Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2017 Apr 1.
Published in final edited form as: Psychol Violence. 2016 Apr;6(2):271–279. doi: 10.1037/a0039411

Longitudinal Change in Women's Sexual Victimization Experiences as a Function of Alcohol Consumption and Sexual Victimization History: A Latent Transition Analysis

Amanda E B Bryan a, Jeanette Norris a, Devon Alisa Abdallah a, Cynthia A Stappenbeck b, Diane M Morrison c, Kelly C Davis c, William H George d, Cinnamon L Danube a, Tina Zawacki e
PMCID: PMC4873161  NIHMSID: NIHMS690959  PMID: 27213101

Abstract

Objective

Women's alcohol consumption and vulnerability to sexual victimization (SV) are linked, but findings regarding the nature and direction of the association are mixed. Some studies have found support for the self-medication hypothesis (i.e., victimized women drink more to alleviate SV-related distress); others have supported routine activity theory (i.e., drinking increases SV vulnerability). In this study, we aimed to clarify the interplay between women's prior SV, typical drinking, and SV experiences prospectively over one year.

Method

Participants (N = 530) completed a baseline survey and weekly follow-up surveys across Months 3, 6, 9, and 12.

Results

Latent class analysis (LCA) suggested that women could be classified as victimized or non-victimized at each assessment month; 28% of participants were classified as victimized at one or more assessment months. Latent transition analysis (LTA) revealed that childhood sexual abuse and adult SV history each predicted greater likelihood of being victimized during the year. Typical drinking during a given assessment month was associated with (1) greater likelihood of victimized status at that assessment month and (2) greater likelihood of having transitioned into (or remained in) the victimized status since the previous assessment month. Furthermore, victimized status at a given assessment month predicted a higher quantity of subsequent drinking.

Conclusion

These findings indicate a reciprocal relationship between typical drinking and SV, supporting both the self-medication hypothesis and routine activity theory, and suggesting that hazardous drinking levels may be one important target for both SV vulnerability reduction and interventions for women who have been sexually victimized.

Keywords: sexual victimization, alcohol consumption, latent class analysis, latent transition analysis

Sexual victimization (SV), defined as unwanted sexual contact, verbal sexual coercion, or attempted or completed rape, is prevalent among young adult women. Over half report some form of SV since age 14, and up to a quarter report attempted or completed rape (Black et al., 2011; Elliott, Mok, & Briere, 2004). Women who experience SV are at increased risk for several adverse outcomes, including substance abuse, post-traumatic stress disorder (PTSD), other anxiety and mood problems, and chronic health problems (Santaularia, Johnson, Hard, Haskett, Welsh, & Faseru, 2014; Zinzow, Resnick, McCauley, Amstadter, Ruggiero, & Kilpatrick, 2012). Characteristics associated with increased SV vulnerability include young age, being single, a high number of sexual partners, alcohol use, and a history of prior SV (e.g., Elliott et al., 2004; Kaysen, Neighbors, Martell, Fossos, & Larimer, 2006; Testa, VanZile-Tamsen, & Livingston, 2007). Few studies, however, have examined whether and how such background risk factors may work in conjunction with women's ongoing experiences and behaviors to predict future SV. In the current study, we aimed to understand SV's associations with background and ongoing factors (age, prior SV, and typical alcohol use) and to reconcile contrasting theories regarding the mechanisms by which alcohol consumption and SV may influence one other.

Prior SV as a vulnerability factor for adult re-victimization

Re-victimization rates among SV victims are high. In general, large prospective studies have shown that women with previous SV have significantly higher vulnerability to subsequent victimization than do non-victims (Classen, Palesh, & Aggarwal, 2005). For example, Najdowski and Ullman (2011) prospectively surveyed 555 SV victims and found that 45% were re-victimized over the following year. Research has also demonstrated that childhood sexual abuse (CSA) history increases vulnerability to adult victimization. A retrospective survey of a nationally representative sample of 8,000 women (Desai, Arias, Thompson, & Besile, 2002) showed that those with a CSA history were nearly twice as likely to have experienced adult SV as those who had not. Prospective surveys have similarly demonstrated that women with a CSA history have higher rates of SV (e.g., Miron & Orcutt, 2014; Siegel & Williams, 2003).

Alcohol consumption as a vulnerability factor for SV

Associations between women's alcohol consumption and SV are complex, especially because typical drinking habits may confer vulnerability differently than event-level consumption. Here we discuss research examining how women's typical drinking habits are associated with the likelihood of becoming sexually victimized over discrete periods of time. We do not intend to imply that women who drink are at fault for being victimized nor that it is victims' responsibility to prevent assault; the responsibility lies firmly with the perpetrator. The risk, however, is borne by the potential victim and we assert that preventive efforts can be valuably informed by understanding how some patterns of alcohol use may be associated with increased vulnerability. Perpetrators bear responsibility for any effects of their own alcohol use on their offending behavior, a topic that is outside the scope of this investigation.

Cross-sectionally, women's typical drinking has been positively associated with SV (e.g., Kilpatrick, Edmunds, & Seymour, 1992; Koss & Dinero, 1989; Siegel & Williams, 2003). These studies, however, do not provide information about the direction of the association. Does alcohol consumption lead to higher vulnerability to SV? Do SV experiences lead women to increase their drinking? Each of these directional hypotheses can be generated from well-supported theories. For example, routine activity theory (Cohen & Felson, 1979), which posits that crime stems from the intersection of multiple contextual factors including behavioral routines, has been used as a framework for explaining the association between women's typical substance use and likelihood of being victimized (Mustaine & Tewksbury, 1999). Conversely, the self-medication hypothesis (Khantzian, 2003)postulates that traumatized individuals drink to relieve psychological distress and has received support in studies of women with SV histories (e.g., Kaysen et al., 2014; Ullman, Relyea, Peter-Hagene, & Vasquez, 2013). In reality, each of these explanations probably plays a role in explaining the relationship between alcohol consumption and SV (Abbey, 2002). For example, there may be a reciprocal relationship between drinking and SV such that drinking confers vulnerability to SV, which in turn confers additional risk for increased drinking.

Prospective longitudinal studies have yielded mixed findings regarding the direction of the alcohol-SV link. Kaysen et al. (2006) found that alcohol-facilitated rape (i.e., penetrative sexual assault that occurred when the woman was too intoxicated to provide consent) was associated with higher alcohol use in years prior and subsequent to an assault. In the largest study to date, Kilpatrick and colleagues (1997) followed a nationally representative sample of 3,000 women over two years and found that SV predicted alcohol abuse but not vice versa. These findings support the self-medication hypothesis but not routine activity theory. However, in a study of first-year college women, Mouilso and Fisher (2012) found the opposite: Consistent with routine activity theory, frequency of drinking and binge drinking predicted subsequent SV, whereas experiencing SV did not predict subsequent drinking. Similarly, Testa and Livingston (2000) surveyed young female sexually active drinkers at two times 12 months apart and found that alcohol problems (but not typical drinking) prospectively predicted SV after controlling for prior SV, but SV did not predict subsequent increases in drinking or alcohol problems. Other prospective studies have found that alcohol use predicted SV but not vice versa (e.g., Gidycz et al., 2007; Messman-Moore, Ward, Zerubavel, Chandley, & Barton, 2015; Parks, Hseih, Taggart, & Bradizza, 2014).

Clearly then, typical drinking is associated with SV vulnerability, but the nature of the association is not yet completely understood. This is in part because of several gaps in the literature that reflect challenges in modeling SV vulnerability. First, retrospective reports of regularly occurring behaviors such as drinking may be unreliable when they are gathered over a long time period (e.g., the past year), yet it is resource-intensive to administer more frequent surveys. Second, definitions of SV vary across studies in ways that may significantly affect their tendencies to predict or be predicted by other factors such as drinking. A researcher may select a more or less inclusive SV definition (e.g., including or excluding verbal coercion); furthermore, SV may be defined categorically (e.g., present or not present; completed rape, attempted rape, or no rape) or continuously (e.g., Davis et al.'s [2014] scoring method for the Sexual Experiences Survey). Third, traditional statistical methods (e.g., stepwise multiple regression) limit researchers' ability to model reciprocal influences. In the current study, we sought to overcome these challenges and shed new light into the directional characteristics of the alcohol-SV link by administering frequent assessments, defining SV according to the identified underlying (latent) structure of the data, and implementing a data analytic strategy that models reciprocal associations across time.

Study overview and hypotheses

We employed a prospective longitudinal design with a baseline survey and four clusters of consecutive weekly follow-up surveys distributed every three months over one year. The study design featured several strengths over previous studies and provided the opportunity to clarify conflicting findings regarding the direction of the alcohol-SV relationship. We used frequent surveys to reduce concerns regarding retrospection and to establish clear temporal links between recent drinking habits and SV experiences. The number of assessment points allowed us to examine whether those links were stable across time, providing a strong foundation for drawing conclusions about whether the associations we observed were truly reciprocal.

To provide a view of the interplay between SV and its vulnerability factors, we combined person- and variable-centered approaches to data analysis (Laurson & Hoff, 2006). Person-centered analyses focus on identifying subgroups of similar individuals; there is evidence that distinct profiles of SV vulnerability can be identified (e.g., French, Bi, Latimore, Klemp, & Butler, 2014; Macy, Nurius, & Norris, 2007). To that end, we used latent class analysis (LCA) and latent transition analysis (LTA) to identify underlying classes of observations (individuals) based on observed responses and to model individuals' movement between classes across time. LTA has not appeared previously in this literature to examine the interplay between women's drinking and SV. We proceeded to incorporate a variable-centered approach to examine SV classifications and transitions as a function of three covariates: age, drinking and prior SV.

Regarding continuity of SV experiences, we hypothesized that (1) SV experiences at each assessment would positively predict SV experiences at the subsequent assessment month, (2) history of CSA would positively predict SV experiences at each assessment month, and (3) history of SV (between age 14 and study enrollment) would positively predict SV experiences at each assessment month. Regarding continuity of alcohol use, we hypothesized that (4) baseline typical alcohol use would positively predict reported drinking quantity during the subsequent assessment month and (5) drinking quantity during each assessment month would positively predict drinking quantity during the subsequent assessment month. We also hypothesized that (6) reported drinking quantity would be associated with greater probability of being victimized at each assessment month and (7) reported drinking quantity would be associated with greater probability of transitioning from non-victimized to victimized status at each assessment month. Finally, we hypothesized that (8) SV experiences at each assessment would positively predict reported drinking quantity at the subsequent assessment month.

Method

Participants

Participants (N = 530) were female drinkers aged 18-30 years recruited from the community for a longitudinal study about alcohol use and social interactions with men. Inclusion criteria were interest in having sex with men, having had sex with men on average at least once a month in the past year, and having had one episode of binge drinking in the past year (consuming at least four drinks in a two-hour period; National Institute on Alcohol Abuse and Alcoholism, 2004). Exclusion criteria were current abstinence from alcohol consumption or endorsement of current or past alcohol use disorder symptoms.

Participants had a mean age of 23.1 years (SD = 3.2 years). Their racial/ethnic identifications were 68.3% European American/White, 9.6% Asian American, 3.8% African American/Black, 18.3% multiracial or other. Forty-four women (8.3%) identified as Hispanic/Latina. Approximately half (50.9%) were current students and 61.1% were employed full- or part-time.

Procedure

Eligibility was determined by an online screening survey. Upon enrollment, participants completed a 60-minute online baseline survey, which included measures of CSA, adult SV, and drinking habits. There were four one-month follow-up assessment periods three, six, nine, and twelve months post-enrollment. Each assessment month consisted of four consecutive weekly surveys about the previous week's sexual and drinking experiences; the first survey of each assessment month included questions about any SV experiences since the previous assessment. Gift card codes were provided as compensation for each completed survey (up to $350 total).

Of 619 women who enrolled in the study and provided baseline data, 530 (86%) were included in this analysis. Those excluded withdrew from the study after baseline (i.e., provided no post-baseline data; n = 18) or remained in the study but were missing data needed for this analysis (n = 71). Of those 71, 23 provided no follow-up SV data because they missed the first weekly survey (in which the SV questions were asked) for all assessment months (n = 5) or chose not to answer the SV questions at all assessment months (n = 18), and 48 were missing one or more covariate values, all of which are needed to compute conditional LTA models.

Included women completed a median of 13 of the 16 possible weekly follow-up surveys; when a weekly survey was missing, the non-missing surveys from that assessment month were used to compute the drinking covariate. A small proportion were missing SV data at one or more assessment months because they missed a Week 1 survey (1%, n = 5) or chose not to answer the SV questions in one or more Week 1 surveys (2%, n = 12). There were no significant differences in baseline SV history, CSA history, typical number of drinks per drinking day, or age between women who discontinued providing SV data partway through the study (i.e., after Month 3 or 6; 3%, n = 16) and those who did not. Furthermore, discontinuing to provide SV data was not significantly associated with having been classified as victimized early on in the study (i.e., at Month 3 or 6; 24.2%, n = 128). Therefore, there was no evidence that attrition had a substantial effect on study results or interpretation.

Measures

Childhood Trauma Questionnaire (CTQ; Bernstein et al., 2003), short form, Sexual Abuse subscale

At baseline, CSA history was assessed with the CTQ Sexual Abuse subscale, which contains five items (e.g., “Someone tried to make me do sexual things or watch sexual things”) rated on 0 (Never true) to 4 (Very often true) scales. Items are averaged with higher scores indicating a more severe CSA history. This subscale had excellent internal consistency reliability in the current sample (Cronbach's α = .92).

Daily Drinking Questionnaire (DDQ; Collins, Parks, & Marlatt, 1985)

At baseline and each weekly follow-up survey, frequency and quantity of alcohol consumption were assessed. Participants indicated on which days in the previous week they had consumed alcohol, and estimated how many alcoholic drinks they had consumed aided by a graphic that showed examples of standard drinks. Daily estimates were averaged to compute typical number of drinks per drinking day.

Sexual Experiences Survey (SES; Koss et al., 2007)

The SES was used at baseline to assess adult SV history (age 14 to study entry) and in the first week of each follow-up assessment month to assess SV occurring since the previous assessment. The SES includes questions about completed and attempted SV outcomes, including unwanted touching and oral, vaginal, and anal penetration by tactic type (e.g., verbal coercion, intoxication, physical force). For each assessment month, each outcome was coded as binary to indicate whether the participant had experienced that outcome (0 = not reported; 1 = reported). To include lifetime SV history as a baseline covariate, a continuous score (Davis et al., 2014) that combines the frequency and severity of outcomes by tactic type was computed (possible range 0-63).

Data Analysis

Statistical analyses were conducted in Mplus Version 7.2 (Muthén & Muthén, 1998-2012). In LCA, observed binary or categorical indicators are used to model discrete latent classes, the number of which is determined by the researcher. Probabilities of membership in each class are calculated for each individual. Interpretation and labeling of classes are based on patterns of item endorsement (i.e., proportions of individuals in each class who endorsed each item). Selection of the best number of classes is based on multiple criteria, including model fit statistics and interpretability of classes (Lanza, Patrick, & Maggs, 2010). LCA is commonly used as the measurement model for LTA (Nylund, 2007). In this analysis, cross-sectional LCA at each time point was used as an exploratory strategy to identify victimization profiles and to guide selection of a measurement model.

LTA is an application of LCA to longitudinal data in which latent class models are computed for each time point simultaneously and transition probabilities (i.e., the probabilities of transitioning among classes between time points) are estimated. LTA also classifies individuals into patterns of class membership across time; the number of possible patterns is defined by the number of classes and time points. Compared with other longitudinal structural equation modeling frameworks, LTA is advantageous for modeling trajectories of class membership when the construct of interest (in this case, SV) is assumed to be categorical in nature and to change over time in a discrete, rather than continuous, manner (Lanza & Collins, 2008). In the current study, for instance, a majority (54.8%) of the participants who were victimized during their participation were victimized at only one of the four time points; thus, a framework in which change over time is modeled continuously (e.g., latent growth curve modeling) would not describe the form of the phenomenon well.

The analysis proceeded as follows. First, cross-sectional LCAs were conducted separately at each time point to (a) explore the underlying class structure of victimization status, (b) examine whether class structures appeared stable across time, and (c) inform selection of a measurement model for the LTA. Second, LTA was conducted to model victimization status across time and to describe the probabilities of transitioning between victimization statuses. Third, time-invariant covariates (age at study entry, adult SV history, and CSA history) were added as predictors of victimization status at each time point. Fourth, number of standard drinks per drinking day was included as a time-varying covariate that both predicted simultaneous class membership and was predicted by previous class membership. Finally, in a separate LTA model1, drinking was used to predict transition probabilities, which is conceptually a moderation question (i.e., whether the association between victimization status at Time T and victimization status at Time T+1 differs depending on drinking at Time T+1). Maximum likelihood estimation with robust standard errors (MLR estimator in Mplus) was used in all models.

Missing data were handled in two ways. First, in conditional LTA models, Mplus excludes participants who are missing a value on any covariate, so those participants were removed prior to analysis2. Second, if a participant was missing SV data at one or more assessment months, her non-missing data were still included as long as she had provided SV data for at least one assessment month (and complete covariate data at all months). Missing SV observations were handled using full information maximum likelihood estimation, which provides the estimated variance-covariance matrix and infers probable class membership at the missing time points using all available data.

Results

Frequency of sexual assault endorsement

At baseline, 26.8% of participants reported some history of CSA (i.e., a CTQ score >0). Among those who reported CSA, the mean CTQ score was 1.34 (SD = 1.09). The majority of participants (75.7%) reported SV at least once since age 14 (i.e., a lifetime SES score >0). Among those who reported lifetime SV, the mean SES score was 24.82 (SD = 18.37).

Number of responses at each follow-up assessment month and proportions of respondents who endorsed each binary SV outcome are shown in Table 1. Unwanted contact was the most frequently endorsed outcome at each time point, although the proportion of participants endorsing it dropped by over half between Month 3 (25%) and Months 9 and 12 (both 11%). Attempted and completed anal penetration were the least endorsed outcomes at each time point.

Table 1. Numbers of Respondents to and Proportion Endorsing SV Outcomes Across Time.

Month 3 Month 6 Month 9 Month 12
Outcome N Prop. N Prop. N Prop. N Prop.
Unwanted contact 447 0.25 470 0.17 480 0.11 450 0.11
Oral penetration 447 0.08 470 0.07 482 0.05 450 0.04
Vaginal penetration 446 0.10 470 0.07 482 0.05 450 0.05
Anal penetration 446 0.04 468 0.03 482 0.01 449 0.02
Attempted oral penetration 447 0.12 469 0.09 482 0.04 449 0.03
Attempted vaginal penetration 447 0.10 469 0.07 482 0.05 450 0.03
Attempted anal penetration 447 0.04 468 0.04 482 0.02 449 0.02
Open in a new tab

Note. Sample size variations within a time point are the result of item non-response.

N = number of participants who provided a response

Prop. = proportion of respondents who endorsed outcome

Drinking quantity

At baseline, participants reported drinking a mean of 11.6 standard drinks per week (SD = 7.5) with a mean of 3.4 (1.5) drinking days per week and 3.5 (1.7) drinks per drinking day. In follow-up assessments, participants drank a mean of 4.1 drinks per drinking day (SD = 2.1 drinks) at Month 3, 4.1 (2.4) at Month 6, 3.7 (2.5) at Month 9, and 3.5 (2.5) at Month 12.

Latent class analysis (LCA)

Two-, three-, and four-class solutions were computed separately for each assessment month using the seven binary SV outcome variables as indicators. Fit statistics were considered in light of the interpretability of each solution. Absolute model fit was assessed using the likelihood ratio chi-square (G2), which, if significant, indicates misfit between the model and the data. Entropy, an index of classification accuracy where values closer to 1 indicate better accuracy, was also examined. Relative model fit (between solutions with different numbers of classes at a given month) was assessed using the Bayesian Information Criterion (BIC), of which the model with the lowest value has the best relative fit, and the bootstrapped likelihood ratio test (BLRT; Collins & Lanza, 2010), which, if significant for a solution with k classes, indicates better model fit than a solution with k-1 classes. Fit statistics are summarized in Table 2.

Table 2. Model Fit Indicators for 2-, 3-, and 4-Class LCA Solutions by Assessment Month.

Assessment month # Classes High/low na Absolute model fit indicators Relative model fit indicators
G2b Entropy BIC BLRTc
3 2 367/80 .29 0.924 1526.496 <.001
3 367/28 .89 0.930 1552.341 <.001
4 367/11 .99 0.939 1578.958 <.001
6 2 408/62 .41 0.915 1396.448 <.001
3 408/19 .71 0.935 1426.180 .04
4 408/15 .96 0.953 1453.732 .05
9 2 427/45 .99 0.943 962.020 <.001
3 438/14 .99 0.966 991.521 .04
4 438/5 .99 0.967 1028.025 .99
12 2 417/33 .99 0.968 805.769 <.001
3 416/13 .99 0.941 822.448 <.001
4 416/5 .99 0.953 860.358 .99
Open in a new tab

Note. BIC = Bayesian information criterion; BLRT = bootstrapped likelihood ratio test.

a

Highest and lowest class size produced by each solution.

b

Values listed for G2 are p-values. A p-value greater than .05 indicates good model fit.

c

Values listed for BLRT are p-values, with values <.05 bolded. A significant BLRT indicates that the model has better fit relative to a model with k-1 classes.

A two-class solution (Figure 1a-d) fit the data well at all four assessment months (G2p's > .05) and showed good classification accuracy (entropies > .90). It was highly interpretable, yielding classes that could clearly be labeled “victimized” and “non-victimized.” The victimized class appeared to characterize women who experienced either attempted or completed SV.

Figure 1.

Figure 1

Open in a new tab

Conditional item probabilities (i.e., the probabilities that an individual in a given class endorsed each item) for a two-class LCA solution across four assessment months.

Note: Class 1 = victimized; Class 2 = non-victimized.

Three- and four-class solutions had similarly good classification accuracy and absolute model fit, and BLRT values suggested that the three-class solution had relatively better model fit than the two-class solution; BIC values, however, favored the two-class solution at all assessment months. In the three- and four-class solutions, the classes did not appear to have consistent profiles across the four months, suggesting that transitions between classes would be uninterpretable if a three- or four-class solution was applied longitudinally in an LTA model. Furthermore, these solutions produced some very small class sizes (as few as five participants; Table 2). Sparseness (small expected cell counts) in the very large contingency tables in LTA can cause estimation problems, which can be exacerbated by the inclusion of covariates (Collins & Lanza, 2010). Because our primary research questions concerned prediction of victimization status with covariates, and because of its superior interpretability, we opted to retain the two-class solution as an LTA measurement model.

Latent Transition Analysis

Based on LCA results, a two-class LTA was computed. Measurement invariance was assumed to ensure estimation of the same number and profile of classes across time, which allowed interpretation of transition probabilities to be straightforward and equivalent at each transition point.

Prevalence of latent statuses

At each assessment, the non-victimized status had substantially more members than the victimized status. Of 16 possible “patterns” of victimization status across time, we observed 15. The most common pattern was non-victimized across all four assessment months (N = 384, 72.5%). The next most common was the pattern in which a woman was classified as victimized at Months 3 and 6 and non-victimized at Months 9 and 12 (N = 36, 6.8%). Fourteen women (2.6%) were classified as victimized in all four assessment months.

Latent transition probabilities

Estimated latent transition probabilities between adjacent assessment months are displayed in Table 3. In support of Hypothesis 1, women who were non-victimized in a given assessment month had a high probability of remaining non-victimized at the subsequent assessment month. Women who were victimized in one assessment month, however, were approximately equally likely to remain victimized at the subsequent assessment month (probability .434-.555, depending on the month) as to transition to the non-victimized status (probability .445-.566).

Table 3. Estimated Latent Transition Probabilities for the 2-Class LTA Model.
Non-victimized Victimized
Month 6 latent status
Month 3 latent status
 Non-victimized .923 .077
 Victimized .445 .555
Month 9 latent status
Month 6 latent status
 Non-victimized .966 .034
 Victimized .566 .434
Month 12 latent status
Month 9 latent status
 Non-victimized .960 .040
 Victimized .558 .442
Open in a new tab

Covariate predictors of victimization status (Table 4)

Table 4. Logistic Regression Coefficients for 2-class LTA with Covariate Predictors of Victimized Status.
Effect Coefficient S.E. z p Odds Ratio
Age -0.070 0.029 -2.434 0.015 0.932
SV history 0.033 0.004 7.545 0.000 1.033
CSA history 0.196 0.087 2.249 0.025 1.216
Drinking 0.380 0.089 4.265 0.000 1.462
Open in a new tab

Note. Drinking = average number of standard drinks per drinking day.

For each covariate, predictive paths to victimization status were constrained to be equal across assessment months, both to aid model estimation and because their effects were not hypothesized to differ depending on time.

Time-invariant predictors

Age at study entry significantly predicted victimization status: The older a woman was, the lower her odds of being in the victimized class at any assessment month (OR = .93, 95% CI .881-.986). In support of Hypothesis 2, CSA history also significantly predicted victimization status: The higher a woman's CSA score, the higher her odds of being in the victimized class at any assessment month (OR = 1.22, 95% CI 1.03-1.44). Similarly, in support of Hypothesis 3, the greater a woman's lifetime SV history at baseline, the higher her odds of being in the victimized class at any assessment month (OR = 1.03, 95% CI 1.02-1.04).

Time-varying predictors

Drinking at each assessment significantly predicted drinking at the subsequent assessment (Hypotheses 4 and 5; p <.001). Drinking also significantly predicted concurrent class membership (Hypothesis 6): The more a woman drank at any assessment month, the more likely she was to be classified as victimized at that month (p< .001). For each increase of one drink per drinking day, there was an approximately 4-10% increase in the probability of being classified as victimized.

Separately, to examine Hypothesis 7, drinking was used to predict transition probabilities. For women who were classified as non-victimized at a given assessment, a higher level of drinking was associated with a significantly higher probability of transitioning to victimized status at the subsequent assessment: Non-victimized women who drank 1 SD above the mean (approximately six drinks per drinking day) had an average estimated 93% probability of remaining non-victimized, compared with 98% estimated probability for non-victimized women who did not drink (OR = 3.75, 95% CI 2.04-7.11). Furthermore, for women who were victimized at a given month, a higher level of drinking was associated with a significantly higher probability of subsequently being victimized: Victimized women who drank 1 SD above the mean had a 35% estimated probability of remaining victimized at the subsequent assessment, compared with 22% for victimized women who did not drink (OR = 1.91, 95% CI 1.05-3.26).

These analyses suggest that drinking, in addition to being associated with victimized status, also influenced the probabilities of transitioning into or remaining in the victimized status. Women non-victimized at one assessment were more likely to become victimized the higher their drinking level over the next assessment month, and women victimized at one assessment were more likely to remain victimized the higher their drinking level over the next month.

Latent victimization status as a predictor of subsequent drinking

Differences between non-victimized and victimized women's subsequent mean number of drinks per drinking day at each assessment month were computed and tested for significance. In support of Hypothesis 8, at each month victimized women subsequently drank more than non-victimized women (all p's < .05 except Month 6, p = .073) with the differences between the means ranging from 0.4 to 1.3 drinks per drinking day. For example, women who were victimized at Month 9 drank more at Month 12 than women who were non-victimized at Month 9.

Discussion

This study addressed discrepancies in the literature regarding temporal relationships between young women's background risk factors (SV history and age), typical alcohol consumption, and ongoing vulnerability to adult SV over one year. In support of both the self-medication hypothesis and routine activity theory, we found a reciprocal association between typical drinking quantity and SV, such that women who drank more (i.e., one standard deviation above the mean, or about six drinks per drinking day) were more likely to subsequently be victimized during the year and that being victimized during the year was associated with a higher level of subsequent drinking. Furthermore, between consecutive assessment periods, higher drinking quantity was associated with greater probability of remaining in, or transitioning into, victimized status. Finally, prior victimization (both CSA and prior adult SV) increased women's likelihood of being victimized during the year and older age appeared to be a protective factor against being victimized.

Taken together in the context of a strong longitudinal research design and data analytic strategy, these findings bridge gaps between previous studies regarding the direction of the link between SV vulnerability and alcohol consumption. The reciprocal association we identified is consistent with a cycle of vulnerability in which women who engage in risky levels of drinking are victimized, subsequently continue or increase their drinking, and are thus at elevated risk for becoming re-victimized. It is notable that we observed these associations across relatively frequent assessments (every three months), suggesting an active cycle that exerts proximal influences on young women's vulnerability. Reciprocal associations may represent causal mechanisms or the influences of other factors; for example, SV experiences may directly influence women's drinking by a mechanism such as self-medication, or other ongoing contextual factors may explain the association. We emphasize that these findings do not place the blame for sexual assault on the victim – responsibility lies with the perpetrator – but we note them as potential explanations for the positive association between alcohol use and SV.

Our findings provide further evidence that women's prior SV predicts future SV; that is, there is continuity in women's victimization “trajectories” over time. It is notable that we observed a high rate of prior SV in the women who entered the study: Three quarters of the participants reported a history of SV at baseline. Although we intentionally recruited women who engaged in binge drinking and who therefore were likely to be vulnerable to SV, our binge drinking criterion was fairly minimal (one episode in the past year). That we observed such a high prevalence of SV even in this moderate-risk sample underscores the importance of identifying factors that can inform risk reduction efforts.

Limitations

Limitations to this analysis should inform further studies. First, because of the available sample size and SV endorsement rates, we were limited to using two SV classes in our analyses, yet finer distinctions between types of SV may exist. We modeled the SV latent class structure in an empirical (not theory-driven) manner, and results reflect the items used and sample characteristics. An analysis using different items (e.g., perpetrator tactics) or a different sample (e.g., less victimized) might have produced a different class structure. Second, we observed declines in both typical drinking quantity and SV rates over the course of participation, possibly because participating rendered an unintended intervention effect. Repeatedly answering detailed questions about drinking and sexual experiences may have increased women's awareness of risky situations and influenced their behavior. Third, we did not examine how alcohol consumption and SV are associated at the event level, so we cannot draw conclusions about alcohol's role in any given instance of SV. Fourth, it was outside the scope of this analysis to examine mechanisms that might underlie associations observed.

Research implications

This analysis illuminated important questions that remain to be addressed. First, does event-level drinking predict SV vulnerability differently than typical drinking? It is possible, for example, that a woman's event-level deviation from her own typical drinking is predictive of risk over and above her typical drinking (e.g., Neal & Fromme, 2007). Given the reciprocal association we observed between typical drinking and SV, it would be useful to investigate whether event-level drinking and SV are also reciprocally linked. Future studies should examine the relative contributions of typical and event-level drinking to SV vulnerability using frameworks (like LTA) that can model temporal associations. Second, does the reciprocal association between drinking and SV differ depending on whether SV is alcohol-involved or not? There is evidence, for example, that assault characteristics such as alcohol involvement influence the development of subsequent PTSD symptoms (Peter-Hagene & Ullman, 2015), which may in turn predict different drinking trajectories. Additionally, further research examining finer gradations of assault type and perpetrator tactic is key to identifying women with high vulnerability to SV, hazardous drinking, and associated outcomes.

Clinical and policy implications

Although in this study we have clarified that there is a reciprocal relationship between SV and alcohol consumption, which of these factors initiates the feedback loop may differ depending on a woman's circumstances. Some women may start to drink heavily to cope with PTSD symptoms resulting from a sexual assault (Kaysen et al., 2014; Ullman et al., 2013), whereas other women may drink heavily in a social setting and get sexually assaulted by a perpetrator who targets vulnerable women (Abbey, Zawacki, Buck, Clinton, & McAuslan, 2004; Testa & Livingston, 2009). Thus, a woman seeking treatment for either SV or alcohol use should be assessed for the other as well. For many women, treatments that address both the negative outcomes resulting from SV and heavy drinking may be appropriate (for a review, see Najavits & Hien, 2013) and may interrupt the “snowball” effects of increased drinking and SV vulnerability. Teaching positive coping strategies during treatment may both improve a woman's mental health (Falsetti & Resnick, 2000) and reduce drinking (Stappenbeck, Hassija, Zimmerman, & Kaysen, 2015) which may lower vulnerability to future SV.

Understanding more about the exact nature of the reciprocal relationship between alcohol consumption and SV is key to intervening effectively with victimized women. Treatment implications may differ depending on whether the problematic drinking preceded or followed the victimization. Women who are victimized after drinking often feel guilty, blaming themselves rather than the perpetrator for his sexual aggression (Brown, Testa, & Messman-Moore, 2009). Recognizing that a woman needs help can be challenging; it may not be readily apparent when drinking is motivated by avoidance of SV-related distress. Widespread public education campaigns not just about the dangers of drinking heavily, but also about why some people drink to excess, may help raise awareness and promote appropriate help-seeking. Women, too, can look out for each other in social settings, warning each other about men who exhibit rape-supportive attitudes and behaviors (Cue, George, & Norris, 1996) or intervening if a woman appears too intoxicated to recognize a potential perpetrator (Norris, Nurius, & Graham, 1999).

The current study provided a stepping stone toward understanding women's alcohol-related vulnerability to SV and highlighted remaining questions. This and future studies have the potential to provide a foundation for effective SV risk reduction programs for vulnerable women and to inform treatment efforts for victimized and heavy-drinking women.

Acknowledgments

This research was supported by grant R01AA014512 from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) to Jeanette Norris. The authors thank Katie Witkiewitz for providing statistical expertise during the preparation of this manuscript.

Footnotes

1

Models were computed separately because Mplus cannot calculate the effects of a covariate on transition probabilities if that covariate is also functioning as a dependent variable in the same model.

2

LCA and unconditional LTA models were also computed including the 17 participants who were missing covariate data but had provided SV data. The results confirmed that excluding those participants from the final models did not change the interpretation of the class structure, suggest that a different number of classes should be used, or meaningfully affect transition probabilities.

References

  1. Abbey A. Alcohol-related sexual assault: A common problem among college students. Journal of Studies on Alcohol. 2002;S14:118–128. doi: 10.15288/jsas.2002.s14.118. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Abbey A, Zawacki T, Buck PO, Clinton AM, McAuslan P. Sexual assault and alcohol consumption: What do we know about their relationship and what types of research are still needed? Aggression and Violent Behavior. 2004;9:271–303. doi: 10.1016/S1359-1789(03)00011-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bernstein DP, Stein JA, Newcomb MD, Walker E, Pogge D, Ahluvalia T, et al. Zule W. Development and validation of a brief screening version of the Childhood Trauma Questionnaire. Child Abuse & Neglect. 2003;27:169–190. doi: 10.1016/S0145-2134(02)00541-0. [DOI] [PubMed] [Google Scholar]
  4. Black MC, Basile KC, Breiding MJ, Smith SG, Walters ML, Merrick MT, et al. Stevens MR. Centers for Disease Control and Prevention; 2011. The National Intimate Partner and Sexual Violence Survey (NISVS): 2010 Summary Report. Retrieved from http://www.cdc.gov/violenceprevention/nisvs/ [Google Scholar]
  5. Brown AL, Testa M, Messman-Moore TL. Psychological consequences of sexual victimization resulting from force, incapacitation, or verbal coercion. Violence Against Women. 2009;15:898–919. doi: 10.1177/1077801209335491. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Classen CC, Palesh OG, Aggarwal R. Sexual revictimization: A review of the empirical literature. Trauma, Violence and Abuse. 2005;6:103–129. doi: 10.1177/1524838005275087. [DOI] [PubMed] [Google Scholar]
  7. Cohen LE, Felson M. Social change and crime rate trends: A routine activity approach. American Sociological Review. 1979;44:588–608. [Google Scholar]
  8. Collins LM, Lanza ST. Latent Class and Latent Transition Analysis with Applications in the Social, Behavioral, and Health Sciences. Hoboken, NJ: John Wiley & Sons, Inc; 2010. [Google Scholar]
  9. Collins RL, Parks GA, Marlatt GA. Social determinants of alcohol consumption: The effects of social interaction and model status on the self-administration of alcohol. Journal of Consulting and Clinical Psychology. 1985;53:189–200. doi: 10.1037/0022-006X.53.2.189. [DOI] [PubMed] [Google Scholar]
  10. Cue KL, George WH, Norris J. Women's appraisals of sexual assault risk in dating situations. Psychology of Women Quarterly. 1996;20:487–504. doi: 10.1111/j.1471-6402.1996.tb00318.x. [DOI] [Google Scholar]
  11. Davis KC, Gilmore AK, Stappenbeck CA, Balsan MJ, George WH, Norris J. How to score the Sexual Experiences Survey? A comparison of nine methods. Psychology of Violence. 2014;4:445–461. doi: 10.1037/a0037494. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Desai S, Arias I, Thompson MP, Basile KC. Childhood victimization and subsequent adult revictimization assessed in a nationally representative sample of women and men. Violence and Victims. 2002;17:639–653. doi: 10.1891/vivi.17.6.639.33725. [DOI] [PubMed] [Google Scholar]
  13. Elliott DM, Mok DS, Briere J. Adult sexual assault: Prevalence, symptomatology, and sex differences in the general population. Journal of Traumatic Stress. 2004;17:203–211. doi: 10.1023/B:JOTS.0000029263.11104.23. [DOI] [PubMed] [Google Scholar]
  14. Falsetti SA, Resnick HS. Treatment of PTSD using cognitive and cognitive behavioral therapies. Journal of Cognitive Psychotherapy: An International Quarterly. 2000;14:261–285. [Google Scholar]
  15. French BH, Bi Y, Latimore TG, Klemp HR, Mutler EE. Sexual victimization using latent class analysis: Exploring patterns and psycho-behavioral correlates. Journal of Interpersonal Violence. 2014;29:1111–1131. doi: 10.1177/0886260513506052. [DOI] [PubMed] [Google Scholar]
  16. Gidycz CA, Loh C, Lobo T, Rich C, Lynn SJ, Pashdag J. Reciprocal relationship among alcohol use, risk perception, and sexual victimization: A prospective analysis. Journal of American College Health. 2007;56:5–14. doi: 10.3200/JACH.56.1.5-14. [DOI] [PubMed] [Google Scholar]
  17. Kaysen D, Atkins DC, Simpson TL, Stappenbeck CA, Blayney JA, Lee CM, Larimer ME. Proximal relationship between PTSD symptoms and drinking about female college students: Results from a daily monitoring study. Psychology of Addictive Behaviors. 2014;28:62–73. doi: 10.1037/a0033588. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kaysen D, Neighbors C, Martell J, Fossos N, Larimer ME. Incapacitated rape and alcohol use: A prospective analysis. Addictive Behaviors. 2006;31:1820–1832. doi: 10.1016/j.addbeh.2005.12.025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Khantzian EJ. The self-medication hypothesis revisited: The dually diagnosed patient. Primary Psychiatry. 2003;10:47–48. 53–54. [Google Scholar]
  20. Kilpatrick DG, Acierno R, Resnick HS, Saunders BE, Best CL. A 2-year longitudinal analysis of the relationships between violent assault and substance use in women. Journal of Consulting and Clinical Psychology. 1997;65:834–847. doi: 10.1037/0022-006X.65.5.834. [DOI] [PubMed] [Google Scholar]
  21. Kilpatrick DG, Edmunds CN, Seymour AK. Rape in America: A report to the nation. Arlington, VA: National Victim Center; 1992. [Google Scholar]
  22. Koss MP, Abbey A, Campbell R, Cook S, Norris J, Testa M, et al. White J. Revising the SES: A collaborative process to improve assessment of sexual aggression and victimization. Psychology of Women Quarterly. 2007;31:357–370. doi: 10.1111/j.1471-6402.2007.00385.x. [DOI] [Google Scholar]
  23. Koss MP, Dinero TE. Discriminant analysis of risk factors for sexual victimization among a national sample of college women. Journal of Consulting and Clinical Psychology. 1989;57:242–250. doi: 10.1037/0022-006X.57.2.242. [DOI] [PubMed] [Google Scholar]
  24. Lanza ST, Collins LM. A new SAS procedure for latent transition analysis: Transitions in dating and sexual risk behavior. Developmental Psychology. 2008;44:446–456. doi: 10.1037/0012-1649.44.2.446. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Lanza ST, Patrick ME, Maggs JL. Latent transition analysis: Benefits of a latent variable approach to modeling transitions in substance use. Journal of Drug Issues. 2010;40:93–120. doi: 10.1177/002204261004000106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Laursen B, Hoff E. Person-centered and variable-centered approaches to longitudinal data. Merrill-Palmer Quarterly. 2006;52 [Google Scholar]
  27. Macy RJ, Nurius PS, Norris J. Latent profiles among sexual assault survivors: Implications for defensive coping and resistance. Journal of Interpersonal Violence. 2007;22:543–565. doi: 10.1177/0886260506298841. [DOI] [PubMed] [Google Scholar]
  28. Messman-Moore T, Ward RM, Zerubavel N, Chandley RB, Barton SN. Emotion dysregulation and drinking to cope as predictors and consequences of alcohol-involved sexual assault: Examination of short-term and long-term risk. Journal of Interpersonal Violence. 2015;30:601–621. doi: 10.1177/0886260514535259. [DOI] [PubMed] [Google Scholar]
  29. Miron LR, Orcutt HK. Pathways from childhood abuse to prospective revictimization: Depression, sex to reduce negative affect, and forecasted sexual behavior. Child Abuse & Neglect. 2014;38:1848–1859. doi: 10.1016/j.chiabu.2014.10.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Mouilso ER, Fischer S. A prospective study of sexual assault and alcohol use among first-year college women. Violence and Victims. 2012;27:78–94. doi: 10.1891/0886-6708.27.1.78. [DOI] [PubMed] [Google Scholar]
  31. Mustaine EE, Tewksbury R. A routine activity theory explanation for women's stalking victimizations. Violence Against Women. 1999;5:43–62. doi: 10.1177/10778019922181149. [DOI] [PubMed] [Google Scholar]
  32. Muthén LK, Muthén BO. Mplus User's Guide. Seventh. Los Angeles, CA: Muthén & Muthén; 1998-2012. [Google Scholar]
  33. Najavits LM, Hien D. Helping vulnerable populations: A comprehensive review of the treatment outcome literature on substance use disorder and PTSD. Journal of Clinical Psychology. 2013;69:433–479. doi: 10.1002/jclp.21980. [DOI] [PubMed] [Google Scholar]
  34. Najdowski CJ, Ullman SE. The effects of revictimization on coping and depression in female sexual assault victims. Journal of Traumatic Stress. 2011;24:218–221. doi: 10.1002/jts.20610. [DOI] [PubMed] [Google Scholar]
  35. National Institute on Alcohol Abuse and Alcoholism. NIAAA council approves definition of binge drinking. NIAAA Newsletter. 2004;3:3. [Google Scholar]
  36. Neal DJ, Fromme K. Event-level covariation of alcohol intoxication and behavioral risks during the first year of college. Journal of Consulting and Clinical Psychology. 2007;75:294–306. doi: 10.1037/0022-006X.75.2.294. [DOI] [PubMed] [Google Scholar]
  37. Norris J, Nurius PS, Graham TL. When a date changes from fun to dangerous: Factors affecting women's ability to distinguish. Violence Against Women. 1999;5:230–250. doi: 10.1177/10778019922181202. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Nylund KL. Latent Transition Analysis: Modeling Extensions and an Application to Peer Victimization. 2007 Retrieved from https://www.statmodel.com/download/nylunddis.pdf.
  39. Parks KA, Hseih YP, Taggart C, Bradizza CM. A longitudinal analysis of drinking and victimization in college women: Is there a reciprocal relationship? Psychology of Addictive Behaviors. 2014;28:943–951. doi: 10.1037/a0036283. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Peter-Hagene LC, Ullman SE. Sexual assault-characteristics effects on PTSD and psychosocial mediators: A cluster-analysis approach to sexual assault types. Psychological Trauma: Theory, Research, Practice, and Policy. 2015;7:162–170. doi: 10.1037/a0037304. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Santaularia J, Johnson M, Hart L, Haskett L, Welsh E, Faseru B. Relationships between sexual violence and chronic disease: A cross-sectional study. BMC Public Health. 2014;14:1–7. doi: 10.1186/1471-2458-14-1286. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Siegel JA, Williams LM. Risk factors for sexual victimization of women: Results from a prospective study. Violence Against Women. 2003;9:902–930. doi: 10.1177/1077801203255130. [DOI] [Google Scholar]
  43. Stappenbeck CA, Hassija CM, Zimmerman L, Kaysen D. Sexual assault related distress and drinking: The influence of daily reports of social support and coping control. Addictive Behaviors. 2015;42:108–113. doi: 10.1016/j.addbeh.2014.11.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Testa M, Livingston JA. Alcohol and sexual aggression: Reciprocal relationships over time in a sample of high-risk women. Journal of Interpersonal Violence. 2000;15:413–427. doi: 10.1177/088626000015004005. [DOI] [Google Scholar]
  45. Testa M, Livingston JA. Alcohol consumption and women's vulnerability to sexual victimization: Can reducing women's drinking prevent rape? Substance Use and Misuse. 2009;44:1349–1376. doi: 10.1080/10826080902961468. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Testa M, VanZile-Tamsen C, Livingston JA. Prospective prediction of women's sexual victimization by intimate and nonintimate male perpetrators. Journal of Consulting and Clinical Psychology. 2007;75:52–60. doi: 10.1177/088626000015004005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Ullman SE, Relyea M, Peter-Hagene L, Vasquez AL. Trauma histories, substance use coping, PTSD, and problem substance use among sexual assault victims. Addictive Behaviors. 2013;38:2219–2223. doi: 10.1016/j.addbeh.2013.01.027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Zinzow HM, Resnick HS, McCauley JL, Amstadter AB, Ruggiero KJ, Kilpatrick DG. Prevalence and risk of psychiatric disorders as a function of variant rape histories: Results from a national survey of women. Social Psychiatry and Psychiatric Epidemiology. 2012;47:893–902. doi: 10.1007/s00127-011-0397-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES