Paper 9 Ok

C 2006) AIDS and Behavior, Vol. 10, No. 1, January 2006 (
DOI: 10.1007/s10461-005-9053-7

Pathways to Sexual Risk Reduction: Gender Differences and Strategies for Intervention Douglas Longshore,1,2,5 Judith A. Stein,3 and Dorothy Chin4 Published online Dec. 30, 2005

Using the AIDS risk reduction model as our conceptual framework and structural equation modeling as our analytic tool, we tested psychosocial antecedents of sexual risk reduction among heterosexually active men and women who use illegal drugs. With baseline sexual risk behavior controlled, stronger commitment to safer sex predicted less sexual risk behavior for both men and women. For men but not women, greater AIDS knowledge predicted safer sex commitment. For women but not men, higher self-efficacy predicted stronger commitment to safer sex, and peer norms favoring sexual risk reduction predicted higher self-efficacy. Intervention for men should focus on increasing safer sex commitment and AIDS knowledge. Intervention for women should promote safer sex commitment by raising self-efficacy for sexual risk reduction. KEY WORDS: HIV; sexual risk reduction; gender differences.

INTRODUCTION

ance in sexual risk-taking and are amenable to intervention (DiClemente and Peterson, 1994; Fishbein et al., 1991). We used the AIDS risk reduction model (ARRM; Catania et al., 1990) as a conceptual framework to guide a prospective analysis of sexual risk reduction in a sample of 600 heterosexually active, HIV-negative persons involved in illegal drug use (337 men and 263 women). The sexual risk reduction measure was a latent variable based on number of sex partners and risky sexual behaviors such as engaging in sex while on drugs and having sex with high-risk partners. These indicators represent behavior by which women as well as men may be able to exert indirect control over their sexual risk. The ARRM is a synthesis of the Theory of Reasoned Action (Ajzen and Fishbein, 1980), Social Cognitive Theory (Bandura, 1997), and the Health Belief Model (Rosenstock, 1990). According to the ARRM, risk behavior change is a three-stage process. First, people come to perceive their behavior as a possible source of HIV infection. Second, they develop a conscious commitment to behavior change; and, third, they act on this commitment. Progress through the three stages depends upon other

Sexual risk-taking is common among users of illegal drugs (Corby et al., 1996). In studies tracking HIV risk behavior over time, drug users have reported no significant reduction in their number of sex partners, and few drug users report using condoms consistently (e.g., Longshore et al., 1998a). Even among drug users who participated in interventions to promote behavior change, sizable and lasting declines in sexual risk-taking have typically not occurred (e.g., Longshore, 1992; Rhodes et al., 1996; Stephens et al., 2000). Hence there remains an urgent need to improve HIV preventive interventions for drug users and to identify factors that explain vari1 Drug

Abuse Research Center, University of California, Los Angeles, California. 2 Drug Policy Research Center, RAND Corporation, Santa Monica, California. 3 Department of Psychology, University of California, Los Angeles, California. 4 Department of Psychiatry and Biobehavioral Sciences, University of California, Los Angeles, California. 5 Correspondence should be directed to Douglas Longshore, Drug Abuse Research Center, University of California Los Angeles, Los Angleses, California; e-mail: [email protected].

93 C 2006 Springer Science+Business Media, Inc. 1090-7165/06/0100-0093/0


94 psychosocial factors. For example, AIDS knowledge regarding virus transmission routes, symptoms, etc., is posited as a factor that influences perceived infection risk (stage 1). Self-efficacy, or confidence in one’s ability to perform risk-reducing behaviors, is a predictor of both commitment to behavior change (stage 2) and actual behavior change (stage 3). Behavior change can be analyzed as an end-point (such as condom use) or a series of incremental steps (such as raising the idea of condom use with a sexual partner) leading to the end-point. Finally, some factors in the ARRM are not stage-specific but instead are hypothetically relevant as motivators of movement across stages. These include social factors such as peer norms regarding risk behavior and cues to action such as information received from an HIV education program. It is important to test predictors of behavior change in theory-based multivariate models rather than in isolation (Fisher et al., 1995). When multivariate models are applied, the relationship between any given predictor and risk behavior is less liable to misspecification, and overall implications for theory and practice should be clearer. The ARRM model offers two important advantages over other psychosocial models. First, as a synthesis of other models, it reflects the accumulated wisdom regarding many domains of health behavior. Second, the ARRM was formulated as a framework for research specifically on HIV risk behavior. Only a few studies have used the ARRM to test pathways to risk reduction among drug users (e.g., Longshore et al., 1996, 1997; Malow and Ireland, 1996; Malow et al., 1994), and evidence regarding ARRM factors is mixed. For example, the ARRM and other theories suggest that behavior change is more likely among people who believe their current behavior confers greater risk, yet both cross-sectional and prospective studies among drug users have found no relationship between perceived infection risk and more frequent condom use and fewer sex partners (Malow et al., 1993; McCusker et al., 1993; Robles et al., 1995). On the other hand, self-efficacy has correlated with safer sex in cross-sectional studies by Kok et al. (1991), Longshore et al. (1994), and Malow et al. (1993). Interventions to reduce sexual risk-taking must take gender into account. Also, it is important to try to resolve the mixed findings regarding gender differences in the relationship between self-efficacy and sexual risk reduction (Longshore et al., 1998b; McCusker et al., 1993; Montoya, 1998; St. Lawrence

Longshore, Stein, and Chin et al., 1998; Wingood and DiClemente, 1998). Literature reviews suggest that decisions to engage in protected or unprotected sex may be made most often by the male partner (Amaro, 1995; McCusker et al., 1993; Rhodes et al., 1996). In addition, for women, direct attempts to reduce sexual risk via condom use may conflict with values of intimacy and trust and with gender roles of modesty and submissiveness (see also Flores-Ortiz, 1994; Hobfoll et al., 1994; Lear, 1995; Wingood and DiClemente, 1998). On the other hand, women may exert indirect control over sexual activity via mate selection and managing the pace and circumstances of sexual intimacy (e.g., Fisher and Gray, 1988; Peplau et al., 1993, 1977). Cabral et al. (1998), Decker and Rosenfeld (1995), Kline et al. (1992), Longshore et al. (1998b), Montoya (1998), and Riehman (2000) support the perspective that drug-using women may be able to exert substantial control over sexual risk and underscore the importance of intervention that focuses on women’s self-efficacy for controlling sexual risk. On the other hand, in research by McCusker et al. (1993), St. Lawrence et al. (1998), and Wingood and DiClemente (1998), self-efficacy was relevant to sexual risk behavior (condom use) among men but not women. An important limitation of most of these studies is that the data were cross-sectional. It was therefore not possible to establish temporal order in the findings. The present study addresses this limitation in its use of a prospective design. With the ARRM as our conceptual framework, we tested pathways to sexual risk reduction among men and women separately. Predictor measures in the present study essentially replicated those in Longshore et al. (1998b). However, the prior study’s outcome measure was a latent variable based on condom use during vaginal, oral, or anal sex, while the outcome measure in the present study was a broader latent construct—sexual risk reduction—defined on the basis of number of sex partners and risky sexual behaviors such as engaging in sex while on drugs and having sex with high-risk partners. As discussed above, condom use is a direct form of control over sexual risk-taking. In most cases, a woman succeeds in reducing her sexual risk via condom use only if she persuades her male partner to cooperate during the “immediacy” of the sexual act (Hughes, 2000). The advantage of the broader sexual-risk construct is that it reflects variation in indirect control over sexual risk-taking; i.e., it is based on behaviors that do not require the

Psychosocial Antecedents cooperation of one’s partner and can be accomplished at some distance from the emotional and normative influences that often prevail in sexual situations. A further distinction between Longshore et al. (1998b) and the present study is that the sample for the latter is younger and mostly (91%) noninjection drug users. Thus, while both studies used the same theoretical base, the same predictors, and the same analytic approach, the distinctive contribution of this new study is to determine whether gender differences seen in Longshore et al. (1998b) can be crossvalidated in relation to sexual risk reduction more broadly construed and in a different at-risk population.

95 Table I. Sample Characteristics Men (n = 337)

Women (n = 263)

This study is based on a secondary analysis of data collected between 1991 and 1995 for an evaluation of Treatment Alternatives to Street Crime (TASC) programs in five U.S. cities. TASC programs assess the drug treatment needs of offenders in the criminal justice system, refer offenders to treatment and other services, and monitor their progress. Offenders may have been charged with a crime (related or unrelated to drug use) or may have been sent to TASC by a probation or parole officer who suspects that they have a drug problem. Evaluation design and findings are reported in Anglin et al. (1999).

Age (M years) Race/ethnicity African American (%) Non-Hispanic White (%) Hispanic (%) Other (%) Education
Participants

Note: Gender differences in employment were assessed with a Chisquare test; other differences, with analysis of variance. ∗p < .05.∗∗ p < .001.

METHOD

Data required for this analysis were complete for a sample of 263 heterosexually active HIVnegative women. (Identification of heterosexually active men and women was based on self-reported intercourse only with the opposite sex over a 30-day recall period at both intake and six-month followup.) To ensure comparable statistical power between the male and female subsamples, we selected every third heterosexually active, HIV-negative man from the pool of 1011 men and thus arrived at an analytic subsample of 337 men. Table I reports characteristics of each subsample. The racial/ethnic breakdown among men was 63% African Americans; 31% non-Hispanic Whites; and 7% others, mostly Hispanics. Their mean age was 27.6 years. Similarly, 63% of the women were African Americans and 32% non-Hispanic Whites; 5%, mostly Hispanics, reported another race/ethnicity. The mean age for

27.6

27.9

62.6 30.9 4.7 1.8

63.1 32.3 3.1 1.5

53.4 25.8 19.3 1.2

47.0 25.9 24.5 1.1

4.7 76.9 16.9

4.9 69.7 19.9

13.9 44.5 24.6 8.9 7.1 0.9

20.5 35.7 13.3 3.8 22.1 3.8

91.2 45.1 52.8 26.8

86.7 58.3 47.3 24.2

38.9 15.0 1.0 9.7

33.3 25.8 1.9 11.0

women was 27.9 years. Women were more likely than men to have never been employed or to have had sales/service jobs. Men reported more involvement in marijuana use; women, in use of crack cocaine. The groups were similar on indicators of criminal history. We did not include employment and drug use characteristics in the analysis because they are not part of the ARRM and were not related to sexual risk behavior outcomes in earlier research with this dataset (Longshore et al., 1998c). Measures Latent variables were developed to represent constructs in the AIDS Risk Reduction Model. Unless otherwise noted, measured items were Likert-type with response values ranging from

96 strongly disagree (1) to strongly agree (4). For the most part, items were the same as or very similar to those used in Longshore et al. (1998b). Three ARRM constructs—fear of AIDS, interpersonal cues to action, and perceived response efficacy of risk reduction strategies—were unavailable in this data set. None of them was a significant predictor of commitment to change or actual change among either men and women in Longshore et al. (1998b), so we do not believe their absence weakens this new analysis in any important way. In addition, response efficacy was very highly correlated with AIDS knowledge among both men and women in Longshore et al. (1998b); thus it had no independent explanatory value. All other constructs in the ARRM were available for analysis and are described below. Perceived infection risk was indicated by four items. Three were Likert-type items: “I’ve already done things that could have exposed me to AIDS,” “I never do anything that could give me AIDS” (reverse scored), and “I am not the kind of person who would get AIDS” (reverse scored). The fourth item measured the person’s perceived chance of contracting AIDS. The scale ranged from no chance (0%) to a sure chance (100%). We rescaled these values to 0–4. AIDS knowledge was measured as the sum of “strongly agree” responses to 10 items concerning routes of virus transmission. Items included, for example, “Only gay men get AIDS” (reverse scored) and “You cannot get AIDS by sharing food.” Self-efficacy was indicated by four items reflecting the person’s confidence in his or her ability to exert control over sexual risk-taking. Sample items are: “If I had sex with someone new, I would say there are things I will not do because of AIDS,” and “Using a condom is easy for me.” Safer sex commitment was indicated by one item: “In the month ahead, I may have sex with someone I’ve known only a short time” (reverse scored). Peer norms was measured with three items regarding behavioral norms for risk reduction. An example item was “My friends have really cut down the number of people they have sex with.” To indicate informational cues to action we used three yes/no items: having received face-to-face AIDS education from any source (e.g., drug treatment counselor) in the past 2 years, having ever been tested for HIV antibody, and having ever been encouraged to get tested. Sexual risk behavior, assessed at both baseline and 6-month follow-up, was a latent variable consisting of a behavioral risk index and number of sex

Longshore, Stein, and Chin partners. The risk index was the sum of responses indicating use of drugs immediately before or during sex, giving sex for drugs or drugs for sex, giving sex for money or money for sex, and having sex partners who were drug injectors. The recall period for these responses and number of sex partners was the past 30 days.

Analyses Our hypothesized model conformed to specifications in Catania et al. (1990) regarding the relevance of factors at each ARRM stage. Stage 1 predictors and the (concurrently measured) stage 1 end-point, perceived infection risk, were treated as exogenous. Stage 1 predictors included peer norms, AIDS knowledge, and informational cues to action. Baseline sexual risk behavior was entered as a control factor. Self-efficacy was treated as a predictor of safer sex commitment at stage 2, which in turn predicted sexual risk behavior at stage 3. Both self-efficacy and safer sex commitment were also tested as potential mediators of the effects of stage 1 factors. The model was tested separately for women and men. Predictors were allowed to correlate, and the initial path model included all possible predictive paths. Structural equation modeling is a popular method of testing predictive models with many endogenous variables. We used the EQS program (Bentler, in press) for this purpose. Goodness of fit of the models was evaluated through the robust comparative fit index (RCFI), the Satorra-Bentler scaled χ2 statistic, and the root mean squared error of approximation (RMSEA). The Satorra-Bentler χ2 statistic and the RCFI are preferable to maximum likelihood estimates when data are multivariately kurtose (Curran et al., 1996). Both gender subsamples were multivariately kurtose (men = 39.9; women = 37.4). The RCFI, which ranges from 0 to 1, indicates the improvement in fit of the hypothesized model compared to a model of complete independence among the measured variables. Values of 0.90 and higher are adequate and indicate that 90% or more of the covariation in the data is reproduced by the hypothesized model (recent research indicates that values closer to 0.95 are preferable; see Hu and Bentler, 1999). Finally, the RMSEA is a measure of fit per degrees of freedom, controlling for sample size, and values less than .06 indicate a good fit between the hypothesized model and the observed data (Hu and Bentler, 1999).

Psychosocial Antecedents In model development, we performed initial confirmatory factor analyses for each gender subsample. These analyses tested the adequacy of the proposed measurement model (factor structure) and proposed relationships among the latent constructs. Subsequent path models positioned the exogenous latent variables of perceived infection risk, peer norms, AIDS knowledge, informational cues to action, and baseline sexual risk behavior as predictors of self-efficacy and safer sex commitment. In turn, all factors predicted the outcome, follow-up sexual risk behavior. Nonsignificant paths and correlations among predictors were dropped until only significant paths and correlations remained. To assess gender differences, we performed multiple group comparisons for both the confirmatory factor analysis and path models in which various parameters were constrained to equality between the male and female subsamples. We also examined gender differences in latent construct means.

RESULTS Confirmatory Factor Analysis Table II presents factor loadings for the hypothesized latent factors in each gender subsample. All manifest variables loaded significantly on their hypothesized latent factors. After minimal model modifications, fit indices were acceptable for both groups. Among the men, Satorra-Bentler χ2 (141, N =337) = 190.17, RCFI = 0.93, RMSEA = 0.03. Among women, Satorra-Bentler χ2 (140, N=263) = 193.32, RCFI = 0.91, RMSEA = 0.04. Based on results from the LaGrange multiplier (LM) test (Chou and Bentler, 1990), three supplementary correlated error residuals were added to the model for women and two supplementary correlated error residuals were added for men. Table III shows correlations among the factors for men and women.

97 tor structure (measurement model) was constrained to equality between groups led to a relatively small but significant decrement in fit: χ2 (301, N = 600) = 418.36, RCFI = 0.92; χ2 difference = 34.83, 20 df, p ≤ .05. After a constraint was released on one indicator of self-efficacy (“I would never talk to a sex partner about using condoms”), the difference between the groups was nonsignificant (χ2 difference = 27.34, 19 df). Apart from response tendencies on the one item reported above, we concluded that relationships between the measured and latent variables were similar across gender and that the factors had the same meanings for both men and women. Thus, interpretation of the path analyses for men and women will not be complicated by any differences in the way men and women responded to the interview items or in the factor structure. In our most restrictive analysis, factor covariances along with the measurement model were constrained to be equal across groups. This tested whether there were gender differences in relationships among the latent variables. This model also fit the data reasonably well: Satorra-Bentler χ2 (328, N = 600) = 445.18, RCFI = 0.90. The adjusted χ2 difference between this model and the unconstrained model (62.01, 47 df) was not statistically significant. The adjusted χ2 difference between this model and the constrained measurement model (35.07, 28 df) also was not significant. Although the factor covariance structure as a whole was not different for men and women, the LM test indicated that three covariances were significantly different for men and women (critical value χ2 ≥ 3.84, 1 df, p = 0.05). These were the covariances between peer norms and AIDS knowledge (higher for men, χ2 = 5.24), self-efficacy and AIDS knowledge (higher for women, χ2 = 10.05), and safer sex commitment and AIDS knowledge (higher for women, χ2 = 4.84). In a comparison of latent means, we found that the women had higher means for perceived infection risk (z = 2.92), informational cues to action (z = 3.02), and safer sex commitment (z = 6.62).

Multiple Group Comparisons Path Analysis For sequential multiple group comparisons, a confirmatory factor analysis multiple-group model with no equality constraints served as a comparison for further models. This model had acceptable fit statistics: Satorra-Bentler χ2 (281, N = 600) = 383.53, RCFI = 0.92, RMSEA = 0.025. A model in which the fac-

Trimmed models depicting only the significant paths for each gender are shown in Figs. 1 and 2. Both models had excellent fit indexes. For women, Satorra-Bentler χ2 = 190.17, 156 df, RCFI = 0.94, RMSEA = 0.03. For men, Satorra-Bentler χ2 = 172.90, 157 df, RCFI = 0.98, RMSEA = 0.02. The

98

Longshore, Stein, and Chin Table II. Summary Statistics and Factor Loadings of Measured Variables by Gender Men (n = 337) Mean (SD)

Factor loadinga

Mean (SD)

Factor loadinga

2.5 (0.8) 2.3 (0.8) 2.6 (0.8) 0.7 (0.7)

0.61 0.72 0.44 0.52

2.6 (0.8) 2.4 (0.8) 2.9 (0.7) 0.8 (0.9)

0.64 0.71 0.59 0.57

2.6 (0.6) 2.6 (0.6) 2.5 (0.7) 3.0 (3.0)

0.33 0.61 0.67 1.00

2.6 (0.6) 2.5 (0.5) 2.5 (0.7) 3.3 (3.1)

0.38 0.41 0.63 1.00

0.2 (0.4) 0.1 (0.4) 0.1 (0.2)

0.65 0.85 0.44

0.3 (0.5) 0.2 (0.4) 0.1 (0.2)

0.66 0.74 0.36

0.3 (0.2) 0.5 (0.6)

0.51 0.67

0.3 (0.3) 0.5 (0.7)

0.62 0.67

3.1 (0.6) 2.8 (0.7) 2.9 (0.7) 3.1 (0.6) 2.7 (0.8)

0.50 0.43 0.37 0.42 1.00

3.2 (0.7) 2.8 (0.8) 3.1 (0.7) 3.1 (0.6) 3.0 (0.7)

0.61 0.33 0.52 0.48 1.00

0.3 (0.2) 0.4 (0.6)

0.66 0.82

0.3 (0.2) 0.4 (0.7)

0.39 0.92

I. Perceived infection risk Already done things Never do anything (R) Not the kind of person (R) % chance II. Peer norms Friends cut down Friends not using condoms (R) Friends not careful (R) III. AIDS transmission knowledge IV. Cues to action Given AIDS prevention info Given condoms, latex protection Referral for HIV test V. Sexual risk behavior–baseline Number of partnersb Risk behavior index VI. Perceived self-efficacy Can say won’t do Condom use easy No sex without condom Never talk (R) VII. Commitment to sexual risk reduction VIII. Sexual risk behavior–follow-up Number of partnersb Risk behavior index a All

Women (n = 263)

factor loadings significant p ≤ .001. transformed due to skewness and kurtosis.

b Log

model for men explained 9% of the variance in sexual risk behavior; for women, 18%. The initial saturated path models for the men and women were compared with multiple-group models. With the factor loadings and regressions held as invariant, there was a significant decrement in fit from a baseline unrestricted model (adjusted χ2 difference = 58.16/35 df). Regressions that were significantly different for men and women included the path from

perceived infection risk to self-efficacy (χ2 = 4.78; −.077 for men and −0.46 for women) and the path from AIDS knowledge to self-efficacy (χ2 = 13.55; 0.42 for men and 0.77 for women). However, both paths were significant for men and women and are included in the final trimmed models. First, regarding the ARRM’s stage 2 end-point, we found that safer sex commitment was predicted by higher self-efficacy and lower baseline sexual risk

Table III. Correlations Among Latent Variables Variable I. Perceived risk risk II. Peer norms III. AIDS knowledge IV. Cues to action V. Baseline sex behavior VI. Self-efficacy VII. Safer sex commitment VIII. Follow-up sex behavior

I — −0.14 0.22∗∗∗ 0.11 0.17∗ −0.31∗∗ −0.15∗ 0.22∗∗∗

Note: Men above the diagonal, women below. ∗ p ≤ 0.05. ∗∗ p ≤ 0.01.∗∗∗ p ≤ 0.001.

II

III

−0.19∗ — −0.05∗∗∗ 0.15 −0.11 0.40∗∗∗ 0.09 −0.05

0.14∗ 0.12 — 0.10 −0.08 0.69∗∗∗ 0.31∗∗∗ −0.04

IV

V

VI

VII

0.04 0.14∗ 0.06 — 0.14 0.17 0.04 0.11

0.23∗∗

−0.64∗∗∗

−0.11 −0.03 0.09 −.07 −0.31∗∗∗ 0.06 — −0.26∗∗∗

−0.35∗∗∗ 0.02 0.08 — 0.03 −0.31∗∗∗ 0.37∗∗∗

0.32∗∗∗ 0.29∗∗∗ 0.17∗ −0.13 — 0.28∗∗∗ −0.01

VIII 0.20∗∗ −0.01 −0.02 −0.00 0.27∗∗ −0.06 −0.12∗ —

Psychosocial Antecedents

99

Fig. 1. Significant regression paths predicting HIV risk behaviors among 337 substance-using men. Large circles represent latent variables, rectangle represents a measured variable. Regression coefficients are standardized (∗ p ≤ 0.05,∗∗ p ≤ 0.01,∗∗∗ p ≤ 0.001).

behavior among the women. Safer sex commitment was predicted by greater AIDS knowledge and lower baseline sexual risk behavior among the men. Thus, as in Longshore et al. (1998b), persons of either gender who engaged in more sexual risk-taking at baseline were less likely to express a commitment to sexual risk reduction in the future. Self-efficacy was higher among both men and women with lower perceived infection risk and greater AIDS knowledge. For women, positive peer norms also predicted greater self-efficacy. For men, higher scores on informational cues to action predicted greater selfefficacy. Second, regarding the stage 3 end-point, we found modest but significant negative paths from safer sex commitment to follow-up sexual risk behavior for both men and women. Thus, with controls for baseline sexual risk behavior, subsequent risktaking was lower among persons who expressed a stronger commitment to practice safer sex. This relationship was substantively the same across gender. Among women, no other psychosocial factors predicted follow-up sexual risk behavior, although base-

line behavior was a strong predictor. Among men, perceived infection risk positively predicted Followup sexual risk behavior. The latent variable for baseline sexual risk behavior did not predict men’s sexual risk behavior at follow-up. However, the separate indicator of behavioral risk was a significant predictor of the more general sexual risk behavior factor, whereas number of partners did not predict variation in the more general factor. DISCUSSION Most studies of sexual risk reduction among drug users have been cross-sectional and have not tested correlates of risk reduction in a theory-based multivariate context. In contrast, this study conducted a theory-based prospective test of predictors of sexual risk reduction among drug-using men and women. Importantly, our outcome measure reflected variation in indirect control over sexual risk-taking, i.e., it was based on behaviors that do not require cooperation by one’s partner and can be accomplished outside the immediacy of sexual relations.

100

Longshore, Stein, and Chin

Fig. 2. Significant regression paths predicting HIV risk behaviors among 263 substance-using women. Large circles represent latent variables. Regression coefficients are standardized (∗ p ≤ 0.05,∗∗ p ≤ 0.01,∗∗∗ p ≤ 0.001).

Findings demonstrated the usefulness of the ARRM as a theoretical “test bed” for comparing psychosocial antecedents of sexual risk reduction across gender. Alternative models might have been equally plausible (MacCallum et al., 1993). However, structural equation modeling can provide more persuasive evidence of causality than other non-experimental methods, especially when, as was the case here, the proposed model is based on an explicit theory and provides a good fit to the data. Specifically, we were able to derive the hypothesized latent factors in both gender groups. Fit indices were acceptable for both groups. Thus, interpretation of the model for men and women was not complicated by any differences in the way men and women responded to the interview items or in factor structure. Future work is needed to determine whether the ARRM or other psychosocial models can be used for this purpose in populations other than heterosexual men and women involved in illegal drug use. Measures of baseline sexual risk behavior (the latent factor for women and the behavioral risk index for men) were a significant predictor of follow-up sexual risk behavior. This finding underscores the

importance of accounting for baseline behavior in psychosocial models like the ARRM. It has been shown repeatedly that measures of past behavior are related to future behavior “over and above measures of attitude or intention” (Verplanken et al., 1998, p. 113). The causal significance of this relationship remains at issue, however. Some have argued that past behavior is merely a proxy for unmeasured causal factors and has no theoretical significance. According to Ajzen (1987, p. 41), “it serves no useful purpose to include past behavior . . . in causal models of human action.” Others have argued that past behavior does have theoretical significance as routinized conduct or habit and that models of human behavior should incorporate the psychosocial or contextual factors that lead to habit formation (Aarts et al., 1998). This issue is beyond our scope, but we can point to one relevant finding from our analysis. For both gender groups, past risk behavior predicted not only future risk behavior but also lower safer sex commitment. This finding is inconsistent with an argument that habitual conduct occurs without any process of conscious reasoning or planning and suggests instead that the influence of habit on future behavior may be

Psychosocial Antecedents mediated by intentions, even if the person no longer accesses the background cognitions that underlie those intentions before making a commitment to act (see Ouellette and Wood, 1998). The model explained only 9% of the variance in men’s sexual risk behavior and 18% of the variance in women’s. It is clear that factors in addition to those in the ARRM framework must be considered in developing a more comprehensive understanding of sexual risk behavior in each gender. At the same time, remaining within the ARRM framework, we uncovered gender-related similarities and differences in the psychosocial process of sexual risk reduction. Safer sex commitment led to reduction in sexual risk behavior in both gender groups. Women scored higher than men on safer sex commitment, but the effect of this factor was similar for both groups. It should be noted that our measure of safer sex commitment was based on a single indicator, for which the degree of reliability was unknown. We would be more confident in the results obtained for safer sex commitment if its measure had demonstrably stronger psychometrics and covered a wider range of relevant behavioral intentions. In Longshore et al. (1998b), paths to condom use from safer sex commitment (a two-item latent variable) were somewhat stronger for both men and women, perhaps because the measure of safer sex commitment in that study was more reliable. On the other hand, the magnitude of the path coefficients was similar for both genders in each study. Accordingly, while less confident that we have captured the true magnitude of the relationship between safer sex commitment and sexual risk reduction, we are confident of the finding that the relationship exists for both men and women and that it does not differ dramatically between them. The path from self-efficacy to safer sex commitment was significant for women but not men. Moreover, both of these factors were related to AIDS knowledge more closely for women than for men. These findings suggest that women realize the importance of sexual risk avoidance and are aware that their sexual risk taking is not entirely under volitional control (see also Montoya, 1998). Other studies have found a relationship between self-efficacy and sexual risk behavior among men but not women (St. Lawrence et al., 1998; Wingood and DiClemente, 1998). Those studies focused on direct control of sexual risk via condom use and are therefore, less persuasive regarding behaviors reflecting indirect control over sexual risk. Moreover, most prior

101 studies were cross-sectional, leaving causal direction unclear. The prospective study by Longshore et al. (1998b) found that self-efficacy predicted subsequent condom use among women and not men. This finding presumably reflects the fact that women’s control over condom use is significantly constrained, whereas men’s control over condom use is not. In the present study, self-efficacy predicted women’s commitment to safer sex but did not predict the behavioral outcome, which was designed to reflect variation in indirect control over sexual risk-taking. Thus, women may feel no less able than men to manage aspects of sexual risk reduction that do not require a partner’s cooperation. However, self-efficacy remains relevant for women in a way that does not pertain to men; it influences women’s commitment to safer sex. As Ajzen (1991) has argued, self-efficacy not only reflects perceived constraints on behavior but also has motivational properties. People who believe they are constrained from engaging in a behavior are unlikely to form strong intentions to engage in it even if they are otherwise disposed to do so. In short, while selfefficacy may influence women’s behavior when the issue is direct control over sexual risk, its relevance for women’s indirect control over sexual risk appears to be motivational; its influence is on behavioral intentions. Finally, in the meta-analysis by Sheeran and Abraham (1999), condom use norms were more strongly related to condom use among women than among men. We found that norms affected selfefficacy for risk reduction broadly construed (not specifically self-efficacy for condom use) among women but not men—a finding consistent with Sheeran and Abraham (1999) in suggesting that peer norms matter more for women than for men when the issue is sexual risk-taking. The importance of the social context for women has been amply illustrated in a number of research areas. For example, across virtually all cultures, women have been found to be more collectivistic than men (Triandis, 1993, 1995); they conceive of their social relationships in a more interdependent manner. Thus it is not surprising that a woman’s perception of what her peers are doing influences her perceived control over what she herself is doing. Gender differences in the pathways to sexual risk reduction imply that intervention to promote sexual risk reduction should be designed to address such differences. Commitment to safer sex was related to sexual risk reduction among drug-using

102 men. Intervention with this population should therefore target safer sex commitment. Examples of commitment-building exercises include counterattitudinal advocacy, declarations or written statements of commitment, and behavior contracts (Strecher et al., 1995). Interventions promoting cognitive accessibility of risk-reduction intentions are another potentially effective strategy (Norris and Ford, 1995). In addition, although drug users’ knowledge regarding virus transmission routes is generally high (Silbersiepe and Hardy, 1997), the path from AIDS knowledge to safer sex commitment—weaker for men than for women—suggests that men’s intervention ought to target their AIDS knowledge. It may be important to note that scoring high on our AIDS knowledge measure required “strongly agree” responses. “Agree” responses did not count toward a high score. Thus, the measure may have tapped not only one’s knowledge of virus transmission routes but also one’s degree of confidence in that knowledge. Intervention for men should therefore provide definitive information regarding routes of transmission and seek to raise men’s beliefs that this information is reliable. Finally, men’s sexual behavior at follow-up was more risky if their perceived risk of infection at baseline was higher. This path can be read in either of two ways. It may mean that risky sexual behavior is less likely to decline among men with a fatalistic attitude toward their infection risk. Alternatively, it may signify greater stability in risk behavior among men with a “taste for risk,” i.e., men who enjoy and seek out high-risk activity. The connection between perceived risk and subsequent behavior would, in that case, represent a realistic appraisal of the risk associated with one’s ongoing sexual lifestyle. These two interpretations are relevant to intervention in divergent ways. Thus, we cannot derive any clear implication. The difficulty in reading effects of perceived infection risk has been widely noted (e.g., Kowalewski et al., 1997). More work is needed before intervention strategies can effectively address perceived infection risk. For drug-using women, safer sex commitment was related to sexual risk reduction. Women’s interventions should seek to raise commitment to safer sex by strategies such as those described above for men. But they should also explicitly address selfefficacy. Self-efficacy for exerting direct control over sexual risk, i.e., for persuading men to use condoms, appears to be a pathway to risk reduction for women (Longshore et al., 1998b). However, the present study indicates that self-efficacy as a motivator for women’s

Longshore, Stein, and Chin indirect control over sexual risk behavior is another such pathway. It is important to emphasize indirect as well as direct control in women’s intervention because the former (i.e., condom use) is often problematic, requiring assertive communication, negotiation, and perhaps confrontation with male partners (Amaro, 1995; Chin, 1999). Attempts to exert direct control may therefore come into conflict with values of trust and intimacy and with gender roles that many women consider appropriate. When there is a marked power disparity between the couple, direct control is particularly problematic and perhaps even dangerous (Wingood and DiClemente, 1998). Interventions that over-emphasize the use of condoms as a risk reduction strategy may, for women unable or unwilling to use them, set up a false or unpalatable choice between capitulation to unprotected sex or abstinence from any sex. Moreover, an overemphasis on use of condoms is unwise if it means devoting inadequate time to risk reduction strategies that women can control more readily, e.g., having sex with fewer partners, avoiding sex with men known or thought to be engaged in high-risk behavior such as drug injection, abstaining from drug or alcohol use during sex, and slowing the pace of sexual intimacy. Self-efficacy was influenced by women’s perceptions regarding peer norms for sexual risk reduction. This finding suggests that interventions with women should attempt to increase peer support for exerting control over sexual risk. Use of a small-group format, led by an HIV educator, is probably most appropriate in such interventions. The educator can raise various strategies for risk reduction, such as condom use and avoidance of sex. Participants as a group, on the basis of their experience, can identify the pros and cons of these strategies. The sharing of successful personal experience would serve to highlight the common concerns of participants and might provide a major boost to their sense of self-efficacy. The most effective strategy may be one that addresses both self-efficacy and peer norms. Traditional conceptions of individual control or agency have been criticized for their neglect of more collectivistic bases of control (Van Uchelen, 2000). The concept of collective efficacy arose from the recognition that feelings of personal efficacy can be derived from one’s relationship with a collective—defined as a small group of identifiable individuals, as a specific neighborhood or community, or, more abstractly, as a group of people who share the same behavioral norms (Bandura, 1997). To the extent that one feels connected to such a collective, a higher sense of

Psychosocial Antecedents personal efficacy may be derived. Because women are more collectivistic than men (Triandis, 1995), a more interdependent approach to intervention may be more effective with women. ACKNOWLEDGMENTS Data collection and analysis were supported by grant numbers N01-DA-1-8408 and P01-DA01070 from the National Institute on Drug Abuse. The authors wish to thank Evelyn Calderon, Amber Horning, Gisele Pham, and Elizabeth Teshome for assistance with manuscript preparation. REFERENCES Aarts, H., Verplanken, B., and Knippenberg, A. (1998). Predicting behavior from actions in the past: Repeated decision making or a matter of habit? Journal of Applied Social Psychology, 28, 1355–1374. Ajzen, I. (1987). Attitudes, traits, and actions: Dispositional prediction of behavior in personality and social psychology. In L. Berkowitz (Ed.), Advances in experimental social psychology, Vol. 2 (pp. 1–63). CA: Academic Press. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179–211. Ajzen, I., and Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice Hall. Amaro, H. (1995). Love, sex, and power. American Psychologist, 50, 437–447. Anglin, M. D., Longshore, D., and Turner, S. (1999). Treatment alternatives to street crime: An evaluation of five programs. Criminal Justice and Behavior, 26, 168–195. Bandura, A. (1997). Self-efficacy: The essence of control. New York: W. H. Freeman. Bentler, P. M. (in press). EQS 6 structural equations program manual. Encino, CA: Multivariate Software, Inc. Cabral, R. J., Pulley, L., Artz, L. M., Brill, I., and Macaluso, M. (1998). Women at risk of HIV/STD: The importance of male partners as barriers to condom use. AIDS & Behavior, 2, 75– 85. Catania, J. A., Kegeles, S. M., and Coates, T. J. (1990). Towards an understanding of risk behavior: An AIDS risk reduction model (ARRM). Health Education Quarterly, 17, 52–92. Chin, D. (1999). HIV-related sexual risk assessment among Asian/Pacific Islander American women: An inductive model. Social Science & Medicine, 49, 241–251. Chou, C.-P., and Bentler, P. M. (1990). Model modification in covariance structure modeling: A comparison among likelihood ratio, Lagrange Multiplier, and Wald tests. Multivariate Behavioral Research, 25, 115–136. Corby, N. H., Jamner, M. S., and Wolitski, R. (1996). Using the theory of planned behavior to predict intentions to use condoms among male and female injecting drug users. Journal of Applied Social Psychology, 26, 52–75. Curran, P. J., West, S. G., and Finch, J. F. (1996). The robustness of test statistics to nonnormality and specification error in confirmatory factor analysis. Psychological Methods, 1, 16–29. Decker, S., and Rosenfeld, R. (1995). “My wife is married and so is my girlfriend”: Adaptations to the threat of AIDS in an arrestee population. Crime and Delinquency, 41, 37–53. DiClemente, R. J., and Peterson, J. L. (1994). Changing HIV/AIDS risk behaviors: The role of behavioral interven-

103 tions. In R. J. DiClemente, and J. L. Peterson (Eds.), Preventing AIDS: Theories and methods of behavioral interventions (pp. 1–4). New York: Plenum. Fishbein, M., Bandura, A., Triandis, H. C., Kanfer, F., Becker, M. H., Middlestadt, S. E., and Eichler, A. (1991). Factors influencing behavior and behavior change: Final report— Theorists’ Workshop. Rockville, MD: National Institute of Mental Health. Fisher, W. A., and Gray, J. (1988). Erotophobia–erotophilia and sexual behavior during pregnancy and postpartum. Journal of Sex Research, 25, 379–396. Fisher, W. A., Fisher, J. D., and Rye, B. J. (1995). Understanding and promoting AIDS-preventive behavior: Insights from the theory of reasoned action. Health Psychology, 14, 255–264. Flores-Ortiz, Y. G. (1994). The role of cultural and gender values in alcohol use patterns among Chicana/Latina high school and university students: Implications for AIDS prevention. International Journal of the Addictions, 19, 1149–1171. Hobfoll, S. E., Jackson, A. P., Lavin, J., Britton, P. J., and Shepherd, J. B. (1994). Women’s barriers to safer sex. Psychology and Health, 9, 233–252. Hu, L.-T., and Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55. Hughes, R. (2000). “Friendships are a big part of it”: Social relationships, social distance, and HIV risks. Substance Use & Misuse, 35, 1149–1176. Kline, A., Kline, E., and Oken, E. (1992). Minority women and sexual choice in the age of AIDS. Social Science & Medicine, 34, 447–457. Kok, G., de Vries, H., Mudde, A. N., and Strecher, V. J. (1991). Planned health education and the role of self-efficacy: Dutch research. Health Education Research, 6, 231–238. Kowalewski, M. R., Henson, K. D., and Longshore, D. (1997). Rethinking perceived risk and health behavior: A critical review of HIV prevention research. Health Education & Behavior, 24, 313–325. Lear, D. (1995). Sexual communication in the age of AIDS: The construction of risk and trust among young adults. Social Science & Medicine, 41, 1311–1323. Longshore, D. (1992). AIDS education for drug users: Existing research and new directions. Journal of Drug Issues, 22, 1–16. Longshore, D., Anglin, M. D., Annon, K., and Hsieh, S. (1998a). Long-term trends in self-reported risk behavior: Injection drug users in Los Angeles, 1987–1995. Journal of Acquired Immune Deficiency Syndromes, 18, 64–72. Longshore, D., Danila, B., Hsieh, S., and Anglin, M. D. (1994). Reducing HIV infection risk among injection drug users: The effect of methadone maintenance on number of sex partners. The International Journal of the Addictions, 29, 741–758. Longshore, D., Stein, J. A., and Anglin, M. D. (1996). Ethnic differences in the psychosocial antecedents of needle/syringe disinfection by drug users. Drug & Alcohol Dependence, 4, 183–196. Longshore, D., Stein, J. A., and Anglin, M. D. (1997). Psychosocial antecedents of needle/syringe disinfection: A theory-based prospective analysis. AIDS Education & Prevention, 9, 442– 457. Longshore, D., Stein, J. A., Kowalewski, M., and Anglin, M. D. (1998b). Psychosocial antecedents of unprotected sex by drug-using men and women. AIDS & Behavior, 2, 293– 306. Longshore, D., Turner, S., and Anglin, M. D. (1998c). Effects of case management on drug users’ risky sex. The Prison Journal, 78, 6–30. MacCallum, R. C., Wegener, D. T., Uchino, B. N., and Fabrigar, L. R. (1993). The problem of equivalent models in applications of covariance structure analysis. Psychological Bulletin, 114, 185–199.

104 Malow, R. M., Corrigan, S. A., Cunningham, S. C., West, J. A., and Pena, J. M. (1993). Psychosocial factors associated with condom use among African-American drug abusers in treatment. AIDS Education & Prevention, 5, 244–253. Malow, R. M., and Ireland, S. J. (1996). HIV risk correlates among non-injection cocaine dependent men in treatment. AIDS Education & Prevention, 8, 226–235. Malow, R. M., West, J. A., Corrigan, S. A., Pena, J. M., and Cunningham, S. C. (1994). Outcome of psychoeducation for HIV risk reduction. AIDS Education & Prevention, 6, 113– 125. McCusker, J., Stoddard, A. M., Zapka, J. G., and Zorn, M. (1993). Use of condoms by heterosexually active drug abusers before and after AIDS education. Sexually Transmitted Diseases, 20, 81–88. Montoya, I. (1998). Social network ties, self-efficacy, and condom use among women who use crack cocaine: A pilot study. Substance Use & Misuse, 33, 2049–2073. Norris, A. E., and Ford, K. (1995). Condom use by low-income African American and Hispanic youth with a well-known partner: Integrating the health belief model, theory of reasoned action, and the construct accessibility model. Journal of Applied Social Psychology, 25, 1801–1830. Ouellette, J. A., and Wood, W. (1998). Habit and intention in everyday life: The multiple processes by which past behavior predicts future behavior. Psychological Bulletin, 124, 54–74. Peplau, L. A., Hill, C. T., and Rubin, Z. (1993). Sex role attitudes in dating and marriage: A 15-year follow-up of the Boston Couples Study. Journal of Social Issues, 49, 31–52. Peplau, L. A., Rubin, Z., and Hill, C. T. (1977). Sexual intimacy in dating relationships. Journal of Social Issues, 33, 86–109. Rhodes, T., Stimson, G. V., and Quirk, A. (1996). Sex, drugs, intervention, and research: From the individual to the social. Substance Use & Misuse 31, 375–407. Riehman, K. S. (2000, June). How couples’ power relations affect treatment retention and outcomes: Preliminary data from a pilot study. Paper presented at the Annual Meeting of the RAND Drug Policy Research Center Advisory Board, Santa Monica, CA. Robles, R., Cancel, L. I., Colon, H. M., Matos, T. D., Freeman, D. H., and Sahai, H. (1995). Prospective effects of perceived risk of developing HIV/AIDS on risk behaviors among injection drug users in Puerto Rico. Addiction, 90, 1105–1111.

Longshore, Stein, and Chin Rosenstock, I. M. (1990). The health belief model: Explaining health behavior through expectancies. In K. Glanz, F. M. Lewis, and B. K. Rimer (Eds.), Health behavior and education (pp. 39–62). San Francisco: Jossey-Bass. Satorra, A., and Bentler, P. M. (2001). A scaled difference chisquare test statistic for moment structure analysis. Psychometrika, 66, 507–514. St. Lawrence, J. S., Brasfield, T. L., Eldridge, G. D., Little, C. E., Reitman, R., and Shelby, M. C. (1998). Factors influencing condom use among African-American women: implications for risk reduction interventions. American Journal of Community Psychology, 26, 7–28. Sheeran, P., and Abraham, C. (1999). Psychosocial correlates of heterosexual condom use: A meta-analysis. Psychological Bulletin, 125, 90–132. Silbersiepe, K. A., and Hardy, A. M. (1997). AIDS knowledge and risk perception of cocaine and crack users in a national household survey. AIDS Education & Prevention, 9, 460– 471. Stephens, R. C., Kwiatkowski, C. F., and Booth, R. E. (2000). The impact of the NIDA cooperative agreement programs on HIV risk among crack and injection drug users. Advances in Medical Sociology, 7, 241–259. Strecher, V. J., Seijts, G. H., Kok, G. J., Latham, G. P., Glasgow, R., DeVellis, B., Meertens, R. M., and Bulger, D. W. (1995). Goal setting as a strategy for health behavior change. Health Education Quarterly, 22, 190–200. Triandis, H. (1993). Collectivism and individualism as cultural syndromes. Cross-Cultural Research, 27, 155–180. Triandis, H. (1995). Individualism and collectivism. Boulder: Westview Press. Van Uchelen, C. (2000). Individualism, collectivism, and community psychology. In J. Rappaport, and E. Seidman (Eds.), Handbook of community psychology (pp. 65–78). New York: Plenum. Verplanken, B., Aarts, H., Knippenberg, A., and Moonen, A. (1998). Habit versus planned behaviour: A field experiment. British Journal of Social Psychology, 37, 111– 128. Wingood, G. M., and DiClemente, R. J. (1998). Partner influences and gender-related factors associated with noncondom use among young adult African American women. American Journal of Community Psychology, 26, 29–51.