross‐validated using the complementary sample in which the other partner served as the focal individual.

3. Results

3.1. Individual Measurement Models

Three individual BCFAs were conducted for each of the partner‐report CRSI subscales (Manage, Care, and Know skills). All three models demonstrated good fit to the data. Consistent with the original CRSI, the partner‐report Manage subscale was fit as a two‐factor model (i.e., positive engagement and avoiding aggression), with three items loading on the first factor and two on the second factor. Factor loadings for this model ranged from 0.59 to 0.92 (p < 0.01; BRMSEA = 0.08; BCFI = 0.99; BTLI = 0.95; BNFI = 0.99). The partner‐report Care model included four items, with factor loadings ranging from 0.74 to 0.82 (p < 0.001; BRMSEA < 0.01; BCFI = 0.99; BTLI = 0.99; BNFI = 0.99). The partner‐report Know model also demonstrated good fit, with factor loadings ranging from 0.81 to 0.90 (p < 0.001; BRMSEA < 0.01; BCFI = 0.99; BTLI = 0.99; BNFI = 0.99). These measurement models were cross‐validated using the complementary sample (i.e., the partner not randomly selected as the focal individual), yielding nearly identical results. For example, the partner‐report Know model produced factor loadings ranging from 0.80 to 0.90 (p < 0.001; BRMSEA < 0.01; BCFI = 0.99; BTLI = 0.99; BNFI = 0.99).

3.2. Confirmatory Factor Structure of Self‐ and Partner‐Reported CRSI

After confirming the factor structure of the individual partner‐report subscales (Manage, Care, and Know), we evaluated the full CRSI‐SP measurement model. The original CRSI full Bayesian model had nine factors, including two factors for Self‐Care (i.e., empowerment and healthy lifestyle) and two factors for Manage (i.e., positive engagement and avoiding aggression). The CRSI‐SP extended the original model by adding four partner‐report latent factors: one each for Care and Know and two for Manage, consistent with the original two‐factor structure of the self‐report Manage subscale, resulting in a 13‐factor BCFA model. All latent constructs were allowed to covary, and some items were allowed to covary in the model based on the original model structure (Adler‐Baeder et al. ²⁰²²). The CRSI‐SP measurement model demonstrated good fit to the data (p < 0.001; BRMSEA = 0.05; BCFI = 0.93; BTLI = 0.92; BNFI = 0.92), supporting the factor structure that included the additional partner‐report constructs. Standardized factor loadings ranged from 0.45 to 0.89. Cronbach's alpha for the CRSI‐SP was 0.90. Analysis using the complementary sample yielded nearly identical results (i.e., factor loadings ranged from 0.44–0.90; p < 0.001, BRMSEA = 0.05, BCFI = 0.92, BTLI = 0.91, BNFI = 0.92). The full measurement model for the CRSI‐SP is shown in Figure 1.

FIGURE 1.

Full measurement model for CRSI‐SP. CRSI‐SP, Couple Relationship Skills Inventory‐Self and Partner Report.

3.3. Discriminant Validity Testing

Covariances among the latent constructs were generally moderate, ranging from 0.08 to 0.78 (M = 0.41). Covariances between the self‐reported and partner‐reported constructs ranged from 0.42 to 0.76, suggesting that self‐reported and partner‐reported constructs are similar but distinct. The highest covariance was observed between the self‐ and partner‐report Care constructs, whereas the lowest covariance was observed between the self‐ and partner‐report Know constructs. These findings suggest that the partner‐report subscales were related to, yet empirically distinct from, their corresponding self‐report constructs, supporting discriminant validity.

3.4. Measurement Invariance

3.4.1. Gender

To evaluate whether the CRSI‐SP measurement structure was equivalent across gender, we conducted a multi‐group measurement invariance analysis in which gender defined the comparison groups. The configural model demonstrated acceptable fit (CFI = 0.91, TLI = 0.90, SRMR = 0.04, RMSEA = 0.05), supporting configural invariance across gender. Metric invariance was then evaluated. Due to our large sample size, we used comparison of the Goodness‐of‐Fit Indexes (GFIs) between invariance models as the tests of invariance (Cheung and Rensvold 2002). Differences of GFIs between the configural model and the metric were minimal: ΔCFI = 0.002, ΔRMSEA = 0.001, and ΔSRMR = 0.001, which fell within the cutoff changes of ≤0.01 for ΔCFI, ≤0.015 for ΔRMSEA, and ≤0.03 for ΔSRMR suggested by Chen (2007), indicating metric invariance. The scalar invariance model also demonstrated comparable fit to the configural model (ΔCFI = 0.008, ΔRMSEA = 0.002, ΔSRMR = 0.002), which also fell within the cutoff ranges that suggest invariance (≤0.01 for ΔCFI, ≤0.015 for ΔRMSEA, and ≤0.01 for ΔSRMR; Chen ²⁰⁰⁷), indicating that intercepts can be considered equivalent across genders. Thus, scalar invariance was also supported, suggesting that not only men and women interpret the measure items similarly, but also any observed group differences would truly reflect differences in the latent construct rather than systematic item bias.

3.4.2. Relationship Status

Upon establishing scalar invariance across genders, we then conducted another set of measurement invariance models to test whether the CRSI‐SP measure was invariant across relationship status (married vs. unmarried). The configural model also demonstrated acceptable fit for married and unmarried individuals (CFI = 0.91, TLI = 0.90, SRMR = 0.04, RMSEA = 0.05), supporting configural invariance across relationship status. Comparison between GFIs of the configural and metric invariance models indicated metric invariance across relationship statuses for the measure (ΔCFI = 0.001, ΔRMSEA < 0.001, ΔSRMR = 0.001), and comparison between GFIs of the metric and scalar invariance models indicated scalar invariance across relationship statuses for the measure (ΔCFI = 0.002, ΔRMSEA < 0.001, ΔSRMR < 0.001). The establishment of scalar invariance indicated that both factor loadings and item intercepts were equivalent across married and unmarried individuals. Consequently, the latent variables are comparable across groups, allowing valid comparisons of latent factor means across groups. Overall, findings from the measurement invariance models suggested that individuals with the same level of the latent construct would be expected to have similar observed item scores regardless of their genders and marriage statuses.

3.5. Predictability of Self‐Report Only and Self‐ and Partner‐Reports Combined

To compare the predictive utility of the original self‐report CRSI and the CRSI‐SP, we estimated two structural equation models using either the original CRSI or the CRSI‐SP to predict the composite individual‐ and couple‐level relationship quality scores derived from the QMI, CSI, and CD. The original CRSI model included nine latent variables, whereas the CRSI‐SP included 13 latent variables following the addition of the four partner‐report constructs. At the individual level, both the original CRSI (CFI = 0.95, TLI = 0.94, RMSEA = 0.04, SRMR = 0.04) and the CRSI‐SP (CFI = 0.91, TLI = 0.90, RMSEA = 0.05, SRMR = 0.04) demonstrated comparable and acceptable model fit. The CRSI‐SP explained slightly more variance in individual relationship quality (R ² = 0.59) than the original self‐report CRSI (0.57). For models predicting couple‐level relationship quality, calculated by averaging both partners' individual scores, the model fit for the original CRSI (CFI = 0.95, TLI = 0.94, RMSEA = 0.04, SRMR: 0.04) and the CRSI‐SP (CFI = 0.92, TLI = 0.90, RMSEA = 0.05, SRMR = 0.04) was also comparable and acceptable. However, the CRSI‐SP explained substantially more variance in couple‐level relationship quality (R ² = 0.66) than the original self‐report CRSI (R ² = 0.51).

Analyses using the complementary sample cross‐validated these findings. At the individual level, both the original CRSI (CFI = 0.95, TLI = 0.94, RMSEA = 0.04, SRMR = 0.04) and the CRSI‐SP (CFI = 0.92, TLI = 0.90, RMSEA = 0.05, SRMR = 0.04) demonstrated comparable and acceptable model fit, and the CRSI‐SP explained slightly more variance in individual relationship quality (R ² = 0.61) than the original self‐report CRSI (R ² = 0.60). Similarly, comparable and acceptable model fit was demonstrated at the couple level. The original CRSI (CFI = 0.95, TLI = 0.94, RMSEA = 0.04, SRMR = 0.04) explained 54% of the variance in couple‐level relationship quality, whereas the CRSI‐SP (CFI = 0.92, TLI = 0.90, RMSEA = 0.05, SRMR = 0.04) explained 67%.

To formally evaluate whether the four partner‐report factors provided incremental explanatory value beyond the nine original self‐report factors, we compared two nested models while holding the full 13‐factor CRSI‐SP measurement structure constant. In the restricted model, the paths from the four partner‐report factors to relationship quality were fixed to zero; in the full model, these paths were freely estimated. For both the individual‐level and couple‐level relationship quality models, the full model that included the partner‐report factors fit significantly better than the restricted model containing only the original CRSI self‐report factors. For the individual‐level outcome, the full model fit significantly better than the restricted model, Δχ ²(4) = 238.30, p < 0.001. Information criteria also favored the full model: AIC = 359,423.23 versus 359,653.53 (ΔAIC = 230.30) and BIC = 360,520.32 versus 360,727.27 (ΔBIC = 206.96). Results were similar for couple‐level relationship quality, with the full model again demonstrating better results than the restricted model, Δχ ²(4) = 204.50, p < 0.001. Information criteria likewise favored the full model: AIC = 359,415 versus 359,612 (ΔAIC = 197) and BIC = 360,512 versus 360,686 (ΔBIC = 174). Together, these findings indicate that the partner‐report factors contributed meaningful incremental explanatory value beyond the original nine self‐report dimensions. Improvements in both model fit and information criteria suggest that this added explanatory value was achieved without unnecessary increases in model complexity.

4. Discussion

Few multi‐informant measures of relationship skills exist, and those that do are often unidimensional, validated using relatively homogenous samples, and provide limited guidance regarding how to utilize multi‐informant data. Accordingly, the current study expanded the established, reliable, and valid Couple Relationship Skills Inventory (CRSI; Adler‐Baeder et al. ²⁰²²) by including reports from an individual's partner (i.e., 13 additional items). Using data from a racially and economically diverse sample of cisgender individuals in committed relationships, BCFA supported the factor structure for the CRSI‐SP. The CRSI‐SP demonstrated good reliability and validity, and current findings provide preliminary evidence that it may offer modestly improved predictive utility for relationship quality at both the individual and couple levels compared to the original CRSI. The current findings support the use of the CRSI‐SP as a multi‐dimensional, multi‐informant measure for racially and economically diverse cisgender individuals in committed relationships. Furthermore, measurement invariance analyses indicated that the CRSI‐SP functioned equivalently across gender and relationship status, supporting its use for comparison across these groups.

The current study first demonstrated that the partner‐report subscales provide reliable and valid assessments of partners' perceptions of an individual's relationship behaviors. The partner‐report Care, Know, and Manage subscales demonstrated good psychometric properties and may be used individually as predictors or outcomes in future research. Because most couple relationship research relies primarily on individuals' self‐reports, incorporating partner‐reports may provide a more comprehensive assessment of relationship skills. Although Interdependence Theory (Kelley and Thibaut 1978) and prior empirical research (Homburg et al. 2012; Sanford ²⁰¹⁰) emphasize the value of incorporating others' perspectives when assessing relationship behaviors, partner‐reports remain underutilized in studies of couple dynamics. Together, these findings provide researchers and practitioners with additional options for incorporating partner‐reports into assessment of relationship skills through the CRSI‐SP.

The current study also demonstrated that combining individuals' self‐reports with partner‐reports of those same individuals resulted in a reliable and valid multi‐informant measure of relationship skills. Building on the original nine‐factor CRSI (Adler‐Baeder et al. ²⁰²²), the CRSI‐SP incorporated four additional partner‐report factors through 13 items. Specifically, partner‐report factors were added for Care, Know, and the two Manage dimensions (avoiding aggression and positive engagement), while maintaining good overall model fit. The moderate covariances among the self‐ and partner‐report constructs supported discriminant validity, indicating that the subscales were related yet empirically distinct, with little evidence of redundancy. Accordingly, researchers may use the individual self‐ and partner‐report subscales independently or combine them into an overall CRSI‐SP score, depending on the research question. Although stronger covariances might have been expected because the self‐ and partner‐reported subscales assess the same underlying constructs, the moderate covariances indicate that the two perspectives provide some distinct information. In post hoc analyses, models that combined self‐ and partner‐report items into the same latent factors demonstrated poor fit, further supporting the decision to model self‐ and partner‐report constructs separately. The results of these additional post hoc analyses are available from the first author upon request.

When considered together, the CRSI‐SP explains more variance in relationship quality than the original CRSI. Specifically, the CRSI‐SP explained slightly more variance in individual‐reported relationship quality and substantially more variance in couple‐level relationship quality (i.e., partner reports averaged) than the original self‐report CRSI, highlighting the value of incorporating partner‐reports into relationship assessment. Notably, the improvement in predictive utility was more pronounced for couple‐level relationship quality than for individual‐level relationship quality, suggesting that incorporating both partners' perspectives may be particularly valuable when evaluating couple‐level functioning that also considers both partners' perspectives. Future research should examine whether this improvement in predictive utility extends to other important relationship outcomes. Accordingly, when feasible, researchers and practitioners should consider using the 45‐item CRSI‐SP and collecting data from both partners to obtain a comprehensive assessment of relationship skills for research and program evaluation.

4.1. Directions for Measure Utilization

The flexibility of the CRSI‐SP allows researchers and practitioners to tailor its use to different research and evaluation objectives. Depending on the research question and completeness of the data, the CRSI‐SP may be scored as an overall measure or with individual subscales. Because the CRSI and CRSI‐SP were developed from a conceptual framework (Futris and Adler‐Baeder ²⁰¹³) to assess distinct relationship skill domains, examining individual subscales may provide useful insight into specific strengths and areas of growth. The psychometric findings further support treating the self‐ and partner‐report constructs as related but distinct dimensions, allowing researchers to examine how each perspective contributes to relationship functioning. Because the CRSI‐SP was developed using a classical test theory approach to measurement development (Algina and Penfield 2009), all items should be weighted equally when calculating scores. Finally, the racially and economically diverse sample and the measurement invariance findings support the use of the CRSI‐SP with economically and racially diverse cisgender men and women across different relationship statuses.

4.2. Limitations

The original CRSI, and subsequently the CRSI‐SP, were developed within the context of an applied research project rather than as a study designed specifically for measure development; therefore, more traditional measurement development methods were used (described in Adler‐Baeder et al. ²⁰²²). More recent efforts have focused on newer approaches to measurement design, including item response theory (IRT) and big data analytics. Future research employing these newer approaches may further refine and strengthen both the CRSI and the CRSI‐SP. Another limitation is the potential for common method biases. Although a strength of the current study is the inclusion of partner reports, the primary outcome of relationship quality still relies on self‐reported data from both partners. Thus, shared method variance may have inflated the observed associations between the CRSI‐SP and relationship quality. Future research could consider observational or behavioral data for outcome measures to reduce reliance on self‐report data and remove some error associated with common method biases.

Additionally, the stability of the factor structure over time was not examined, and assessing shifts in partners' perceptions of relationship behaviors after therapy or psychoeducational efforts seems to be an important next step. Also, discriminant validity was evaluated using standardized covariances rather than nested model comparisons; therefore, more formal model comparison statistics (e.g., Δχ ²) were not available. This limits more robust statements related to the potency of the CRSI‐SP's predictability above and beyond self‐report alone. Although measurement invariance was supported across gender and relationship status, future research could evaluate invariance across additional demographic characteristics (e.g., race/ethnicity, socioeconomic status, and sexual identity) as more diverse samples become available. Finally, although this study improved upon prior validation efforts by including a racially and economically diverse sample, participants were recruited from the southeastern United States and primarily consisted of heterosexual parents. Future research should examine the CRSI‐SP in geographically, culturally, and demographically diverse populations, including sexual and gender minority couples, to further evaluate its generalizability.

4.3. Next Steps

This study represents an initial step in evaluating the use of partner‐reports to assess a focal individual's relationship skills; however, additional work is needed to further evaluate and expand approaches for measuring and utilizing multi‐informant data from self‐ and partner‐reports (Busby and Gardner 2008). In the current study, we utilized an inter‐rater reliability approach by combining self‐reports from the focal individual with partner‐reports of that same individual. Future research could compare this approach with the insider's perspective model, which combines an individual's reports of themselves and their partner to capture perceptions of relationship functioning. Future studies could also evaluate the couple‐level model (Snyder 1997), which combines both partners' self‐ and partner‐reports into a single couple‐level construct. Another promising direction is to examine congruence between partners' reports to determine whether agreement or disagreement regarding relationship skills has implications for relationship functioning. Finally, although the CRSI‐SP demonstrated good reliability and validity across the demographic groups examined in this study, future research should evaluate whether its measurement properties and predictive utility differ across additional populations (e.g., socioeconomic status, relationship length, race/ethnicity, sexual identity).

4.4. Implications and Applications

Because healthy, stable relationships are strong predictors of optimal health and well‐being (Pinker 2015), accurately assessing relationship skills is essential for both research and practice. The current findings indicate that using both self‐ and partner‐reports through the CRSI‐SP provides modestly improved prediction of relationship quality compared with relying on self‐reports alone. Most existing assessments rely on a single informant or focus on a limited set of relationship behaviors. By assessing multiple relationship skill domains from both partner perspectives, the CRSI‐SP addresses both of these limitations. These findings may be particularly relevant in therapy and relationship education programs, where discussing similarities and differences in partners' perceptions of relationship skills may facilitate communication and insight (e.g., how a partner prefers to show and receive care). The CRSI‐SP may also provide a useful framework for structured discussions about relationship strengths and areas of growth. Overall, these findings support the CRSI‐SP as a reliable and valid multidimensional, multi‐informant measure of relationship skills. More importantly, the CRSI‐SP may advance applied and clinical evaluation efforts, which often rely primarily on individuals' experiences in couple relationship education or therapy, by incorporating partners' perspectives to provide a more holistic, dyadic understanding of relationship functioning and change.

Data Availability Statement

Research data are not shared.