The Spanish adaptation of the relationship power inventory

Power dynamics are fundamental when negotiating conflicts. However, no instrument for measuring power in romantic relationships has been adequately adapted to Spanish culture. The goal of this research was to adapt the Relationship Power Inventory (RPI; Farrell et al., 2015) to Spanish culture and language, filling this gap by providing a rigorous instrument for evaluating this construct. Study 1 was conducted to obtain evidence based on Spanish adaptation of RPI content. Once the Spanish adaptation of the RPI was built and we obtained validity evidence based on the test content, in Study 2, the scale was administered to two different samples of the adult population following a cross-validation approach. Specifically, in Sample 1 ( N = 400), the training sample, a statistical analysis and an exploration of the dimensional structure and reliability of the measure were carried out. In Sample 2 ( N = 755), the validation sample, the internal structure of the scale was confirmed, and evidence of external validity and generalization was obtained. The exploratory and confirmatory factor analysis showed a good fit for the four-factor structure. These dimensions were invariant to gender and had adequate validity based on their relationship with other variables (dependence on the partner, conflict-resolution strategies, and psychological well-being). In sum, the Spanish version of the RPI (SARPI) is a reliable instrument with sufficient valid evidence to provide accurate measurement of power differences in the context of romantic relationships.

One of the contexts in which power dynamics play a particularly important role is romantic relationships.In particular, although people's goals, desires, and happiness in these relationships inevitably depend on their partners' cooperation and investment, both members of a relationship often do not share the same goals or interests, resulting in conflict (Clark & Mills, 2012;Kelley & Thibaut, 1978).When this occurs, individuals may try to influence or control their romantic partners to achieve their interests, which is a risk factor when an imbalance of power exists between the two partners (Kim et al., 2019;Morales-Marente, 2005).Specifically, individuals may respond aggressively and violently toward their partners as a way of imposing their goals when they perceive they have power over their partners or, conversely, may increase submissive responses and tolerance of abuse when they perceive greater power in their partners (Cross et al., 2019;Overall et al., 2016;Pietromonaco et al., 2021).As a result, these power dynamics can severely affect the individuals' health and well-being (Kifer et al., 2013;Leone & Conroy, 2019;Martín-Lanas et al., 2019).Thus, it is vitally important to analyze power dynamics because they can have serious consequences for the members of romantic relationships.
The Dyadic Power Social Influence Model (DPSIM; Simpson et al., 2015) proposed that power in romantic relationships is a dyadic property of the relationship and that to analyze its effects, it is necessary to study four relationship components depending on who is exercising power (the person/their partner) and when they exercise power (during the decision-making process /at the final outcome of the decision or discussion): (a) one's own control or influence during discussions or decision-making, for example, by proposing ideas (self-process power); (b) one's own control or influence over the final outcome of the discussion or decisionmaking (self-outcome power); (c) the partner's influence or control during discussions or decision-making (partnerprocess power); and (d) the partner's control or influence over the final outcome of such discussions (partner-outcome power; Farrell et al., 2015).
In recent years, there has been a growing interest in identifying variables closely related to the power components of romantic relationships described previously.This research focuses specifically on the relationship of power components to partner dependence, conflict resolution strategies, and well-being.On one hand, dependence on the partner is an antecedent or base of power.In fact, previous studies have shown that traditionally, people possess more power-process and outcome-to influence their partners and promote their goals when they are less dependent on their partners-either anxious, exclusive, or emotionalwhereas when they are more dependent on their partners, they possess less power or are unable to resist their attempts to influence their partners (Kelley & Thibaut, 1978).In other words, while negative correlations have been found between personal power-process and outcome-and dependence, positive associations emerged for dependence and partner power (Farrell et al., 2015).On the other hand, these power dynamics are also associated with different resolution strategies for conflicts that appear in romantic relationships and with the levels of health and well-being of each member.Specifically, prior research has shown a positive association between personal power-process and outcome-and the use of destructive and even violent responses toward the partner during conflicts (exit or neglect; Cross et al., 2019;Overall et al., 2016), whereas a negative association has been found with constructive responses (expression or loyalty; Alonso-Ferres et al., 2021;DeMarree et al., 2014).Conversely, lacking power or perceiving greater power in the romantic partner-process and outcome-is positively associated with passive responses (loyalty or neglect) that tolerate partner abuse, and negatively associated with active responses (expression or exit) that confront conflict-either constructively or destructively (Pietromonaco et al., 2021).Finally, previous work has also confirmed that lacking power or perceiving greater power in a partner-process and outcome-negatively impacts the components of an individual's psychological well-being (i.e., self-acceptance, personal growth, positive relations, autonomy, environmental mastery, and purpose in life), while possessing power has salutary effects (Kifer et al., 2013;Leone & Conroy, 2019;VanderDrift et al., 2019).
The relevance of power in different contexts, but especially in romantic relationships, has triggered the development of different measures to assess this construct.The most prominent tests are the Generalized Sense of Power Scale (ESPG; Willis et al., 2016), the Sexual Relationship Power Scale (SRPS; Pulerwitz et al., 2000), the Purchasing Decision-Making Table (Davis & Rigaux, 1974), and the Relationship Power Inventory (RPI; Farrell et al., 2015).On the one hand, the ESPG evaluates the sense of power in interpersonal relationships (e.g., work context); however, it does not assess power dynamics specific to romantic relationships.On the other hand, the SRPS assesses how often a person has the capacity to make decisions regarding sexual relations, and the Purchasing Decision-Making Table evaluates who is the economic decision-maker in the relationship.These measures, although developed in a romantic interpersonal context, studies power only in a specific domain: sexual decisions and economic decisions, respectively.Therefore, the RPI (Farrell et al., 2015) is the only measurement instrument that aims to assess dyadic power in a romantic interpersonal context.
Given the essential impact of power in relationships, it is crucial to have an instrument to assess this construct.The benefits of the RPI in measuring relationship power go beyond the other instruments for several reasons: (a) It is drawn from the dyadic power social influence model (Simpson et al., 2015), a robust theoretical framework widely supported by researchers in the area of power dynamics in romantic relationships (Kim et al., 2019); (b) it operationalizes the four theoretical power dimensions accepted (selfoutcome power, partner-outcome power, self-process power, and partner-process power), and adequate reliability for all RPI dimensions was empirically confirmed in the original study (α ≥ 0.85); (c) the RPI dimensions were not associated with social desirability, removing potential measurement bias (Farrell et al., 2015); and (d) regarding external validity evidence, research has confirmed the associations between the RPI components and dependence, conflict resolution, and well-being that have been described previously.Despite these advantages, the available data on its internal structure are still limited.In particular, the psychometric properties of the RPI have been assessed only in the original study in which Farrell et al. (2015) proposed a dimensional model-that is, fourfactors (self-outcome power, partner-outcome power, selfprocess power, and partner-process power) nested in two second-order factors (self-power and partner power)-without testing other competitive and more parsimonious models that are theoretically reasonable.In addition, although previous literature has found that men tend to be the power-holders in relationships (Felmlee, 1994), no prior research has provided any evidence regarding the invariance across genders in the measurement of power.
In Spain, although an abundant bibliography is available on the relevant aspects and consequences of power following the aforementioned conceptualization of the construct (e.g., Alonso-Ferres et al., 2021;Martín-Lanas et al., 2019;Morales-Marente, 2005), psychometrically sound measures are clearly still needed to assess the construct of power in the context of romantic relationships.Of all the scales mentioned, the RPI is the only available instrument that measures this construct in romantic relationships (Kim et al., 2019).Therefore, this study aims to provide a comprehensive, adapted, reliable, and valid assessment of the RPI through two large Spanish samples.

Research Overview
Two studies were conducted to examine internal consistency and to obtain validity evidence to support the interpretation of RPI scores for the Spanish population.Specifically, in Study 1, we sought to translate the RPI (Farrell et al., 2015) into Spanish (called the Spanish Adaptation of the RPI; SARPI).Next, we aimed to obtain evidence based on Spanish adaptation of the RPI content.The main goal of Study 2 was to provide validity evidence based on the internal structure, measurement invariance, and relationships with the above-mentioned variables (i.e., dependence, conflict resolution, and psychological well-being) to support the interpretations of the RPI scores for proposed uses in the Spanish population.Such adaptation will give research on this construct in Spain an instrument with good psychometric properties and an opportunity to compare findings across different countries and cultural settings.In addition, the Spanish adaptation of the RPI would enrich the still insufficient data on the psychometric guarantees of the original version and, moreover, it would deepen the internal structure, examining competitive models of RPI through two independent samples collected at different times and in another cultural and linguistic context.The data and syntax are available on the Open Science Framework.

Method
Participants A panel of 6 professionals was selected to serve as content validity experts for the Spanish adaptation of the RPI.An important criterion for the panel of experts selected was that they were not familiar with the RPI and they were experts in the fields of social psychology (i.e., romantic relationship dynamics and power construct) and psychometrics.Each of the 6 experts was in academia.Five of them have doctoral degrees and one a master's degree in psychology.Two of the experts engaged in research on the dynamics of romantic relationships, one of them in power, and three were experts on test construction.Professional experience ranges from three to 15 years.

Procedure
We adapted the RPI to the Spanish linguistic and cultural context following the guidelines proposed by Muñiz et al. (2013) and the recommendations of the Standards for Educational and Psychological Testing (American Educational Research Association et al., 2014;International Test Commission, 2017).
First, we obtained the necessary permission from the authors of the original RPI scale to carry out the Spanish adaptation.
Second, to ensure that culturally idiosyncratic expressions of the construct were included for the Spanish context, the 20 items were translated into Spanish using a committee approach (Harkness, 2003).Thus, a group of four professionals from different areas of expertise-linguistic, psychological target construct, and psychometrics-prepared, discussed, and revised the translation to fit the Spanish culture and language.Specifically, a bilingual specialist translated the original items of the RPI to Spanish, and then, two different specialists on the construct and psychometrics reviewed the translation introducing any changes considered necessary.After that, during a meeting, the research team in full, including a new expert in test construction, discussed the best translation to assure a balanced treatment of psychological, linguistic, and cultural considerations (Hambleton & de Jong, 2003).
Third, for the purpose of obtaining validity evidence based on content and avoiding construct biases for the Spanish RPI items, we utilized an expert-appraisal method.Specifically, we recruited a panel of six experts in the fields of social psychology and psychometrics.The role of the expert was to assess technical accuracy (i.e., comprehension, ambiguity, clarity) as well as the belonging and representativeness of the items for the intended dimensions (Sireci & Faulker-Bond, 2014).The above-mentioned aspects were evaluated through a "matching task" and Likert-type rating scales.Specifically, each expert had the semantic definition of relationship power and each dimension.Their evaluation of the domain definition and representation consisted of identifying the power dimension (i.e., self-outcome power, partner-outcome power, self-process power, and partner-process power) to which each item had been classified as belonging and indicating the degree of representativeness-from 1 = nothing representative to 4 = very representative-to which the item reflected a behavior indicative of such dimension.Subsequently, the item's appropriateness was assessed on a four-point Likert scale considering comprehension (i.e., to what extent the item was correctly understood; 1 = incomprehensible language, 4 = understandable language), ambiguity (i.e., whether the item could be interpreted in another way; 1 = varied interpretations, 4 = a single interpretation), and clarity (i.e., to what extent the item was concise/accurate/direct; 1 = nothing concise, 4 = very concise).Finally, the experts were given the opportunity to make comments regarding the accuracy of each item.
Data Analyses Content validity indices (CVIs) and Cohen's kappa coefficient (κ) were calculated for the different items to obtain a quantitative indicator of agreement regarding each item's adequacy through the areas of content-based validity (i.e., representativeness, comprehension, ambiguity, clarity).A CVI > 0.79 indicated that the item would be appropriate, between 0.70 and 0.79 that it might need revision, and < 0.70 that it should be eliminated.Likewise, Kappa values above 0.74, between 0.60 and 0.74, and between 0.40 and 0.59 were considered excellent, good, and fair, respectively (Zamanzadeh et al., 2015).

Results
As Table 1 shows, all items were mostly correctly classified into their dimensions by all experts.Likewise, all items showed excellent content validity and interrater agreement indices (CVIs > 0.78, κ > 0.74; Zamanzadeh et al., 2015), except Items 6 and 10 regarding comprehension and representativeness, respectively (CVIs = 0.67, κ = 0.67).These items were reviewed and slightly edited based on experts' comments following DeVellis' (2012) recommendations.Specifically, according to the experts' suggestions, the expression "bring up issues" was translated as "exponer problemas" instead of "plantear problemas".Likewise, "tend to" was translated as "suelo" instead of "tiendo a" to facilitate the representativeness and comprehension of such items to Spanish idiosyncratic language and culture.

Study 2
Once we completed the development of the Spanish adaptation of the RPI in Study 1 and we obtained validity evidence based on the test content, the scale was administered to two samples of adult population in order to (a) carry out a psychometric study of their properties, and (b) obtain validity evidence based on the internal structure, measurement invariance, and relationships with other variables (i.e., dependence, conflict resolution, and psychological well-being).

Method
Participants To achieve the aims of this study, two samples were collected at different times.In both samples, participants had to meet the following inclusion criteria to participate in the study: (a) be aged 18 years or older; (b) be involved in an exclusive relationship for at least 3 months before the data collection date; and (c) be Spanish citizens.Sample 1 was composed of 400 adults (200 males & 200 females) between the ages of 18 and 79 years (M = 33.32,SD = 14.59), and they had been involved in their relationships for an average of 10.01 years (SD = 12.67).In this sample, 37.3% were married, 27.3% were cohabitating, and 35.5% were dating.Sample 2 consisted of 755 adults (304 males & 451 females) with ages ranging from 18 to 78 years (M = 30.22,SD = 14.02).They were involved in their relationships for an average of 10.26 years (SD = 11.65).Most participants were dating (58.8%), followed by married (22.9%), and cohabitating (18.3%).

Instruments
The Spanish Adaptation of the RPI (SARPI; Farrell et al., 2015) was used in Samples 1 and 2. The SARPI is a 20-item measure that assesses self-outcome power, partner-outcome power, self-process power, and partnerprocess power.Respondents were asked to rate personal and partner power on a 7-point Likert-type scale ranging from 1 (never) to 7 (always).Higher scores indicated a greater power for the evaluated dimension.The internal consistency of this measure was adequate in the original version (Farrell et al., 2015): α self-outcome power = 0.94, α self-process power = 0.85, α partner-outcome power = 0.95, and α partner-process power = 0.87.
The Spouse-Specific Dependency Scale (SSDS; Valor-Segura et al., 2009) was administered in Sample 2. The SSDS consists of 17 items, scored with a 6-point Likerttype scale ranging from 1 (totally disagree) to 6 (totally agree).It measured three dimensions that showed adequate reliability in the version of the scale adapted to the Spanish population: anxious attachment (α = 0.90), exclusive dependency (α = 0.89), and emotional dependency (α = 0.89).In this study, internal consistency was also adequate.The Cronbach's alphas were 0.82, 0.79, and 0.77, respectively.
The Accommodation Among Romantic Couples Scale (ERCP; Valor-Segura et al., 2020) was used in Sample 2. This instrument has 27 items that assess four types of conflict-facing strategies: voice, loyalty, exit, and neglect.Responses are rated on a 9-point Likert-type scale ranging from 1 (never does that) to 9 (always shows that type of behavior).The internal consistency of this measure in the adapted Spanish version was good: α voice = 0.82, α loyalty = 68, α exit = 0.87, and α neglect = 0.75.For this study, Cronbach's alphas were 0.74, 0.65, 0.86, and 0.79, respectively.[Cuando tomamos decisiones en nuestra relación, finalmente mi opinión tiene más peso que la de mi pareja] 100% 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 5.I have more control over decision making than my partner does in our relationship.[Tengo más control que mi pareja sobre las decisiones en nuestra relación] 83% 16% 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 9.I have more influence than my partner does on decisions in our relationship.
Procedure We selected two independent samples composed of Spanish adults in Study 2. We followed a cross-validation approach with Samples 1 and 2. Before its larger administration, we used Sample 1, the training sample, as a pilot study for a preliminary evaluation of the psychometric properties and factorial structure of the SARPI, including item analysis, reliability assessment, and an examination of the items' distribution patterns and underlying dimensions.Sample 2, the validation sample, was employed to support the score interpretations of the SARPI by (a) confirming its dimensionality with validity evidence based on internal structure; (b) checking that the measure was invariant across gendergender being one of the bases of power imbalance (e.g., Alonso-Ferres et al., 2019;Pietromonaco et al., 2021); and (c) replicating the linkage between the relationship power and the above-mentioned target variables (i.e., validity evidence based on the relationship with other variables).We recruited both samples at different times using a convenience nonprobabilistic sampling method.For Sample 1, at several public places, trained members of the research team requested that participants collaborate by individually completing a survey in paper-pencil format.Data from Sample 2 were obtained using online administration.Specifically, we posted online advertisements on social media such as Facebook.Participants who met the inclusion criteria were given access to a subject-recruitment platform (i.e., Qualtrics) to complete the multimeasure survey.Previous studies have shown that online sampling methods are as valid as the paper-pencil method (e.g., Topolovec-Vranic & Natarajan, 2016).Following the Declaration of Helsinki, before completing the questionnaire, the participants in both studies read and signed an informed consent form in which they received information emphasizing the principles of voluntariness, confidentiality, and the anonymity of their answers, and advising them that they could interrupt the questionnaire at any time.
Data Analysis First, we explored the missing data in samples 1 and 2. While Sample 2 had no missing values, Sample 1 had a small number of missing data (ranging from 0.5 to 1.2).Given the small portion of missing values and the fact that data were missing completely at random (Little's MCAR test, χ 2 (96) = 383.05,p = 0.309), we used SPSS's expectation maximization algorithm (EM) to replace missing values in our sample.This technique provided unbiased parameter estimates and improved the statistical power of analyses (Scheffer, 2002).Then, we used Sample 1 to obtain descriptive statistics for the items and to observe whether they fit the normal distribution.We tested the multivariate normal distribution assumption using the Mardia test in the R software (version 3.6.3;R Core Team, 2017).Subsequently, after checking the matrix data with the Kaiser-Meyer-Olkin coefficient (KMO) and Bartlett's test-that is, whether there was an adequate intercorrelation between items-, we also performed an exploratory factor analysis (EFA) in Sample 1, the training sample.This allowed examination of the items' distribution patterns and the underlying dimensions using principal axis estimation and direct oblique rotation (Clarkson & Jennrich, 1988;Osborne, 2014).Given the non-normal distribution of the data, we drew from a measurement and theoretical model that supports self and partner power as interconnected.We used the above-mentioned factor extraction and rotation methods (Lloret-Segura et al., 2014).We retained the dimension numbers based on a parallel analysis (Garrido et al., 2013) and the Goodness of model fit (Brown, 2015).Specifically, we used the chi-square (χ 2 ) test, the root mean square error of approximation (RMSEA) with a 90% confidence interval, the Tucker-Lewis index (TLI), and the standardized root mean square residual (SRMR).Values less than or equal to 0.08, and less than or equal to 0.06 for RMSEA indicated an excellent and a good fit, respectively (MacCallum et al., 1996).TLI values higher than 0.95 and between 0.90 and 0.95, indicated the excellent and acceptable fit of the model to the data, respectively (Hu & Bentler, 1999;Kaplan, 2000).Otherwise, values less than or equal to 0.08 for SRMR indicated an excellent fit (MacCallum et al., 1996).Once we established the dimensions, we calculated discrimination indices for each item and the internal consistency (i.e., McDonald's Omega total coefficients) for each dimension.In addition, we calculated loadings, communalities, and percentage of explained variance for each item by its related dimension, as well as eigenvalues for the factor solution that was suggested by Kaiser criterion and parallel analysis in EFA.We conducted the analysis with R software (R Core Team, 2017).Sample 2, the validation sample, was collected once we verified that no further revision of the adapted test was necessary, and we established the factor solution of the scale in the training (pilot) sample.We first obtained descriptive statistics for the items to observe whether they fit the normal distribution.Subsequently, we used Sample 2 to establish the factorial validity of the scale.We compared several competitive models-among others, the model of the original version of the RPI-using confirmatory factor analysis (CFA).To confirm the SARPI's dimensionality, we used R software (R Core Team, 2017) to perform the analysis.We conducted the CFA using the robust maximum likelihood estimation method.We assessed the model fit using the chi-square (χ 2 ) test, comparative fit index (CFI), the Tucker-Lewis index (TLI), the root mean square error of approximation (RMSEA) with a 90% confidence interval, and the standardized root mean square residual (SRMR).As the RMSEA and TLI, CFI and SRMR follow similar criteria.Thus, CFI values higher than 0.95 and between 0.90 and 0.95, indicated excellent and acceptable fit of the model to the data, respectively (Hu & Bentler, 1999;Kaplan, 2000).Next, we calculated the internal consistency for each dimension through McDonald's Omega total coefficients.Furthermore, we also tested different levels of measurement invariance in Sample 2 across gender using multigroup CFAs.The less restrictive, or configural, model determined whether men and women conceptualized relationship power in the same way, estimating the same model for both groups without constrained parameters (i.e., loadings, thresholds, and item variances).The metric invariance model introduced constraints-that is, factor loadings for both models were equal-confirming whether men and women understood the items on the SARPI equally.A scalar model incorporated constrained thresholds for both models to observe whether the latent factors showed the same item scores for men and women.Subsequently, we assigned strict invariance model fixed loadings, thresholds, and item variances at the same value across groups so that the measurement error could be discerned as equal between men and women.The cutoff values proposed by Cheung and Rensvold (2002) to support a more restrictive invariance measurement model were the changes in CFI of less than or equal to 0.010 and in RMSEA of less than or equal to 0.015.Last, we computed validity evidence based on the relationship with other related variables in Sample 2 using Jamovi (Version 1.2.16.0;The jamovi project, 2021).Pearson correlation and attenuated correlation (to eliminate measurement error)-and their 95% confidence intervals-were calculated between SARPI dimensions and the SSDS, the ERCP, and PWBS (American Educational Research Association et al., 2014).

Results
Preliminary Analyses As Table 2 shows, skewness and kurtosis values for the observed variables (i.e., item) were acceptable in samples 1 and 2. However, the Kolmogorov-Smirnov test (univariate normality) was significant for all items in both Samples (ps < 0.001), as was the Mardia test (multivariate normality) in Sample 1 (MS = 3856.84,p < 0.001; MK = 51.29,p < 0.001) and Sample 2 (MS = 4489.19,p < 0.001; MK = 64.03,p < 0.001), indicating that the samples did not follow a strictly normal distribution.Exploratory Analysis Bartlett's test (χ 2 = 6578, df = 190, p < 0.001) and the KMO coefficient (0.92) in Sample 1 pointed toward an adequate intercorrelation among these items, allowing the interpretation of the factorial solutions.As Fig. 1 shows, the cross between slopes pointed out that the eigenvalues from random data started to exceed the eigenvalues from research data after the fourth factor, suggesting the parallel analysis of a four-factor solution.Similarly, it is supported acceptably by the goodness-of-fit indices (χ2[116] = 458.66,p < 0.001, TLI = 0.93, RMSEA = 0.068, 90% CI [0.05, 0.07], SMRS = 0.03).The EFA resulted in a four-factor solution (see Table 3)-these are consistent with the four dimensions of relationship power-that explained 67.9% of the total variance.Such factors showed eigenvalues of 4.86, 4.81, 2.22, and 1.66, respectively.Likewise, the items presented factor loadings higher than 0.40 in the dimensions proposed by the original version of the RPI and suitable discrimination indices (> 0.50).Commonalities also showed that the common variance of items explained by each dimension was between 30% (for item 10 by self-process power) to 85% (for item 16 by partner-outcome power; see Table 3).
Evidence Based on Internal Structure Through CFAs, overall goodness-of-fit indices for the different competitive models were calculated in Sample 2. Specifically, we checked (a) a model with four first-order factors (M1; self-outcome power, self-process power, partner-outcome power, and partner-process power) nested into two correlated second-order factors  (self-power and partner power); (b) a model with four first-order correlated factors (M2; i.e., self-outcome power, self-process power, partner-outcome power, and partner-process power); and (c) a model with two first-order correlated factors (M3; i.e., selfpower and partner power).The first model was consistent with the original proposal, whereas the latter two models were based on empirical results obtained previously and on the original theory (see the Supplemental Material for each model factor structure and specification).As Table 4 shows, M1 and M2 fit better to the data in comparison with M3.However, there are small differences in the goodness-of-fit indices between M1 and M2.We performed a likelihood ratio test to compare the fit between M1 and M2.We found a significantly better fit of M2 in comparison to M1 (χ 2 = 14.90, df = 46, p = 0.001).In addition, results suggested that M1 might be misspecified, and thus, it is not appropriate for the data (i.e., presents negative variances in the self-process power and partner-process power factors, as well as factor loadings greater than 1 in the standardized solution).Together, these results indicated that the correlated four-factor model (M2) fit better to the data in comparison with the other proposed models and provided a more parsimonious factorial solution (Flanagan et al., 2012).Based on that, the four first-order correlated factors model (M2) was endorsed as the latent structure of the scale for the Spanish population.
The CFA for M2 confirmed a nonsignificant correlation among partner-outcome power, self-outcome power (r = − 0.10, p = 0.18), and self-process power (r = − 0.13, p = 0.06).Moreover, modification indices were high for correlations between the error measurement for items 6 and 12 (MI = 61.43)as well as the correlation between both items (r = 0.68).After reviewing these items, redundant contentalmost the same wording and terminology-was found and not shared with any other item of their respective power's dimension.Specifically, items 6 and 12 belong to the partner-process dimension and share the expression "issues in our relationship".We understand that part of the uniqueness that is not a random error is shared by both items.This fact supported the re-specification of the model (Brown, 2015).Therefore, we recalculated the correlated model with four first-order factors (M2) by removing nonsignificant correlations between dimensions that were not related and introducing correlations between the measurement error of item 6 and 12 as new parameters.The final fit was excellent: χ 2 (165) = 586.34,p < 0.001, CFI = 0.95, TLI = 0.94, RMSEA = 0.058, 90% CI [0.05, 0.06], SRMR = 0.06.
Figure 2 shows the factor structure of the SARPI.All factor loadings were higher than 0.80 and statistically significant.Likewise, all dimensions showed excellent internal consistency: self-process power (ω = 0.80), self-outcome power (ω = 0.94), partner-process power (ω = 0.82), and partner-outcome power (ω = 0.93).5 shows, Sample 2 supported the configural and metric invariances.Thus, we can conclude that men and women conceptualized the construct of power in romantic relationships and interpreted the items in the same way.Sample 2 did not support the scalar invariance model.Therefore, following Byrne et al.'s (1989) recommendations, the intercept of item 11 was allowed to vary across groups based on the high value in the modification index (MI = 25.0,p < 0.001).After leaving the intercept free for this item between the groups, the CFI and RMSEA did not change above the cutoff points (Cheung & Rensvold, 2002).Thus, the partial scalar invariance model was finally supported.Finally, taking into consideration both the partial scalar invariance model and the strict factorial invariance model, we can conclude that both means and variances of the four SARPI dimensions could be compared between men and women.

Validity Evidence Based on Relationships with Other Variables
We tested the validity evidence based on the relationships with other variables in Sample 2. As the values and 95% confidence intervals of the attenuated correlations in Table 6 indicate, self-process and outcome powers correlated negatively with partner dependence (except for anxious attachment).In addition, higher self-process and outcome powers were related to a greater use of destructive (exit and neglect) and less constructive-passive (loyalty) conflict-resolution strategies.Although the constructive-active strategy (voice) and self-outcome power were not associated, a significant and positive relationship was found between the constructive-active strategy (voice) and self-process power.Self-process power was also positively related to all Table 4 CFA Fit Indices for Competitive Models of the SARPI N 2 = 755; M1 = original model of four factors (self-process power, self-outcome power, partner-process power, and partner-outcome power) nested into two second-order correlated factors (self-power and partner-power); M2 = four-factor correlated model (self-process power, self-outcome power, partner-process power, and partner-outcome power); M3 = two-factor correlated model (self-power and partner-power) dimensions of psychological well-being (except positive relationships), whereas self-outcome power correlated positively with the dimensions of self-acceptance and autonomy.
The partner-process and outcome powers were related to greater dependence on the partner and greater use of passive (loyalty and neglect) and less constructive-active (voice) Note.so = self-outcome power; sp = self-process power; po = partner-outcome power; pp = partner-process power conflict-resolution strategies.No relationship was found with the destructive-active strategy (exit).Finally, both types of partner power-process and outcome-correlated with lower subjective psychological well-being in all dimensions.These results are similar to those found in previous literature (e.g., Farrell et al., 2015).

General Discussion
Power dynamics are fundamental when negotiating interpersonal conflicts.In particular, during a romantic couple's conflict, the parties attempt to influence each other to find a solution that suits both (Kelley & Thibaut, 1978).However, one may have more or less power to influence the other and thus attempt to regain power through aggressive behaviors (Cross et al., 2019;Overall et al., 2016) or relinquish it by yielding to or adopting the needs of their partner (Pietromonaco et al., 2021).The consequences of such power dynamics can be quite varied, even seriously affecting the individuals' health and well-being (Kifer et al., 2013;Leone & Conroy, 2019;Martín-Lanas et al., 2019).Therefore, the assessment of power dynamics between partners should be a central theme in the study of and intervention in marital conflicts.Because Spain lacks a measurement instrument to assess this construct, the main objective of this research was to adapt the RPI, providing evidence of validity to support the interpretation of its measurements in the general Spanish population.
First, a panel of experts verified the validity of the SARPI based on its content, ensuring that the items adequately captured the content of the construct in the Spanish context.Subsequently, the psychometric properties and internal structure of the scale were evaluated in two different samples based on the model proposed by the authors of the original scale (Farrell et al., 2015).The internal structure of the SARPIfour dimensions nested into two second-order dimensions as proposed in the original study-was not supported after the exploratory study and the CFA.The proposed analyses indicated that, unlike the original model, the alternative four correlated factor model (self-process power, self-outcome power, partner-process power, and partner-outcome power) had a better fit in the Spanish context.It is relevant to highlight that the methodological approach conducted in the research has not been fully reported for the original RPI.Indeed, the dimensional structure of the original scale was accepted even though it had some limitations in its fit and without its corroboration in other contexts different from the U.S. context.In line with the theoretical starting model (dyadic power social influence model; Simpson et al., 2015), the new proposed structure was endorsed as the latent structure of the scale for the Spanish population.This endorsement overcame the deficiencies of the original dimensional structure by eliminating the secondorder factors composed of only two components, which consequently makes the model more parsimonious.Having performed a committee adaptation approach (instead of literal translation or assembly; Van de Vijver & Tanzer, 2004) during the translation process, together with findings from the expert appraisal, psychometrics results, and validity evidence based on internal structure, we are confident about the appropriateness of the SARPI four correlated factor structure.
Next, we obtained evidence of the measurement invariance of the RPI for the first time in any of its language versions.In our analysis, we compared the underlying construct of power between men and women to ensure the invariance of the SARPI across genders and to verify that the theoretical structure of both groups was similar (Felmlee, 1994).With these results, we conclude that men and women conceptualize the latent structure of the scale in the same way and interpret the items in a similar way, and that the means and variances of the four dimensions of the SARPI can be compared between both groups.Therefore, gender-differentiated studies on power dynamics in romantic relationships can be conducted using the adapted scale.
Finally, validity evidence based on the relation to other variables replicated previous correlations obtained between the SARPI, SSDS, ERCP, and PWBS (Farrell et al., 2015).Specifically, higher scores on self-power-process and outcome-are associated with lower dependence, destructive conflict resolution (both active and passive), and greater self-acceptance and autonomy.Conversely, higher scores on partner power-process and outcome-are associated with greater partner dependence, more passive conflict resolution, and lower levels of psychological well-being.Our work shows, according to the DPSIM (Simpson et al., 2015), that power is a unique construct that can be determined by different bases (e.g., dependence) and predicts key outcomes (e.g., conflict resolution and well-being).Specifically, in accordance with the principle of "least interest" (Kelley & Thibaut, 1978), these results support the idea that the person who is less involved in the relationship, and who thus may be less affected by the actions of their partner, possesses relatively greater levels of power to influence their partner and promote their own personal desires.Conversely, the person who is more engaged and whose goals and happiness are more dependent on the relationship is less able to exert influence (Kim et al., 2019;Simpson et al., 2015).This leads them to adopt passive responses toward their partner for fear of losing their relationship (Pietromonaco et al., 2021) and consequently affects their personal well-being (Kifer et al., 2013;Leone and Conroy, 2019).Results drawn from this research suggest the importance of having a specific measure of power in romantic relationships.The SARPI will make it possible to study these power dynamics in couple relationships in the Spanish context and thus stop this possible cycle of negativity.For the same purposes, the SARPI could be used in other Spanish-speaking contexts such as Latin America as long as two recommendations are followed (Van de Vijver, & Tanzer, 2004).First, items should be reviewed drawing from an adaptation method (i.e., changing idioms which are not suitable for a Spanish-speaking cultural or linguistic context, adding or removing items in order to avoid construct or item bias).Second, some validity evidence (e.g., validity evidence based on the internal structure or the relationship with other constructs) might be provided to support that the inferences from the SARPI scores are as suitable as the inferences made using the Spanish version for Spain.
The current findings may be important not only for the psychometric investigation of power in romantic relationships but also for clinical practice.From a theoretical perspective, our results provide a dyadic measure of power that will allow the researcher to evaluate, regardless of the individuals' gender: (a) the relative levels of power between partners, and how the (im)balance of relationship power predicts key outcomes (i.e., health and well-being) at both individual and couple level; and (b) whether power dynamics would have an important role in other challenges of a relationship such us emotions, decision making, or sacrifice during work-family conflicts.Marital therapists could use this information to develop intervention programs for couples seeking to improve relationship success.Additionally, given that power disparities represent one of the main reasons why couples seek therapy (Parker, 2009), therapists could use the SARPI to evaluate whether couples have healthy levels of relational power, and the effectiveness of their intervention programs, as well as to detect critical areas where interventions should be more precise (e.g., process or outcome power).Therefore, this assessment tool may be useful for a broad conceptualization and approach to power dynamics, as well as to show progress and intervention efficacy.
In future research, and with the aim of overcoming the limitations of this study, it would also be interesting to validate the SARPI in other age groups or fields of study, such as the adolescent population, using different sampling procedures to eliminate possible self-selection bias.Finally, given that with the Spanish population, we found a factorial model with certain differences from the one validated with the U.S. population, it would also be interesting to (a) be cautious and further confirm the proposed four correlated factor structure for the SARPI in future studies in Spain but also in other similar cultural contexts, and (b) explore possible cultural differences in the power dynamics that arise in romantic relationships.For example, future research could provide evidence about invariance measurement through the U.S. and Spanish samples to ensure the construct is similar, the understanding of the items is in the same way, and the means of SARPI dimensions can be compared across cultural groups.Having instruments with evidence of validity and reliability in different cultural contexts and languages is a valuable resource for intercultural studies.
In conclusion, this work shows the first evidence of reliability and validity of the adaptation of the RPI to the Spanish language and context, offering a useful instrument to evaluate the power dynamics that occur in such intimate contexts as romantic relationships.

Fig. 1
Fig. 1 Parallel analysis using eigenvalues from SARPI adaptation data (blue slope) and random data (yellow slope)

Fig. 2
Fig.2Factor Structure of the Spanish Version of the Relationship Power Inventory (SARPI).Note.so = self-outcome power; sp = self-process power; po = partner-outcome power; pp = partner-process power

Table 1
Proportion of matches between classification of items into each SARPI dimensions, CVI, and Kappa values for item characteristics

Table 3
EFA loadings of the 20 items of the SARPI on the four rotated factors resulting of factor retention analyses N 1 = 400; h 2 = commonality Higher item lodging for each factor are shown in italics

Table 5
Fit Indices and Comparison of Invariance Models