Do Mindful Partners Make a Difference? A Meta-Analytic Test of the Actor-Partner Interdependence Model in Romantic Couples

Abstract

Objectives:

Mindfulness is consistently linked with individual well-being and relationship functioning, yet it remains unclear systemically whether mindfulness reliably “crosses over” within romantic couples — such that one partner’s mindfulness is associated with other partner’s outcomes after accounting for dyadic interdependence. We conducted a model-testing meta-analysis (MTMA) based on the Actor–Partner Interdependence Model (APIM) to estimate actor (self-self), partner (self-partner), and interdependence (partner similarity) effects of mindfulness on couples’ relationship outcomes.

Methods:

We searched six databases (PsycINFO, PubMed, CINAHL, Scopus, Web of Science, Cochrane Trials) for quantitative dyadic studies linking one partner’s mindfulness to the other partner’s outcomes across relational domains. Thirty-seven independent studies (12,396 couples) met inclusion criteria. We conducted a multivariate meta-analysis to estimate APIM-implied correlations (r*) and derived path coefficients, with 95% CIs from 1,000 bootstrap samples.

Results:

Cross-partner correlations were small but significant (rs* = .16 and .16 for male-to-female and female-to-male associations, respectively). However, corresponding APIM partner effects were close to zero and nonsignificant (βs = −.01), indicating no reliable crossover. Actor effects were significant (βs =.22 and .21 for male and female partners, respectively). Interdependence effects were also significant for partners’ mindfulness (r* = .27) and outcomes (r* = .42), indicating meaningful within-couple covariation.

Discussion:

Findings suggest that mindfulness is reliably associated with individuals’ own relationship functioning and shared dyadic processes, but not with a robust direct partner effect. These results indicated that partner associations may reflect shared couple-level influences rather than direct crossover, highlighting the implications of dyadic modeling and couple-level interventions.

Mindfulness, rooted in Buddhist traditions, has gained widespread attention across the globe over the past few decades (Cramer et al., 2016; Levey & Levey, 2018; Van Dam et al., 2018). With the rise of contemplative science, diverse conceptualizations of mindfulness have emerged (Alvear et al., 2022; Bishop et al., 2004; Van Dam et al., 2018). For instance, mindfulness has been conceptualized as a multidimensional construct comprising several related facets (e.g., present-moment attention, nonjudgmental acceptance, acting with awareness, and/or nonreactivity), depending on the measure and theoretical perspective (Baer et al., 2006; Blanke & Brose, 2017; MacKillop & Anderson, 2007). In the current study, we adopt a widely cited contemporary scholarly definition of mindfulness as “the awareness that emerges through paying attention on purpose, in the present moment, and nonjudgmentally to the unfolding of experience moment by moment” (Kabat-Zinn, 2003, p. 145), and conceptualize mindfulness as encompassing multiple interrelated dimensions that may be operationalized differently across studies.

Does One Person’s Mindfulness Impact Others?

As mindfulness has become increasingly integrated into psychological research and practice, its benefits have been examined not only at the individual level but also within interpersonal contexts. Empirical research has consistently demonstrated the intrapersonal benefits of mindfulness, including improvements in emotional regulation, stress reduction, and overall well-being across clinical and non-clinical populations (Galante et al., 2021; Goldberg et al., 2018, 2021; Goyal et al., 2014). These findings provide a basis for understanding mindfulness as a personal resource that promotes psychological health. However, as mindfulness research continues to evolve, scholars have increasingly turned their attention to its interpersonal implications, particularly within romantic and couples’ relationships. Couples’ relationships are an important intervention target as such relationships function not only as a subsystem within the broader family system that transmits relational patterns across generations, but also as a foundation for fostering other key interpersonal bonds, such as parent–child relationships (Chang & Barrett, 2009).

Recent studies suggest that mindfulness may have reciprocal benefits, or crossover effects in dyadic contexts, where one partner’s mindfulness influences the other’s emotional and relational outcomes (e.g., Birnie et al., 2010; Montes-Maroto et al., 2018; Zheng et al., 2021). These crossover effects likely emerge through the underlying intrapersonal and interpersonal mechanisms by which mindfulness shapes relational processes. Specifically, Karremans et al. (2017) proposed a theoretical model outlining how dispositional mindfulness may support romantic relationships through three key domains: (a) pro-relationship motivation and behavior, such as prioritizing long-term relational goals over immediate self-interest; (b) coping with distress, including responses to conflict and external stressors; and (c) relationship cognition, or the beliefs and perceptions individuals hold about their partner and the relationship (Karremans et al., 2017). These domains provide a useful framework for understanding how mindfulness-related processes within individuals may extend to shape their partners’ experiences and overall relationship dynamics.

While Karremans et al. (2017) provided a compelling theoretical account of how mindfulness could shape relational processes, empirical results are mixed. Some studies have found that individuals’ mindfulness is linked to their partners’ greater relationship satisfaction (e.g., Kappen et al., 2018; Karremans et al., 2020), whereas others report only actor effects, with no evidence of crossover effects in partners’ relationship satisfaction (e.g., Rimsky, 2020). Similar patterns appear in sexual well-being: daily sexual mindfulness has been associated with better same-day sexual outcomes for both members of the couple (Goldberg et al., 2025), yet other daily designs report benefits limited to the mindful individual than their partner (Jarvis et al., 2025). Variation also emerges by mindfulness type: interpersonal (i.e., relationship-specific) mindfulness shows more consistent crossover benefits that extend to both partners than general mindfulness in everyday life (Morin et al., 2024; Stanton et al., 2021). Collectively, these findings indicate that crossover varies by couples’ functioning domains and how mindfulness is operationalized (e.g., state, trait, interpersonal).

In light of these mixed results, no peer-reviewed meta-analysis has synthesized the crossover effects of mindfulness in couple dyads. A recent dissertation by Lam (2025) addressed this gap by meta-analyzing cross-partner correlations between one partner’s mindfulness and the other partner’s relational or psychological outcomes. They reported statistically significant but modest associations (rs =.17–.24) that were comparable across male–female partners, suggesting that mindfulness may operate as a shared dyadic resource. Using correlation coefficients provided a consistent metric across studies that had otherwise reported diverse effect sizes due to different analytic methods (e.g., correlations, regression coefficients, path coefficients). Yet, given that these estimates are correlations, they remain vulnerable to confounding and may overestimate the true magnitude of crossover effects among the dyads. These limitations highlight the need for a model-testing meta-analytic (MTMA; Becker & Aloe, 2019) approach that can disentangle each partner’s unique contribution among the dyad and provide a clearer estimate of how one partner’s mindfulness relates to the other partner’s outcomes. In other words, we assessed whether one’s mindfulness predicts the other partner’s outcomes beyond what the couple shares in common and beyond the individual’s own mindfulness.

Actor-Partner Interdependence Model

One statistical framework well suited for this purpose is the Actor–Partner Interdependence Model (APIM; Kenny et al., 2020). APIM is widely used in relationship science because it directly addresses nonindependence between partners and, importantly, separates the association between a person’s predictor and their own outcome (actor effect) from the association between that predictor and their partner’s outcome (partner effect) within the same model (Cook & Kenny, 2005). This structure makes APIM especially appropriate when the substantive question is whether one partner’s mindfulness predicts the other partner’s outcomes over and above shared couple-level covariance and each individual’s own mindfulness-linked functioning. Reflecting its broad utility, this model has been applied across diverse dyadic contexts, including romantic couples (e.g., Campbell & Kashy, 2002), family relationships (e.g., Afifi et al., 2006; Fales et al., 2014), and organizational dyads (e.g., Benlian & Haffke, 2016), and more recently in studies of mindfulness in romantic relationships (e.g., Harvey et al., 2019). Substantively, APIM allows researchers to test whether one partner’s mindfulness is associated with both their own relationship functioning (actor pathway) and their partner’s functioning (partner pathway), such as marital quality (e.g., Zamir et al., 2017).

As illustrated in Figure 1, simple cross-partner correlations (r_XY′, r_X′Y) are biased indicators of the partner path (p, p′). By the tracing rule, r_XY′ aggregates the direct partner path (p) plus indirect pathways via similarity on the predictor (r_XX′) and covariance between partners’ outcomes (r_YY′). When these components are positive, the raw correlation can overstate the true partner effect. APIM yields partner estimates that adjust for these dependencies, providing a clearer test of whether one partner’s mindfulness is linked to the other’s well-being. This approach allows researchers to produce more accurate estimates of whether mindfulness is linked to partners’ well-being by providing unconfounded estimates of partner influences while adjusting for actor effects.

Figure 1. The Actor–partner Interdependence Model (APIM).

Note. X and X′ denote mindfulness scores for male and female partners, respectively; Y and Y′ denote their perceived outcomes. Curved double-headed arrows indicate bivariate correlations (covariances) between partners on the predictor (r_XX′) and on the outcome (r_YY′). Single-headed arrows indicate hypothesized directional effects (partial regression coefficients): a and a′ are actor effects, and p and p′ are partner effects. Figure adapted from Cook & Kenny (2005, p. 2).

In the current study, because most existing mindfulness research on romantic relationships has focused on heterosexual male–female couples (e.g., McGill et al., 2020; Jaurequi et al., 2023), the studies meeting our eligibility criteria were mostly composed of male–female dyads. Accordingly, in our APIM framework, we treated partners as distinguishable by gender (male vs. female), which allows estimation of separate actor and partner effects for male (a, p) and female (a′, p′) partners and enables tests of gender as a moderator of actor and partner effects.

The Current Study

Despite expanding research on the crossover effects of mindfulness, it remains unclear whether reliable partner effects exist in couple relationships. Individual studies show mixed findings (e.g., Pakenham & Samios, 2013; Roberts et al., 2021; Zhou, 2022), and analytic approaches vary from simple cross-partner correlations to dyadic models, limiting comparability. To address these limitations, we conducted a model-testing meta-analysis using the actor–partner interdependence model (APIM) to obtain pooled estimates of both actor and partner effects of mindfulness on relationship outcomes in couples.

The primary objectives of the current study were to: (a) synthesize quantitative evidence on the association between actors’ mindfulness and partners’ relational outcomes; (b) estimate the magnitude and direction of actor, partner, and interdependence effects using a model-testing meta-analysis based on the Actor–Partner Interdependence Model (APIM); (c) examine whether these effects varied across mindfulness types, outcome domains, and sample characteristics. Central to these aims, we evaluated whether actors’ mindfulness uniquely predicts partners’ outcomes beyond shared couple-level variance and beyond individuals’ own mindfulness. Consistent with our pre-registration, our primary hypothesis was that actors’ mindfulness would be positively associated with partners’ relational outcomes, even after accounting for actor effects and dyadic interdependence. In addition, based on the broader mindfulness literature, we expected actor effects to be positive, such that individuals’ mindfulness would be associated with their own relational outcomes. Moderator analyses were exploratory and examined whether the magnitude of actor, partner, and interdependence effects varied across mindfulness types, outcome domains, and sample characteristics.

In this study, mindfulness is treated as the independent variable and conceptualized across three dimensions: trait mindfulness, state mindfulness, and interpersonal mindfulness. The dependent variables encompass relational outcomes theorized to be influenced by mindfulness (e.g., Karreman et al., 2017), including pro-relationship motivation and behavior, coping with distress, relationship cognition, relationship well-being, and sexual well-being. These domains reflect key processes identified in theory and empirical work as central to couple functioning and provide a comprehensive framework for evaluating the relational impact of mindfulness (e.g., Karreman et al., 2017, 2020; Goldberg et al., 2025).

Finally, given that the strength of partner (crossover) effects may vary across mindfulness type (e.g., Barnes et al., 2007), outcome domain (e.g., Hadden et al., 2014; Karremans et al., 2017), and sociocultural and couple-contextual characteristics (e.g., relationship duration, cultural orientation, gender differences in relational processes; Lam, 2025; Christensen et al., 2006), this study examines exploratory moderators that may these associations. Accordingly, we examined moderators including mindfulness type (i.e., trait, state, interpersonal), outcome domains (i.e., pro-relationship motivation and behavior, coping with distress, relationship cognition, relationship well-being, and sexual well-being), cultural context (independence- vs. interdependence-oriented), mean age, relationship duration, percent racial/ethnic minority, percent married, education, and gender differences. By integrating moderator analyses with a meta-analytic APIM approach, this study aims to clarify not only whether mindfulness “crosses over” to affect partners on a dyadic level, but also under what conditions such effects are more likely to occur. Taken together, these analyses contribute to a more nuanced understanding of mindfulness as both an individual and relational resource, with implications for its role in intimate relationship dynamics and its potential as a couple-level intervention target.

Methods

Protocol and Registration

This study was preregistered through the Open Science Framework (https://osf.io/4ya6m; Lam et al., 2023)². It was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021). We made the following five deviations from our preregistered plan. First, we estimated effects using a meta-analytic APIM framework rather than simple Pearson correlations to model both actor and partner effects while appropriately accounting for dyadic interdependence. Second, to ensure comparability of the mindfulness exposure across studies, we extracted baseline (pre-intervention) mindfulness scores from eligible intervention trials. Third, intervention studies were not included as tests of intervention effects because they address a distinct research question regarding the effects of mindfulness training on partners’ outcomes, whereas the present meta-analysis examines whether one partner’s mindfulness predicts the other partner’s outcomes beyond shared couple-level variance and beyond the individual’s own mindfulness within an APIM framework. Nevertheless, intervention studies that provided baseline mindfulness scores were included to derive pre-intervention actor–partner associations, which are methodologically comparable to the cross-sectional effect sizes synthesized in the present APIM-based meta-analysis. Fourth, we tested the moderator effects of categorical moderators (e.g., gender) by bootstrapping the comparison of coefficient values for separate models run on effect sizes at each level of the moderator variable.

Eligibility Criteria

Studies were included if they met the following conditions: (a) used a quantitative design; (b) included romantic couples (i.e., dyads) with distinguishable partners; and (c) reported statistical associations linking one partner’s (actor’s) mindfulness to the other partner’s outcomes. Eligible operationalizations of the actor’s mindfulness comprised (a) trait mindfulness, (b) state mindfulness, and (c) interpersonal mindfulness. Studies were excluded if they did not include romantic couples, reported only partners’ outcomes or only actors’ mindfulness, or if mindfulness and outcome measures were drawn from the same individual within the dyad, as such designs do not permit evaluation of mindfulness crossover effects. Literature reviews and meta-analyses were excluded. No limitations were placed on publication status and language. Studies were synthesized based on the availability of APIM-relevant effect sizes and the comparability of dyadic role structure across samples. Analyses focused on heterosexual couples (i.e., male–female romantic partners) to ensure consistent distinguishability and comparability for APIM analyses. Other romantic dyads defined by role-based configurations (e.g., patient–spouse or spouse–caregiver dyads), in which partners are differentiated by functional roles rather than gender, were identified but not analyzed as separate subgroups because these role structures are not directly comparable to male–female dyads and because there were too few studies (k < 4) to support a meaningful meta-analytic synthesis.

Information Sources

We conducted a comprehensive search across multiple databases, including the Cochrane clinical trials registry, PsycINFO, PubMed, CINAHL, Scopus, and Web of Science. To reduce potential publication bias and increase the likelihood of capturing null results, we also screened dissertations and records of unpublished clinical trials (i.e., likely “file-drawer” studies). All procedures followed PRISMA recommendations (Page et al., 2011) to maximize transparency and reproducibility of the review process. Our search was conducted from the first available date until February 6, 2025 (see “Protocol and Registration”). We also conducted backward and forward citation searches of the studies identified through the primary search on March 31,2026.

Search Strategy

As we were interested in examining the partner effects of mindfulness in romantic couples, we conducted database searches using the following search terms: (dyad* OR “actor-partner” OR interaction* OR interpersonal OR crossover OR “cross-over” OR interdependence OR “cross-informant” OR extrapersonal OR “cross-partner” OR “partner effect*”) AND (mindful* or meditate*) AND (couple* OR partner* OR “close relation*” OR spous* OR marital* OR mate* OR roman* OR dat*). The complete database-specific search strategies, including field tags, filters, limits, and dates searched are reported in Supplemental Table S1.

To supplement the database search, we conducted citation searching using Citation Chaser (Haddaway et al., 2022) to identify potentially eligible reports that may not have been captured through the primary database. Specifically, we performed both backward citation searching, in which we reviewed the reference lists of identified studies, and forward citation searching, in which we examined newer records that cited those studies (Hirt et al., 2024). Citation-search results were screened using the same eligibility criteria and study-selection procedures as records identified through database searches.

Study Selection

All studies were independently screened and coded by a coding team consisting of the lead author, two doctoral students, and one trained undergraduate assistant. Each title/abstract and full-text report was screened by at least two reviewers independently. Team members engaged in double-coding across studies to balance workload and enhance consistency in study selection and coding decisions. The lead author developed the codebook in collaboration with the senior author to guide the coding team’s initial training. The codebook was updated iteratively based on study review to guide the entire coding process. The coding team met regularly to address coding discrepancies until consensus was reached. When consensus could not be reached through initial discussion, the lead author and senior author were consulted to make the final eligibility decision. Screening decisions were managed using Microsoft Excel to track inclusion/exclusion decisions and resolve discrepancies across reviewers. No automation tools were used to make inclusion or exclusion decisions. Inter-rater agreement based on independent screening decisions was substantial for both title and/or abstract (Cohen’s κ = .71; Fleiss, 2013) and full-text screening (Cohen’s κ = .73). Agreement for categorical moderators was similarly substantial (Cohen’s κ ≥ .75).

During full-text review, we made three clarifications to the inclusion/exclusion rules: (1) studies that presented only a mindfulness subscale embedded within self-compassion measures were excluded as lacking an actor mindfulness construct; (2) studies reporting only indistinguishable dyads (e.g., same-sex couples without role differentiation, or “close other” caregivers in couple samples) were excluded because partner effects could not be estimated; and (3) distinguishable dyad categories represented by too few studies for analysis (e.g., employee–spouse; childbearing parent–spouse) were excluded from quantitative synthesis. Full-text reports that appeared potentially eligible but were excluded are summarized with reasons for exclusion in the PRISMA flow diagram (Figure 2).

Figure 2. PRISMA Flow Diagram.

Note. PRISMA flow diagram illustrating the identification, screening, and inclusion of studies, in accordance with Page et al. (2021).

Data Collection and Data Items

Data was independently coded by the coding team, which met on a regular basis to address coding discrepancies and reached consensus. Each included study was coded by at least two reviewers, with disagreements resolved through discussion. Extracted study-level variables included sample size, relationship characteristics (e.g., relationship duration, marital status), participant demographics (e.g., male and female partners’ age and education), and country of origin. Cultural context was derived from country-level information based on established regional or cultural classifications. To account for variability in the conceptualization of mindfulness across studies, mindfulness variables were coded based on the constructs assessed by study-specific measures and categorized into trait mindfulness, state mindfulness, and interpersonal mindfulness. When available, facet-level indicators (e.g., present-moment attention, nonjudgmental acceptance, acting with awareness, and nonreactivity) were also extracted.

Outcome variables were coded based on the constructs assessed in each study and subsequently grouped into broader theoretically derived domains to ensure comparability in the meta-analytic synthesis. These variables encompassed relational outcomes theorized to be influenced by mindfulness, including pro-relationship motivation and behavior, coping with distress, relationship cognition, relationship well-being, and sexual well-being (e.g., Karremans et al., 2017, 2020; Goldberg et al., 2025). These domains reflect key processes identified in theory and empirical work as central to couple functioning and were used to organize outcomes across studies for synthesis.

Extracted effect-size information included cross-partner correlations, actor correlations, partners’ mindfulness interdependence, partners’ outcome interdependence, sample sizes, and any information needed to compute or verify Fisher’s z-transformed correlations. When multiple eligible mindfulness indicators or outcome measures were reported within a study, all eligible effects were extracted and then aggregated within study before synthesis to reduce dependence among effect sizes.

The inter-rater reliabilities for data coding were excellent (i.e., ICC = .99; Cicchetti, 1994). Authors from eligible studies that did not report the required baseline correlation metrics data were contacted by the lead author during the coding phase. Datasets from eight out of 37 included studies were received from authors (Khaddouma et al., 2018; Laurent et al., 2013, 2016a, 2016b; Lenger et al., 2017; Li et al., 2022; McGill et al., 2020; Schellekens et al., 2017). In cases where the necessary APIM correlations were not available from published materials or author correspondence, the associated effect was coded as missing and omitted from the relevant synthesis; nonetheless, the study was retained in the overall review. Assumptions made when coding unclear study characteristics or effect-size information were documented in the coding file and resolved through consensus.

Effect Measures and Data Preparation

For each eligible study, we recorded as many of the six correlations implied by the APIM (Figure 1) as were reported by the authors. In this notation, X denotes mindfulness and Y denotes the outcome; unprimed terms refer to the male partner and primed terms (′) refer to the female partner. Specifically, each study contributed estimates of some or all of the following correlations: cross-partner correlations (r_XY′; r_X′Y), actor correlations (r_XY; r_X′Y′), partners’ mindfulness interdependence (r_XX′), and partners’ outcome interdependence (r_YY′). All effect sizes were coded so that positive values reflected higher actor mindfulness associated with more favorable partner outcomes. When necessary, effect directions were reverse-coded to maintain interpretive consistency. Analyses were conducted in R (R Core Team, 2025) using the MAd (Del Re & Hoyt, 2022) and metafor packages (Viechtbauer, 2010). Institutionally licensed ChatGPT (OpenAI, 2026) was used to refine R code for formatting and labeling figure outputs (e.g., PRISMA flow diagram), with all outputs reviewed and verified by the authors. ChatGPT was not used to make eligibility decisions, extract data, estimate effect sizes, or interpret results.

Synthesis Methods

All correlation coefficients were transformed to Fisher’s z′ prior to analysis (Hedges & Olkin, 1985). Because many studies reported multiple mindfulness measures, multiple outcomes, or both, dependent effect sizes within studies were aggregated using procedures recommended by Borenstein et al. (2009). Studies were included in a given synthesis if they provided sufficient APIM-relevant correlation information to estimate one or more of the six pooled correlations required for the meta-analytic APIM framework. We then followed Becker’s (2009) procedures to conduct a multivariate meta-analysis, simultaneously deriving meta-analytic estimates of the six pooled z′ coefficients. We converted these z′ estimates to r and computed meta-analytic estimates of the APIM path coefficients based on the resulting correlation matrix. Confidence intervals (CIs) for all effect estimates were computed via percentile bootstrap with 1,000 resamples. We quantified heterogeneity using I², representing the proportion of total variability attributable to between-study differences (Higgins et al., 2003).

Moderator effects were evaluated in two steps. First, we tested univariate moderator effects for each of the six APIM-implied correlations using meta-regression models. Second, for categorical moderators, we estimated separate APIM path models for each category and used bootstrapped confidence intervals to evaluate between-category differences in path coefficients. Specifically, we coded effect-size level moderators (mindfulness type and outcome domain) and one categorical study-level moderator (cultural context). Gender was examined as a moderator of actor effects by bootstrapped contrasts between a and a′ (actor) and p and p′ (partner) coefficients. For continuous moderators, no straightforward comparison of path models was available, so we report only the univariate moderator tests.

Publication bias and sensitivity analyses

We assessed whether publication bias (e.g., selective reporting) influenced the pooled partner-effect estimates. We inspected contour-enhanced funnel plots to visualize asymmetry and to locate any potentially “missing” studies relative to regions of statistical significance using Duval and Tweedie’s trim-and-fill procedure as a descriptive aid (Peters et al., 2008; Duval & Tweedie, 2000). Given the substantial heterogeneity in the current meta-analysis (I² > 90%), trim-and-fill results were interpreted cautiously because the method can spuriously adjust in heterogeneous datasets (Terrin et al., 2003).

In addition, sensitivity analyses were conducted to evaluate the robustness of primary findings to influential outliers. Specifically, primary models were re-estimated with and without effect sizes flagged during outlier screening. To identify potential outliers, we screened for values exceeding 2 standard deviations (SDs) from the mean of the sampling distribution (Hedges & Olkin, 1985) and visually examined contour-enhanced funnel plots for each of the six APIM correlations. Outlier screening and removal were conducted separately for each meta-analytic model corresponding to the six APIM-implied correlations, with between one and two outliers identified per model (k ranging from 29 to 35 studies per model). Sensitivity check indicated only trivial differences in pooled estimates (Δr ranging from −.03 to .00), indicating that the results were not substantially influenced by these values.

Reporting Quality Assessment

To evaluate the quality and transparency of reporting in the included studies, we conducted a reporting quality assessment using checklist items adapted from the STROBE recommendations for observational studies (Melinte et al., 2023). The assessment focused on reporting of study design, participant selection, variable definitions and measurement, missing data, statistical methods, outcome reporting, and interpretation. Only applicable items were rated for each study. For each study, we calculated the number and percentage of applicable items that were fully reported. Two reviewers independently completed the assessment, and discrepancies were resolved through discussion and consensus. Because STROBE is a reporting guideline rather than a formal risk-of-bias instrument, we interpreted these ratings as indicators of reporting completeness and transparency rather than as direct estimates of study validity or bias. Study-level assessment results are provided in the OSF supplemental materials.

Availability of Data

All supplemental materials, including search results, study characteristics, STROBE-informed quality assessments, effect-size level data, are available on the OSF at https://osf.io/hp8gv/ (Lam et al., 2026).

Results

Study Selection

The database search yielded a total of 10,078 citations. We removed a total of 3,757 duplicates and evaluated the remaining 6,321 studies for titles and/or abstracts based on inclusion/exclusion criteria. The screening resulted in 246 studies for full-text review and 88 studies for effect size coding. To supplement the database search, we also conducted backward and forward citation searches of the identified studies. This citation search yielded 2,528 additional records (1,422 from backward searching and 1,106 from forward searching). After removal of 453 duplicates, 2,075 records were screened by title and abstract, 169 full-text reports were assessed for eligibility, and 3 additional reports proceeded to effect size coding. Across all search procedures, 12,606 records were identified, 4,210 duplicates were removed, and 8,396 records were screened. In total, 415 full-text reports were reviewed, and 91 reports were coded for effect sizes. The final dataset comprised 37 unique studies representing 12,396 couples (Figure 2).

Study Characteristics

Characteristics of the 37 included studies are available as Supplemental Table S2. Most studies were conducted in the United States (62.16%; k = 23), followed by China (13.51%; k = 5), and Canada (8.11%; k = 3). Fewer studies were conducted in Hong Kong (5.41%; k = 2), Turkey (5.41%; k = 2), Croatia (2.70%; k = 1), and Spain (2.70%; k = 1). Most studies were conducted in relatively independence-oriented cultural contexts (70.27%; k = 26; United States and Canada), whereas 29.73% (k = 11) were conducted in relatively more interdependence-oriented contexts (China, Hong Kong, Croatia, Turkey, and Spain).

Across studies, male partners were on average 38.77 years old (SD = 9.84), and female partners were on average 36.36 years old (SD = 10.21). The average dyadic age was 38.74 years (SD = 12.30). On average, 81.96% of studies included married couples, and all couples were identified as heterosexual. The mean relationship duration across studies was 16.73 years (SD = 21.08). Overall, 63.53% of male partners and 65.60% of female partners reported having attended at least some college. Racial/ethnic minority status was defined relative to the majority population in the country where the primary study was conducted. Dyad-level racial/ethnic data were reported in 15 studies, all U.S.-based, with an average of 35.39% of couples identified as racial/ethnic minorities. Nine studies reported racial/ethnic data for male partners and 10 for female partners, all conducted in the United States; across these studies, 46.61% of male partners and 51.71% of female partners identified as racial or ethnic minorities. Twelve studies (32.43%) included dyads with children or in caregiving roles. Child-related characteristics varied substantially, including newborns and infants, preschool- and kindergarten-aged children, and adolescents, as well as diverse family contexts such as coparenting, stepfamily, foster care, and military-connected samples. Details are provided in the study-level data available in the OSF supplemental materials.

Reporting Quality of Included Studies

Reporting quality was assessed using applicable items adapted from the STROBE recommendations for observational studies. Across the 37 included studies, the number of applicable items ranged from 25 to 30, reflecting minor differences in study design and reporting applicability. The proportion of fully reported applicable items ranged from 63.0% to 85.7% across studies, with a mean of 73.7% and a median of 74.1%. Overall, these findings suggest moderate to high reporting completeness among the included studies, although no study fully reported all applicable items and reporting transparency varied across studies.

Results of Individual Studies

The aggregated correlation matrices for individual studies are reported in Supplemental Table S3. These matrices summarize the within-study associations among partners’ mindfulness and outcomes by reporting the aggregated APIM correlations available from each study after combining dependent effect sizes, including cross-partner (r_XY′*, r_X′Y*), within-partner (r_XY*, r_X′Y′*), and interdependence correlations (r_XX′*, r_YY′*) that were used to derive the meta-analytic APIM path coefficients in the MTMA models.

Synthesis of Results

MTMA pooled effect size estimates of the main APIM model are displayed in Figure 3. The pooled correlation coefficients from Becker’s multivariate meta-analysis are reported in Supplemental Table S4. Confidence intervals reflect percentile bootstrap estimates based on 1,000 resamples.

Figure 3. Meta-analytic Actor–Partner Interdependence Model (APIM).

Note. APIM summarizes meta-analytic estimates from the model-testing meta-analysis. Paths are denoted as partner effects (β_XY′, β_X′Y) and actor effects (β_XY, β_X′Y′); between-partner covariation (interdependence effects) is shown as r_XX′* (predictors) and r_YY′* (outcomes). Estimates are pooled effects; *p < 05; **p <.01; ***p<.001.

Partner Effects

Bivariate cross-partner correlations in mindfulness between male and female partners were small but statistically significant (r_XY′* = .16 [.12, .20]; r_X′Y* = 16 [.12, .21]), indicating modest associations between one partner’s mindfulness and the other partner’s relationship outcomes (Supplemental Table S4). However, after accounting for dyadic interdependence using the APIM framework, the corresponding partner effects were near zero (β_XY′ = −.01 [−.04, .02]; β_X′Y = −.01 [−.03, .02]). These estimates suggest that one partner’s mindfulness was not reliably associated with the other partner’s outcome once actor and interdependence effects were controlled. Thus, despite significant bivariate correlations, there was no evidence of systematic crossover effects of mindfulness between partners. Heterogeneity was high for both cross-partner correlations (I² = 91.23% for r_XY′; I² = 92.05% for r_X′Y), indicating substantial between-study variability beyond sampling error and suggesting that moderator analyses are warranted.

Actor Effects

Both actor effects were statistically significant (β_XY = .22 [.17, .28]; β_X′Y′ = .21[.16, .26]). These results indicate that individuals who reported higher mindfulness tended to report more favorable outcomes themselves, independent of their partner’s mindfulness. Heterogeneity in our meta-analysis of the corresponding actor correlations was high (I² = 94.86% and 95.99%, respectively).

Interdependence Effects

Correlations between partners’ mindfulness and outcomes (r_XX′* = .27, [.22, .34]; r_YY′* = .42, [.36, .49]) confirmed moderate-to-high interdependence within dyads (Figure 3). These values reflect the shared context and mutual influence inherent in romantic relationships, even when specific actor and partner pathways are controlled. The overall pattern highlights that partners’ outcomes remain statistically dependent, consistent with theoretical expectations of dyadic data structures, but that this interdependence is not explained by direct mindfulness crossover. Heterogeneity for both models was high (I² = 97.05% for r_XX′ and 96.24% for r_YY′, respectively).

Reporting Quality of Included Studies

Reporting quality was assessed using applicable items adapted from the STROBE recommendations for observational studies. Across the 37 included studies, the number of applicable items ranged from 25 to 30, reflecting minor differences in study design and reporting applicability. The proportion of fully reported applicable items ranged from 63.0% to 85.7% across studies, with a mean of 73.7% and a median of 74.1%. Overall, these findings suggest moderate to high reporting completeness among the included studies, although no study fully reported all applicable items and reporting transparency varied across studies.

Reporting Bias Assessment

Trim-and-fill analyses indicated little evidence of small-study or publication bias across the aggregated models. For the partner effects, the contour-enhanced funnel plots with trim-and-fill imputations showed only trivial asymmetry (Figure 4). For the male-to-female partner effect (r_XY′), no studies were imputed to restore symmetry, and the funnel plot appeared broadly symmetrical. The female-to-male crossover association (r_X′Y) showed slight asymmetry, with two studies imputed to restore symmetry but only a negligible change in the pooled estimate (observed r= .19; adjusted r = .20; Δr = .01), arguing against meaningful bias.

Figure 4. Counter-enhanced Funnel Plots with Trim-and-Fill Imputations for Partner (Crossover) Effects.

Note. Contour-enhanced funnel plots with trim-and-fill imputations for the (left) male-to-female (r_XY′) and (right) female-to-male (r_X′Y) partner effects. Filled circles represent observed studies; open circles represent imputed studies added by the Duval and Tweedie (2000) trim-and-fill procedure. Shaded regions indicate conventional p-value contours (p < .10, .05, .01). Both plots show symmetrical distributions with only two imputed studies each, suggesting negligible publication bias.

For the actor effects, the male actor path (r_XY) showed a broadly symmetrical distribution around the meta-analytic mean, with no studies imputed, indicating little risk of publication bias. The female actor path (r_X′Y′) showed a similarly symmetrical distribution, with no meaningful evidence of asymmetry after trim-and-fill (Supplemental Figure S1).

For the interdependence effects, the dyadic mindfulness association (r_XX′) yielded two imputed studies, with no substantive shift in the pooled estimate (observed r= .33; adjusted r = .34; Δr = .02). The dyadic outcome association (r_YY′) showed a broadly symmetrical distribution with no studies imputed (Supplemental Figure S1).

Taken together, the contour-enhanced funnels (Peters et al., 2008) did not indicate a systematic absence of studies in regions of statistical non-significance, suggesting that any minor asymmetry was more likely due to sampling variability than to selective reporting. Across models, these findings indicate minimal evidence of reporting bias across the synthesized effects.

Additional Analyses

Results of all moderator analyses are reported with corresponding statistical estimates and confidence intervals in Supplemental Tables S5–S10.

Gender Moderator

Because gender varies within studies (male partners vs. female partners) rather than between studies, we did not conduct univariate moderator tests for gender. Instead, we tested gender differences directly within the APIM by comparing male versus female partner effects (Δp = β_XY′ − β_X′Y) and actor effects (Δa = β_XY − β_X′Y′) using bootstrapped contrasts. Neither contrast was statistically significant, Δp = −.005 [−.03, .02] and Δa = .007 [−.02, .03], indicating no evidence that actor or partner effects differed by gender.

Effect-size level Moderators

We tested whether mindfulness type (i.e., relationship, state, trait) and outcome domain (i.e., pro-relationship motivation and behavior, coping with distress, relationship cognition, relationship well-being, and sexual well-being) moderated the pooled MTMA correlation estimates (separate univariate moderator analyses). We then estimated subgroup-specific APIM path coefficients (β) from these MTMA correlation estimates and evaluated between-subgroup differences using bootstrapped pairwise comparisons of subgroup MTMA estimates.

Mindfulness Type.

We first tested moderation at the level of the six APIM-implied correlations using univariate meta-regressions. Mindfulness type did not significantly moderate the cross-partner or actor correlations (ps >.10; see Supplemental Table S5), indicating no evidence that the magnitude of partner or actor effects varied across interpersonal, state, or trait mindfulness. In contrast, mindfulness type significantly moderated interdependence effects for partners’ mindfulness (r_XX′: Q = 13.86, p < .001) and outcomes (r_YY′: Q = 17.52, p < .001). Because differences in these interdependence effects can carry forward into the derived APIM estimates, we examined subgroup-specific APIM path coefficients.

We next conducted bootstrap subgroup MTMA analyses to examine whether these differences extended to APIM-derived path coefficients (see Supplemental Table S6). Across mindfulness types, partner effects remained small in magnitude (|β| < .10), despite some statistically detectable pairwise differences. Specifically, partner effects were more negative in the interpersonal-mindfulness subgroup relative to the state and trait mindfulness subgroups (e.g., interpersonal v.s. state: Δβ_XY′ = −.19 [−.32, −.05]); Δβ_X′Y = −.14 [−.27, −.04]; interpersonal v.s. trait: β_XY′ (Δ β_XY′ = −.17 [−.31, −.03]). In addition, interpersonal mindfulness was associated with stronger interdependence effects, reflected in higher r_XX′* and r_YY′* compared to state or trait measures (e.g., interpersonal vs. state: Δr_XX′* = .34 [.23, .50]; Δr_YY′* = .22 [.11, .38]). However, these subgroup differences should be interpreted cautiously given the small number of studies in the interpersonal and state subgroups (k = 5 and k = 4, respectively), and the consistently small magnitude of partner effects across all mindfulness types.

Outcome Domain.

Outcome domain showed evidence of moderation for partners’ outcome interdependence, (r_YY': Q = 13.53, p = .009; see Supplemental Table S7). No evidence of moderation was observed for the cross-partner, actor, or mindfulness interdependence correlations (ps > .10). Relative to coping outcomes (reference category), relationship well-being outcomes were associated with greater outcome interdependence (r_YY': B = .31 [.12, .51]). Given that domain differences in r_YY' can carry forward into the derived APIM effects, we next examined outcome-domain differences in MTMA-derived APIM path coefficients using bootstrap subgroup contrasts.

Bootstrap subgroup MTMA by outcome domain (i.e., pro-relationship motivation and behavior, coping with distress, relationship cognition, relationship well-being, and sexual well-being) indicated small but statistically detectable pairwise differences in partner effects (Supplemental Table S8). Compared with relational well-being, pro-relationship motivation and behavior showed stronger partner effects (Δβ_XY′ = .08 [.02, .25]; Δβ_X′Y = .05 [.01, .16]). Nevertheless, partner effects remained small in magnitude across domains (β_XY′ = −.01; β_X′Y = −.01; domain-specific |β| < .10), indicating that outcome-domain differences did not alter the overall conclusion that partner effects are close to zero. Bootstrap resampling yielded 977 valid draws (of 1,000) due to instability in the sexual well-being subgroup (k = 3); results involving that subgroup should be interpreted with caution.

Study level Moderators

We first tested moderation at the level of the six APIM-implied correlations using univariate meta-regressions (see Supplemental Table S9). Cultural context (i.e., independence-vs. interdependence-oriented) significantly moderated the cross-partner correlations, with interdependent-oriented (vs. independent-oriented) samples showed larger univariate cross-partner correlations (Δr_XY′ = .19 [.06, .30]; Δr_X′Y = .15 [.01, .27]). No evidence of moderation was observed for the actor or interdependence correlations (ps > .10).

We next examined whether cultural context accounted for variability in APIM partner effects using bootstrap subgroup MTMA analysis (see Supplemental Table S10). In contrast to the univariate results, cultural-context differences in partner effects were small and nonsignificant (Δβ_XY′ = .04, [−.09, .17]; Δβ_X′Y = .00, [−.09, .10]). Similarly, cultural context did not reliably moderate actor effects (Δβ_XY = .02, [−.14, .15]; Δβ_X′Y′ = .01, [−.18, .15]).

Cultural context did, however, moderate interdependence effects, with interdependence-oriented couples showing larger partner similarity in both mindfulness (Δr_XX′* = .19, [.06, .30]) and outcomes (Δr_YY′* = .15, [.01, .27]). Thus, cultural context appears related to dyadic interdependence, but not to the APIM partner effects. Nevertheless, given the uneven distribution of studies across cultural contexts (independence-oriented at k = 25; interdependence-oriented at k = 11), these results should be interpreted cautiously.

For the remaining study-level moderators (sample mean age, relationship duration, percent racial/ethnic minority, percent married, education, and gender differences), univariate tests were largely nonsignificant across the six APIM-implied correlations (ps > .10; see Supplemental Table S9), with the exception of a few effects on interdependence effects. Specifically, mean age and racial/ethnic composition were found to moderate interdependence effects, with older samples showing greater partner similarity in both mindfulness (r_XX′: B₁ = .01, p = .03) and outcomes (r_YY′: B₁ = .01, p < .01), and samples with a higher percentage of racial/ethnic minority couples showing greater partner similarity in mindfulness (r_XX′: B₁ = .01, p = .02). Nevertheless, because these moderators did not significantly explain the cross-partner correlations, they provide limited evidence for meaningful moderation of the APIM partner effects.

Sensitivity analyses

For sensitivity analyses, we screened for potential outliers using contour-enhanced funnel plots for each of the six meta-analytic effects. Distributions were broadly symmetrical, and screening identified a small number of potential outliers (one to two per model). To evaluate their impact, we re-estimated all primary models with and without the flagged values. Results were substantively unchanged, with only trivial differences in pooled estimates across all effects (Δr ranging from −.03 to .00). For instance, the male actor correlation showed a negligible shift after exclusion (r_XY* = .27) relative to the full model (r_XY* = .27; Δr = −.003). All remaining actor, partner, and interdependence effects showed comparable stability, indicating that the findings were not substantially affected by outliers.

Discussion

Despite expanding research on the partner (i.e., crossover) effects of mindfulness in couple relationships, it remains unclear whether reliable partner effects can be detected in dyadic data. Individual studies have produced mixed findings (e.g., Pakenham & Samios, 2013; Roberts et al., 2021; Zhou, 2022), and conclusions are further complicated by wide variation in analytic strategies (ranging from cross-partner bivariate correlations to basic and extended dyadic models) that limit comparability across studies. A preliminary quantitative synthesis (Lam, 2025) pooled cross-partner correlations and found small-to-medium statistically significant associations, but these estimates do not provide a clear test of partner effects because bivariate crossover indices are confounded with actor effects and other shared dyadic influences. Building on this limitation, the current study extends prior work by applying a model-testing meta-analysis based on APIM to separate actor, partner, and interdependence effects, thereby providing a more stringent test of whether mindfulness exerts a unique partner (crossover) effect in couple relationships.

We meta-analyzed a sample of 37 studies (n = 12,396 couples) and found that across studies, mindfulness showed a consistent pattern of within-person benefits but little evidence of partner (crossover) effects. Contrary to our hypothesis that actors’ mindfulness would be positively associated with partners’ outcomes after accounting for actor effects and dyadic interdependence, the APIM results indicate that partner effects are essentially null. At the bivariate level, the cross-partner correlations within the dyads suggested a modest but statistically significant association between a partner’s mindfulness and the other partner’s relationship outcomes. However, when dyadic interdependence was modeled using the APIM (i.e., accounting for actor effects and interdependence effects of the couple dyads), the partner effects were essentially null and nonsignificant. This result mirrors prior meta-analytic evidence showing little support for crossover effects from one partner’s work-related variables to the other partner’s family well-being (Matei et al., 2021). One possible interpretation is that these initial associations largely reflect shared couple context rather than a direct crossover process. Because dyadic data are inherently nonindependent, APIM separates shared variance from individual effects (Kenny et al., 2020). After accounting for this shared variance, partner effects may diminish because they are driven by common routines, environments, or relationship climate rather than one partner’s mindfulness directly influencing the other’s outcomes. In this sense, mindfulness may covary within couples without exerting a unique direct effect on partners’ outcomes (see the discussion for interdependence effects). This finding refines earlier meta-analytic conclusion (Lam, 2025) by showing that apparent cross-partner correlations may be overestimated when dyadic interdependence is not explicitly modeled.

In contrast, significant actor effects for both partners suggested that greater individual mindfulness is associated with more favorable intrapersonal outcomes. This pattern aligns with a large body of research linking mindfulness to adaptive coping, emotional regulation, and relationship functioning (Keng et al., 2011; Malin, 2023; Voldstad et al., 2025), as well as prior dyadic studies reporting stronger actor effects despite nonsignificant partner effects (e.g., Jarvis et al., 2025; Stanton et al., 2021). Together, these findings suggest that mindfulness primarily operates as an individual regulatory capacity. Theoretically, mindfulness is thought to influence how individuals attend to and respond to relational experiences through processes such as attentional control, decentering, emotional awareness, and nonreactivity (Bishop et al., 2004; Brown & Ryan, 2003), which shape one’s own coping, responsiveness, and conflict regulation (Karremans et al., 2017), thereby enhancing relationship functioning at the individual level. However, these intrapersonal benefits of mindfulness may not reliably translate into partner outcomes unless they are translated into observable, relationship-relevant behaviors, such as constructive communication, responsiveness, emotion regulation, or reduced negativity during couple interactions (Barnes et al., 2007; Wachs & Cordova, 2007; Karremans et al., 2017). This distinction may help explain why actor effects are significant while direct partner effects are not consistently observed.

Consistent with the above interpretation, the moderate interdependence estimates indicate meaningful shared variance in couples’ mindfulness and functioning. This pattern suggests that similarity within couples may play a larger role than direct crossover effects. Prior research shows that romantic partners often resemble one another across a range of characteristics (Lewis & Yoneda, 2021), and that greater within-couple similarity has been associated with more favorable relational and health-related outcomes (Gonzaga et al., 2010; Lewis & Yoneda, 2021). Accordingly, the observed interdependence may reflect broader dyadic commonalities such as assortative partnering (e.g., choosing similar partners; Watson et al., 2004), health concordance (Meyler et al., 2007; Pauly et al., 2023), relationship climate (Ledermann & Kenny, 2011), and reciprocal processes (Bodenmann, 2005). In this context, mindfulness may be embedded within shared couple processes rather than operating as a direct crossover mechanism. At the same time, crossover effects may be conditional rather than uniformly detectable. Crossover theory suggests that interpersonal influence is shaped by processes such as empathic reactions and shared exposure to stressors, through which one partner’s affect or strain may be transmitted to the other (Westman, 2001). From this perspective, mindfulness may affect partner primarily under specific circumstances such as conflict, stress, caregiving, or recovery from dysregulating interactions, rather than as a stable main effect across all couple contexts (Barnes et al., 2007; Wachs & Cordova, 2007). When aggregated across heterogeneous studies, such conditional effects may attenuate, resulting in near-zero overall partner effects.

Moderator analyses by mindfulness type showed that interpersonal mindfulness yielded slightly more negative partner estimates than state or trait mindfulness, even though partner effects were near zero in the omnibus model. This pattern suggests that partner associations may be measure-contingent and may reflect conceptual overlap that confounds with aspects of relationship processes or evaluations (e.g., heightened attunement during conflicts or stress). Alternatively, the negative partner estimates may reflect compensatory coping or uneven emotional-regulation labor under relationship strain or lower partner functioning (Falconier et al., 2019). Importantly, although mindfulness type did not moderate partner or actor effects, it did significantly moderate interdependence effects, with stronger partner similarity observed for both mindfulness and outcomes. Consistent with this pattern, studies using interpersonal mindfulness measures showed stronger interdependence effects relative to state and trait measures, which may reflect samples of longer-term, more committed couples, for whom shared routines and within-couple concordance are typically stronger (Meyler et al., 2007), potentially contributing to greater dyadic interdependence rather than crossover effects.

Similarly, moderation results by outcome domain showed that pro-relationship motivation and behavior showed slightly larger partner estimates than relational well-being, though all domain-specific estimates remained small in magnitude (|β| < .10), and outcomes with few effect sizes (e.g., sexual well-being, k = 3) produced less stable bootstrap estimation. Cultural context likewise did not reliably moderate APIM partner effects, despite significantly moderating interdependence. Specifically, interdependence-oriented samples showed greater partner similarity in both mindfulness and outcomes. This pattern suggests that cultural context may shape dyadic similarity or shared experiences within couples but does not translate into stronger partner effects once actor and interdependence pathways are accounted for. Given the uneven distribution of studies across cultural contexts (independence-oriented k = 25; interdependence-oriented k = 11), these findings should be interpreted cautiously and primarily as hypothesis-generating (Hempel et al., 2013).

Although the present findings provide little evidence that actors’ mindfulness is positively associated with partners’ outcomes after accounting for actor effects and dyadic interdependence, these conclusions should be interpreted in light of several limitations that may reduce sensitivity to detecting partner effects. First, the included evidence base of the current study was predominantly cross-sectional in nature, which limits conclusions about temporal ordering and makes it difficult to distinguish crossover influences from shared couple-level processes (Kenny et al., 2020). Although a small subset of studies employed longitudinal or repeated-measures designs (k = 7, 18.9%), effect sizes were derived from baseline associations to ensure comparability within the APIM framework; thus, all estimates reflect correlational relationships rather than temporal or causal effects. Second, heterogeneity was substantial across APIM pathways (I² > 90%), indicating considerable between-study variability and limiting the extent to which pooled estimates can be generalized across samples, measures, and study contexts (Higgins & Thompson, 2002). This variability may reflect differences in mindfulness operationalization, outcome domains, and sample characteristics that could not be fully accounted for in the moderator analyses. Third, moderator analyses were often based on small or uneven numbers of studies, limiting our ability to account for heterogeneity and increasing the likelihood of failing to detect true sources of variability (Hedges & Pigott, 2004). Fourth, the current analyses were restricted to studies of heterosexual couples, and most samples were drawn from Western contexts, constraining generalizability to more diverse relationship structures, sexual orientations, gender identities, and cultural contexts (Hilpert et al., 2016). This sampling pattern also mirrors what has been noted in recent meta-analytic work on couples and mindfulness interventions (Voldstad et al., 2025). In addition, many studies did not report key demographic characteristics (e.g., immigration history, socioeconomic status), and racial/ethnic diversity was limited and primarily U.S.-based, further narrowing the representativeness of the evidence base. Fifth, it should be noted that mindfulness was assessed exclusively via self-report measures, which are vulnerable to interpretive differences, social desirability, and ongoing concerns regarding construct validity (Bergomi et al., 2012; Grossman, 2008).

Future research would benefit from designs that can more directly test whether mindfulness transmits across partners over time, including longitudinal dyadic studies and intensive repeated-measures approaches (e.g., daily diary, ecological momentary assessment) that can separate the temporal influence of mindfulness from stable couple-level factors (Ledermann & Kenny, 2011). In particular, randomized controlled trials are needed that either (a) train only one partner in mindfulness or (b) compare individual-focused vs. dyad-focused mindfulness-based interventions would provide stronger tests of causal crossover by examining whether increases in one partner’s mindfulness leads to changes in the other partner’s outcomes. More broadly, such experimental manipulation allows tests of causality beyond what is possible with correlational data. Future research may also examine mindfulness as a mediating mechanism linking individual or contextual predictors to relationship outcomes. More broadly, this perspective is consistent with patterns observed in the current study and suggests that associations between partners’ mindfulness and relationship outcomes may reflect indirect interpersonal pathways rather than direct partner effects. Again, longitudinal and experimental designs will be helpful for testing these indirect pathways and clarifying temporal ordering in dyadic processes. In addition, more consistent reporting of baseline dyadic correlations and covariance structures is needed to strengthen meta-analytic estimates of partner effects, as these parameters are essential for modeling interdependence and reducing bias in APIM-based analyses. At a process level, future studies should examine interpersonal mechanisms (e.g., conflict resolution, communication, relationship cognition) as mediators or moderators, as partner benefits may depend on whether mindfulness translates into observable interactional change (Gesell et al., 2020; Karreman et al., 2017). Researchers may also consider integrating the common-fate model with APIM in a hybrid framework, which enables simultaneous examination of individual-level and dyad-level processes (Ledermann & Kenny, 2011). Substantively, future studies should test interpersonal processes (e.g., conflict resolution, communication) as mediators or moderators, given that partner benefits may depend on whether mindfulness translates into observable interactional change (Gesell et al., 2020). Inclusive sampling for same-sex couples and non-Western couples is needed to strengthen generalizability and evaluate cultural conditions under which crossover effects might be more likely to emerge (Voldstad et al., 2025).

Lastly, although our meta-analytic APIM results did not support a direct mindfulness crossover pathway, this pattern should not be interpreted as evidence that mindfulness lacks interpersonal relevance; rather, it suggests that a simple crossover pathway may not be the primary way dyadic processes unfold in the available literature (Kenny et al., 2020). More importantly, the present study contributes new insight by showing that previously observed cross-partner correlations may be partly driven by shared dyadic variance rather than a unique partner effect, highlighting the importance of modeling interdependence when evaluating relational processes. Dyadic processes could be instead captured by alternative frameworks, including the mutual-influence framework (Woody & Sadler, 2005), in which partners’ outcomes dynamically shape one another over time, or the common-fate model (Ledermann & Kenny, 2011), in which partners’ reports largely reflect shared dyad-level factors (e.g., relationship climate, shared stress exposure) rather than directional transmission (Kenny et al., 2020). For instance, longitudinal evidence from the English Longitudinal Study of Ageing shows that one partner’s recent health behavior change predicts corresponding change in the other partner (Jackson et al., 2015), highlighting the dynamic “co-change” processes that may not be captured in cross-sectional dataset. Consistent with this perspective, the moderate interdependence observed in the present study aligns with the broader health concordance literature documenting substantial within-couple similarity in health and health-related behaviors (Falba & Sindelar, 2007; Meyler et al., 2007). Taken together, these findings suggest that mindfulness in couple relationships may operate primarily through individual pathways and shared dyadic processes, rather than through a direct crossover mechanism.

This distinction has important implications for both research and clinical practice. For researchers, these findings highlight the need to move beyond bivariate or single-path models and to adopt dyadic and longitudinal designs that can better capture how mindfulness unfolds within relational systems over time. For clinical work, the results suggest that interventions may be more effective when targeting the couple as a shared system, rather than assuming that changes in one partner’s mindfulness will directly translate to improvements in the other partner. This perspective is consistent with evidence from systematic reviews of couple-based health behavior change interventions, which show benefits when both partners are engaged in change processes (e.g., McGrath, 2017). Overall, the present study refines prior evidence by clarifying that mindfulness is reliably linked to individual functioning and dyadic interdependence, but not to a robust direct partner effect, thereby offering a more precise understanding of how mindfulness operates within intimate relationships.

Public health significance statement.

Mindfulness is often promoted as beneficial not only for individuals but also for close relationships. This model-testing meta-analysis suggests that while mindfulness was consistently related to a person’s own well-being and relationship functioning, it is not reliably associated with partners’ outcomes once shared couple influences are taken into account. These findings suggest that previously observed partner associations may reflect shared dyadic processes rather than direct crossover effects, and caution against assuming that individual mindfulness training alone will benefit partners. Instead, partner gains may require approaches that directly target couple processes. Most included studies were conducted with heterosexual male–female couples in predominantly Western, independence-oriented cultural contexts, which may limit generalizability to more diverse relationship structures, gender identities, and cultural backgrounds.