Running head: SEXUAL RELATIONSHIP POWER SCALE

The Sexual Relationship Power Scale: A Factor Analysis Nathalie DOUGOUD, RN BSN University of Rochester, School of Nursing

Running head: SEXUAL RELATIONSHIP POWER SCALE Abstract Background: Initial assessment: Initial extraction: Further analyses: Conclusion:

Running head: SEXUAL RELATIONSHIP POWER SCALE

Background The Sexual Relationship Power Scale (SRPS) (appendix 1), was developed in 2000 to measure power in sexual relationships and to investigate both the role of relationship power in sexual decision-making, and its relation to human immunodeficiency virus (HIV) risk (Pulerwitz et al., 2000). It was developed in light of the fact that women of racial minorities were at that time the fastest growing group with HIV infections. In addition, the SRPS aimed to understand how gender-based imbalances and power differentials modulate womens ability to negotiate safer sex (Pulerwitz et al., 2000). Theory Development of the SRPS drew on both structural and psychosocial theories to understand relationship power. The structural Theory of Gender and Power focuses primarily on gender-based imbalances looking at economic equality, male partner control within relationships and social norms related to gender roles (Pulerwitz et al., 2000). Social Exchange Theory examines the interpersonal definition of relationship power and how power is the resistance on the part of one individual that can potentially be overcome by another. Relationship power can be expressed in the ability to control a partners wishes, ability to go against a partners wishes and ability to dominate decision-making (Pulerwitz et al., 2000). Combined, both theories provide insight on how gender-based inequalities manifest themselves in individual relationships. This guided the development of items for the SRPS (Pulerwitz et al., 2000).

SRPS Characteristics The SRPS has two subscales: the Relationship Control Subscale and the DecisionMaking Dominance Subscale. It was developed into both an English version and a Spanish version. Initial assessment of the scale showed a good internal reliability (alpha=0.84 for the English version, 0.88 for the Spanish version). Factor analysis (FA) also confirmed the two subscales and allowed developers to decrease the number of items from 43 to 23 (Pulerwitz et al., 2000). The 15 items of the Relationship Control subscale utilize a four-point Likert scale, while the eight items of the Decision-Making Dominance subscale use a three-point Likert scale (Pulerwitz et al., 2000). Study and Sample The sample used for the purpose of this factor analysis is from the Harlem River Couples Project study database (McMahon, 2007). The Harlem River Couples Project was a three-group randomized control trial (RCT) designed to test the efficacy of HIV intervention modality and intervention content on drug using womens sexual risk with primary partners. The researcher hypothesized that interventions administered to couples rather than to only women only; and interventions that focus on relationships in the context of HIV, will result in a decrease of sexual risk behaviors among drug using women and their primary partners (McMahon, 2013). The sample included 660 individuals who were self-reported HIV negative women IV drug users and their primary male sex partners. Participants were recruited in New York City between 2005 and 2007. Inclusion and exclusion criteria are described more fully in the original study (McMahon, 2013). Participants were given assessments at baseline, three months follow-up and nine months follow-up. For the purpose of this

Running head: SEXUAL RELATIONSHIP POWER SCALE

paper, FA of the SRPS was done on the baseline assessment data due to the greater amount of missing data with the following assessments. Initial Assessment Preliminary assessment was conducted to assess adequacy of the sample size and correlation matrix for a factor analysis. The study included 660 individuals. There were 27 missing cases (4.1%). Because this is less than 5%, missing cases were excluded list wise (Tucker, 2013). Evaluation of the correlation matrix found some moderate correlations of 0.30.572. Because not all the correlations are near zero and there are some moderate correlations, we can assume that there is at least one factor present (Di Iorio, 2005). The determinant evaluates whether the correlation has an inverse. This value ranges from zero to one. A value of one indicates that the correlation is an identity matrix and thus the FA cannot be done. A value of zero indicates the matrix is a linear combination of another aspect and a FA cannot be computed as well (Di Iorio, 2005). Determinants should be between zero and one. The determinant here is 0.05 and the sample is thus appropriate to conduct an FA. Sampling Adequacy There are two tests to evaluate sampling adequacy, the Kaiser-Mayer-Olkin test (KMO) and the Bartletts test. The KMO compares differences from item to item on the correlation matrix to the second matrix of partial correlations (Di Iorio, 2005). This value should be greater than 0.7 to meet criterion. The Bartletts Test examines whether the correlation matrix is an identity matrix. It should be significant at <0.05 (Tucker, 2013). The KMO for this initial assessment is 0.862. This is greater than 0.7 and thus meets criterion for sampling adequacy. The Bartletts test is significant at p=0.000 and also meets criterion for sampling adequacy. Initial Factor Analysis Descriptive statistics of the data showed a mean centered around 2.0 to 3.0 with a standard deviation of less than 1.0. This suggests that the data is normally distributed. Histograms of each item showed that responses followed the normal distribution curve, also suggesting normally distributed data. Because it appears I have normally distributed data, I used the maximum likelihood FA option (Tucker, 2013). Initial extraction was based on Eigen values > 1 and used an un-rotated factor solution. Total Variance Explained and Scree Plot For this initial extraction, five factors were found to explain 52.32% of the total variance. While on the low side, this is considered fairly good as it is greater than 50% (Tucker, 2013). Factor one explained 24.4% of the cumulative variance, factor two explained 35.46% of the cumulative variance, factor three explained 41.89% of the cumulative variance, factor four explained 47.51% of the cumulative variance and factor five explained 52.32% of the cumulative variance. The scree plot (Figure 1) suggests three factors exist based on the vertical line assessment. Chi-Square Goodness of Fit The Chi-Square Goodness of Fit for this initial extraction was significant at 0.000. A significant Chi-Square suggests that there might be more factors that need to be extracted (Di Iorio, 2005). Analysis was thus attempted with six and seven factors. ChiSquare was still significant and factor matrices presented hardly any factor loading.

Running head: SEXUAL RELATIONSHIP POWER SCALE

Because theory and scale development suggest two subscales and thus two factors, it did not seem appropriate to attempt the factor analysis with more factors (Di Iorio, 2005). Instead, it seemed most appropriate to continue with analysis by decreasing the number of factors. Rotation In an un-rotated factor solution, analysis will not optimally identify item clusters and will result in much cross loading and items that do not load highly on any factor (Tucker, 2013). Rotations allows for lines of best fit to re-arrange through the original multidimensional item data in order to optimally go through clusters with common variance and help find a more interpretable factor structure (Tucker, 2013). For this analysis, I tried both oblique and orthogonal rotation and did not find any significant differences. However, because I have scare cross-loading in my initial factor matrix (Tucker, 2013), I chose to stay with an orthogonal rotation. Factor Matrix Initial factor matrix presented some correlations >0.32 and some scarce crossloading. A simple or clean pattern matrix would have no cross-loading of factors as each item would load (>0.32) onto only one factor. In addition, at least two or three items would highly load onto one factor (Tucker, 2013). This would clearly demonstrate which items belong to which factor, no item would belong to multiple factors at once. This is the ideal factor matrix in an FA and is not what was found for this first extraction.

Further Analyses Analysis Based on Scree Plot Based on the scree plot, there should be three factors present in this scale. The analysis was therefore rerun as described above and with orthogonal rotation, but limited to three factors. In the rotated factor matrix (table 1), items one and two loaded (>0.32) on factor three, items three to 15 (without item 14), loaded on factor 1 and items 16-23 loaded on factor 2. Item 14 loaded on none of the factors and items three to eight cross loaded on more than one factor. Analysis Based on Theory While the analysis above shows a cleaner pattern matrix than the initial extraction, it is still not optimal. FA was thus conducted based on theory and the subscales that suggest two factors. In this factor matrix (table 2), items one through 15 excluding item 14, loaded highly on factor one, items 16-23 loaded highly on factor two, item 14 loaded on none of the factors and items six and 22 cross-loaded on both factors. Based on the scale (appendix 1), one would expect factors 1-15 to load highly onto factor one and items 16-23 to load highly onto factor two because that is how the questions are grouped to each subscale. With the exception of item 14, this is what was found. This analysis based on the theory of the scale seems more appropriate and is congruent with the initial developmental FA performed by Pulerwitz and colleagues (2000). Analysis without Item Fourteen Item 14 is my partner always wants to know where I am. While this seems like an appropriate question, it does seem to be an item that would be relevant to any normally functioning couple. Knowing where your partner is may facilitate daily household activities such as taking the children to sports practice or going grocery

Running head: SEXUAL RELATIONSHIP POWER SCALE

shopping and may not necessarily inform power imbalances in the couple. Because it does not seem as relevant as other items and did not load onto any factors, analysis was run without this item. Factor matrix (table 3) had items 1-15 load highly on factor one and items 16-23 load highly on factor two. Items six and 22 still cross load on both factors. Item six is My partner has more say than I do about important decisions that affect us. While this is in the Relationship Control subscale, it does seem to also pertain to the Decision-Making dominance subscale because it relates to who has more to say about making decisions. As such, cross-loading of the item onto both factors makes sense based on the content of the item. It is unclear why item 22 crossloads, however because it addresses the use of condoms in the context of sexual relationship power, it seems appropriate to keep it in the items for the FA. Analysis with Women Only The study from which the data was used for this factor analysis administered the SRPS to both men and women in 330 dyads (660 individuals). In order to keep with the intent of the study and the manner in which the measure was administered, the factor analysis was done on all 660 individuals. However, the SRPS was initially developed as an instrument to be given to women, and was tested and validated within that population. As such, while results of the factor analysis using both men and women closely match those of the initial factor analysis, it seemed appropriate to also run the analysis with just the 330 women in the study. The analysis was run in the same way as described above with men and women. Both the initial assessment and the sampling adequacy met criterion. Initial extraction found five factors explained 62% of the total variance, which is about 10% more than with the men and women included together. Five factors were initially determined for the women only sample, which is the same as for the men and women sample. The rotated factor matrix showed similar results to the one with men and women. Most items did not load on any factors, there were only a couple items that significantly (>0.32) loaded on a factor and there was little cross loading. Scree plot suggested four factors. When the analysis was run again with four factors it found more items that loaded onto a factor but many that still did not. In addition, more cross-loading was present. The analysis was run again constraining to two factors as suggested by theory. Rotated factor matrix (table 4) limited to two factors with women only was very similar to the men and women group: items one through 15 (excluding 14) loaded highly onto factor one and items 16 to 23 loaded highly onto factor two. Item six also cross-loaded on both factors and was expected to do so due to the nature of the question which suggests it could belong to the Decision-making Dominance subscale as opposed the Relationship Control l. Overall, the results for the women only sample are almost identical to the men and women sample analysis with the exception that in the women and men group, both item 6 and item 22 cross-loaded on both factors. Conclusion In general, this factor analysis was consistent with the results found in the initial factor analysis of the scale as presented by Pulerwitz and colleagues (2000). With the exception of item 14 that loaded onto none of the factors and two items that cross-loaded, findings found the same items loaded onto the same factor. Indeed, items one through 15, which are part of the Relationship Control subscale loaded onto one factor; and items 16 through 23, which are part of the Decision-Making dominance subscale loaded highly

Running head: SEXUAL RELATIONSHIP POWER SCALE

onto another factor. This analysis however, does lend itself to two possible modifications of the scale. Firstly, item 14 could be excluded as it does not seem to be a relevant question theoretically and did not load onto any factor in any of the analyses. Secondly, item 6, although still loading more highly onto factor one, could conceivably be moved to factor two both because it does cross-load and because it appears to address more decision-making than relationship control. Lastly, while this scale has, to the knowledge of the author, only been tested in women, the factor analysis using both men and women suggest that items hang very well together in this sample as well. This is a testament to the strength of the instrument. Perhaps future studies could assess validity and reliability of the instrument in a population using both men and women rather than just women.

Running head: SEXUAL RELATIONSHIP POWER SCALE

Appendix 1. Sexual Relationship Power Scale Sexual Relationship Power Scale (Spanish language scale also available) Relationship Control Subscale Each of the following items was scored on a 4-point Likert scale, where 1 = Strongly Agree, 2 = Agree, 3 = Disagree, and 4 = Strongly Disagree. 1. If I asked my partner to use a condom, he would get violent. 2. If I asked my partner to use a condom, he would get angry. 3. Most of the time, we do what my partner wants to do. 4. My partner wont let me wear certain things. 5. When my partner and I are together, Im pretty quiet. 6. My partner has more say than I do about important decisions that affect us. 7. My partner tells me who I can spend time with. 8. If I asked my partner to use a condom, he would think Im having sex with other people. 9. I feel trapped or stuck in our relationship. 10. My partner does what he wants, even if I do not want him to. 11. I am more committed to our relationship than my partner is. 12. When my partner and I disagree, he gets his way most of the time. 13. My partner gets more out of our relationship than I do. 14. My partner always wants to know where I am. 15. My partner might be having sex with someone else. Decision-Making Dominance Subscale Each of the following items was scored in the following manner: 1 = Your Partner, 2 = Both of you Equally, and 3 = You. 16. Who usually has more say about whose friends to go out with? 17. Who usually has more say about whether you have sex? 18. Who usually has more say about what you do together? 19. Who usually has more say about how often you see one another? 20. Who usually has more say about when you talk about serious things? 21. In general, who do you think has more power in your relationship? 22. Who usually has more say about whether you use condoms? 23. Who usually has more say about what types of sexual acts you do? (Pulerwitz et al., 2000)

Running head: SEXUAL RELATIONSHIP POWER SCALE

Figure 1. Scree plot of initial factor analysis extraction with both men and women. Suggests there are three factors.

Running head: SEXUAL RELATIONSHIP POWER SCALE Table 1

Running head: SEXUAL RELATIONSHIP POWER SCALE Table 2.

Note. Rotated Factor Matrix using two factors as suggested by theory and scale development.

Running head: SEXUAL RELATIONSHIP POWER SCALE Table 3.

Table 3. Rotated Factor Matrix as suggested by theory and scale development without item 14.

Running head: SEXUAL RELATIONSHIP POWER SCALE

Table 4. Rotated Factor Matrix using women only as suggested by theory and scale development

Running head: SEXUAL RELATIONSHIP POWER SCALE

References Di Iorio, C. K. (2005) Measurement in health behavior: Methods for research and evaluation. San Francisco, CA: Jossey-Bass. McMahon. J.M., Tortu, S., Pouget, E.R., Torres, L., Rodriguez, W. & Hamid, R. (2013). Effectiveness of Couple-Based HIV Counseling and Testing for Women Substance Users and Their Primary Male Partners: A Randomized Trial. Advances in preventive medicine, 2013. McMahon, J. (2007). The Harlem river couples project, [data file and code book]. Pulerwitz, J., Gortmaker, S. L., & DeJong, W. (2000). Measuring sexual relationship power in HIV/STD research. Sex Roles, 42(7-8),637-660. Tucker, R. (2013, November 13). Factor Analysis in Scale Development. Lecture presented for NUR 513 Research Measurements ,University of Rochester, Rochester, NY.