The Intersection of Intimate Partner Violence Perpetration and Sexual Risk Behavior Among Young Men in Tanzania: A Latent Class Analysis of Patterns and Outcomes

Abstract

Few studies of intimate partner violence (IPV) perpetration and sexual risk behavior among men have examined how multiple dimensions of these behaviors intersect in ways that may uniquely elevate health risks. The current study used latent class analysis to: (1) identify distinct patterns of IPV and sexual risk behavior in a sample of Tanzanian men (n = 985) and (2) examine associations between identified patterns and health outcomes. Four classes were identified: normative (64% of the sample), IPV only (14%), sexual risk only (13%), and comorbid IPV/sexual risk (5%). Compared to men in the normative subgroup, men in the comorbid group had significantly higher odds of STI infection, higher perceived HIV risk, and greater odds of substance use. Findings provide evidence that engaging in IPV and multiple sexual partnerships (i.e., a comorbid pattern) denotes elevated health risks across a range of indicators, suggesting the importance of targeted treatment and prevention efforts for men in this subgroup.

Introduction

HIV/AIDS and intimate partner violence (IPV) are highly prevalent in Sub-Saharan Africa (SSA) and are a major public health and human rights concern [1, 2]. The global burden of HIV disproportionately impacts SSA where 71% of people living with HIV resided and 75% of deaths and 65% of new infections occurred in 2017 [3–5]. Similarly, emotional, physical, and sexual IPV victimization among women is widespread throughout SSA [2] and leads to a range of negative physical, mental, and perinatal health impacts among women [6–9] and children who are exposed to IPV [10].

Numerous studies in SSA have found that risk of HIV is greater among women who have experienced IPV than among women who have not [11–14]. In turn, these syndemic health risks may work together to amplify risk of negative health outcomes [15]. Researchers have posited that the link between HIV and IPV may be because men who engage in IPV also tend to engage in risky sexual behaviors that put both themselves and their partners at risk for infection with HIV as well as other sexually transmitted infections (STIs; [16–18]). Consistent with this notion, a growing body of research has identified an association between IPV perpetration and sexual risk behavior (e.g., condomless sex; multiple sexual partners) and outcomes (e.g., HIV/STIs) among men [19–25]. For example, studies in Ghana and Zimbabwe found that men who reported IPV perpetration were also more likely to report having multiple sexual partners in the past year [19, 22].

While this emerging body of literature suggests a link between IPV perpetration and sexual risk behavior, almost no research has examined the intersection between these behaviors from a typological perspective. Typological perspectives posit there are subgroups of men who engage in qualitatively distinct patterns of IPV and sexual risk that have different etiological origins and sequelae. For example, the control-based typology proposed by Johnson (1995) suggests that IPV and risky sexual behavior likely co-occur among a subgroup of men those who engage in coercive controlling violence, a type of violence characterized by perpetration of severe forms of IPV and is “…a product of patriarchal traditions of men’s right to control their women” ([26], p. 284). For this subgroup, harmful gender norms that prescribe control and dominance over women, sexual entitlement, and hypersexual performance may drive a behavioral profile characterized by engagement in severe forms of IPV and risky sexual behavior. The typology also posits that there is also a distinct subgroup of men in the population who engage in situational IPV, which is less generalized and severe and is tied to specific relationship conflicts that escalate into abuse. For this subgroup, relationship conflict and stress may drive a behavioral profile characterized by involvement in some moderate IPV, but a relatively low likelihood of engaging in risky sexual behaviors [26, 27]. Holtzworth-Munroe and Stuart (1994) similarly proposed a violence typology that distinguishes a subgroup of generally violent/antisocial men who engage in severe forms of IPV and who are likely to engage in concurrent sexual partnerships, from other subgroups of men who engage in less severe and/or generalized violence and are less likely to engage in concurrent sexual partnerships [28]. In an empirically derived typology of masculine identities, Casey et al. (2016) identified two subgroups of men—misogynistic and sex-focused—that both engaged in relatively high levels of risky sexual behavior (sexual partners, one-night stands) but who differed in terms of their level of endorsement of traditional masculinity and involvement in IPV perpetration; the misogynistic subgroup was characterized by high and the sex-focused subgroup by low endorsement of these indicators [29].

While they differ in their particulars, these typological perspectives suggest there are heterogeneous subgroups of men who differ meaningful in terms of their patterns of engagement in IPV perpetration and sexual risk behavior. A better understanding of this heterogeneity may be key to informing the development of tailored prevention approaches and identify subgroups who are at differential risk for negative consequences. Yet, to date, almost no research has examined the intersection of IPV and sexual risk among men in SSA from a typological perspective and/or examined whether men who engage in different patterns of IPV/sexual risk behavior are at differential risk for negative health outcomes. Most studies that have examined the overlap between indicators of IPV perpetration and sexual risk behavior and outcomes have done so using variable-centered approaches (e.g., regression analysis) with a narrow focus on a particular indicator of IPV and sexual risk and with the assumption that the population of interest is homogeneous with respect to how indicators are related to each other and influence outcomes. For example, research assessing the impacts of IPV has tended to use single crude measures of perpetration, such as binary measures capturing any physical and/or sexual aggression, that do not consider that IPV varies meaningfully in both form (emotional, physical, sexual) and frequency [19–25]. Similarly, studies assessing the impacts of sexual risk behavior have tended to use individual behavioral indicators, such as condom use at last sex, thus not accounting for other dimensions of sexual risk, such as engaging in concurrent partnerships [30]. Accounting for the multidimensional nature of IPV and sexual risk may be of key importance for describing these phenomena more holistically and for identifying subgroups who may be at increased risk for negative health outcomes. For example, in a study of South African women, Reyes et al. (2020), identified two distinct patterns of IPV victimization (multiform severe controlling and moderate) that were differentially associated with negative mental health outcomes [31]. Similarly, studies of youth in the US and South Africa have identified multidimensional patterns of sexual risk behavior that differentially predict HIV and STI outcomes [32, 33].

The Current Study

A better understanding of the intersection of various dimensions of IPV and sexual risk behavior can offer a more comprehensive and nuanced illustration of different profiles of behavior common in the population and inform targeted lines of prevention programming. To this end, the focal aim of the current study was to use a person-centered analytic approach (latent class analysis [LCA]), to identify subgroups of young Tanzanian men characterized by engaging in distinct patterns (or profiles) of IPV perpetration (physical, psychological, sexual) and sexual risk behavior (condom use, multiple sexual partnerships, concurrency). In contrast to variable-centered approaches, which are oriented towards identifying common “main effect” associations between pairs of variables that are typically assumed to apply to all individuals, person-centered analyses seek to identify “key patterns of values across variables, where the person, viewed holistically, is the unit of analysis” ([34], p. 256). As such, this analytic approach is uniquely suited to capturing key configurations of variables denoting multiple dimensions of IPV and sexual risk behavior that, in turn, may differentially predict negative health outcomes.

Drawing from the typological perspectives described above, as well as empirical research, we hypothesized that, in addition to a normative group, characterized by not engaging in any IPV or risky sexual behavior, we would identify at least three “risky” behavior profiles: comorbid, characterized by involvement in multiple forms of IPV and risky sexual behavior; IPV-only, characterized by some limited involvement in IPV, but no involvement in risky sexual behavior; and sexual risk only, characterized by involvement in multiple sexual partnerships, but not IPV perpetration [12, 17–19, 25, 26]. To validate the explanatory value of the profiles we further examined associations between profile membership and negative health outcomes including STI infection, perceived HIV risk, and substance use involvement. These health outcomes were selected for study based on research documenting positive associations between sexual risk behavior and/or IPV perpetration and each of these outcomes, although no previous research, to our knowledge, has examined whether and how membership in distinct IPV/sexual risk profiles differentiates risk for these outcomes.

Method

Parent Study Design and Procedure

Secondary analyses were conducted using baseline data from a cluster randomized control trial of a microfinance and peer leadership intervention to prevent sexually transmitted infections and intimate partner violence among men in Dar es Salaam, Tanzania [35, 36]. Participants in the trial are members of venues, locally referred to as “camps,” where men socialize and sometimes engage in small scale enterprise. Camps were identified for inclusion in the trial in four wards (equivalent to US census tracts) in Dar es Salaam using an adaptation of PLACE (priorities for local AIDS control efforts) methodology [37].

To be eligible for inclusion, camps had to have 20–79 members, have been in existence for more than one year, and report no violence incidents with a weapon in the last six months (to ensure researcher safety). There were 172 eligible camps identified in the study area from which 60 were randomly selected for inclusion in the study. Through member rosters completed by leaders of the selected camps, 1581 camp members were identified and assessed for eligibility. Eligible participants had to be at least 15 years old, have been a camp member for at least 3 months, visit the camp at least once a week, plan to reside in Dar es Salaam for the next 30 months, and be willing to provide contact information.

At total of 1,258 men consented and were enrolled in the study (79.6%) after which camp members from one camp (n = 9) were removed from the study because new information rendered them ineligible for participation, resulting in a total sample of 1249 men in 59 camps. For the current study, analyses were restricted to the 985 men who reported ever having had a sexual partner in the past 12 months. Trained interviewers administered questionnaires in Swahili using computer assisted personal interviewing (CAPI). The interviews lasted approximately 60–90 min. To assess the prevalence of STIs urine samples were drawn from consenting participants immediately after the behavioral interview was completed. Due to funding limitations that precluded conducting STI testing on all participants a random sample of 50% of men who reported having ever had sex in their lifetime were invited to provide a urine sample. Participants who were diagnosed with an STI were provided free treatment by a study clinician.

The study was approved by the ethical review committees at the University of North Carolina at Chapel Hill and Muhimbili University of Health and Allied Sciences in Dar es Salaam, Tanzania. Individual written informed consent was obtained from all participants. Further details on study procedures can be found in the study protocol paper [35].

Measures

Latent Class Indicators

Six indicators were created for defining the latent IPV/sexual risk classes. Three of these assessed different forms of IPV—physical, sexual, and psychological—and three assessed different dimensions of sexual risk—inconsistent condom use, number of past year sexual partners, and past year concurrent partnerships.

IPV Perpetration Indicators

Past-year IPV perpetration was assessed using an adapted version of the WHO Violence Against Women instrument [7]. Participants were asked how many times they had perpetrated acts of psychological (e.g., belittled or humiliated their partner in front of others; 4 items), physical (e.g., kicked, dragged, or beaten their partner up; 6 items), and/or sexual IPV (e.g., physically forced their partner to have sex when they did not want to; 3 items). Response options were: never, once, 2–3 times, 4–10 times, and more than 10 times. Scores on items assessing each form of violence were first summed into a composite score. Due to low prevalence, the score for sexual violence was then dichotomized such that a 1 indicated any perpetration and a 0 denoted no perpetration. The composite scores for physical and psychological violence, which were more prevalent but still skewed toward no violence, were recoded into an ordered categorical variables as follows: a composite score of two, indicating a participant had engaged in at least two acts of perpetration or engaged in at least one act two or more times, was coded as “2”; a score of one, indicating a one-time act of perpetration was coded as “1”; and a score of zero, indicating no use of that form of violence in the past year was coded as “0.”.

Sexual Risk Indicators

Inconsistent condom use was assessed based on men’s self-report of condom use with their three most recent sexual partners over the past year. For each partner, men reported how many times they had engaged in sex with the partner over the most recent month of the relationship and how many of these times they had used a condom. Participants were then assigned to one of three categories: consistent use (0), inconsistent use (1), or no use (2), with participants who report using condoms 100% of the time classified as “consistent” users, those who report using condoms greater than 0% and less than 100% of the time classified as “inconsistent” users, and those who reported never using condoms classified as “no use.” This categorical approach was preferred over a continuous variable in the interest of minimizing the effect of recall bias and following recommendations of Noar et al. [38]. Past year number of sexual partners was created based on participants’ reports of the number of people they had sex with in the past year and was scored as 1, 2, or 3 or more partners [the underlying range for this last category was 3–16]. Sexual concurrency was evaluated by self-report of any overlapping sexual partnerships in the past year. Participants were asked to enumerate their three most recent relationships over the past 12 months and to report if they had sex with anyone else during any of these partnerships [39]. A binary indicator was created to denote any concurrency (1) vs. none (0).

Outcome Measures

Outcome measures included indicators of STI infection, perceived HIV risk, and substance use involvement. STI infection was assessed by testing of urine samples for both Chlamydia trachomatis [40] and Neisseria gonorrhoea (NG) using Multiplex PCR. Of the 464 men in the analytic sample who were invited to provide a urine sample, 73% accepted and provided a viable sample (n = 340). Study participants who tested positive for NG and/or CT were considered to have an STI (1), those who tested negative for both NG and CT were considered not to have an STI (0). Perceived risk of HIV was assessed by asking participants “in general, how would you consider your chances of getting HIV/AIDS? Are they small, moderate, great, or no risk at all?” A binary outcome variable was created with responses of no or small risk coded as low risk (0) and responses of moderate or great risk coded as high risk (1). Alcohol intoxication was assessed by asking participants how often they had drunk until intoxicated in the past month and was coded as ever intoxicated (1) or never intoxicated (0). Marijuana use was assessed by asking participants how often they had smoked marijuana in the past 12 months and was coded as ever (1) or never (0) smoked marijuana.

Covariates

Covariates included in analyses of latent class outcomes were age, socioeconomic status, and marital status. Age was calculated based on reported date of birth, or when not available, the reported age in years. Socioeconomic status was evaluated through the Filmer Pritchett Wealth Index [42]. Participants were asked to indicate which of 10 possible assets they owned and a composite score was created by weighting each asset by its factor loading on the first component in a principle components analysis. The weighted composite score for each participant was categorized into terciles based on the entire baseline sample with the lowest tercile classified as “lowest SES” (0), the highest tercile as “highest SES” (2), and the remainder as “middle SES” (1). Marital status was evaluated by asking men if they had ever been married and was coded as (1) if they reported they had ever been married and (0) if they reported never having been married.

Analytic Strategy

We conducted a series of latent class analyses (LCA) in Mplus 7.4 to identify respondents with similar patterns of responses on the six IPV/sexual risk indicators. We first identified the optimal number of classes for the sample by comparing models with increasing number of classes across different statistical fit indices including: the Akaike information criterion (AIC), the sample size adjusted Bayesian information criterion (ssBIC), and the Lo-Mendel-Rubin Likelihood Ratio Test (LMR-LRT). The best-fitting most parsimonious models are those that minimize the AIC and ssBIC and for which adding an additional class results in a significant decrease in model fit as indicated by a p-value of less than 0.05 for the LMR-LRT. We also evaluated classification quality, as indicated by entropy scores, and considered the substantive interpretation of the class profiles [43].

After identifying the optimal unconditional latent class model, we estimated a series of latent class regression models (one for each outcome) using the three-step approach developed by Vermunt [44]. This approach allowed us to examine associations between latent class membership and each outcome while adjusting for both: (1) measurement error due to uncertainty of class classification and (2) the effects of control variables (age, sex, and marital status) on class membership and the outcome measure [44, 45]. A small number of cases (n = 3, < 1%) were missing data on covariates and were thus dropped from covariate adjusted analysis. Full information maximum likelihood (FIML) procedures, which provide for unbiased estimates under the missing at random assumption, were used to deal with missing data on the outcomes, which was minimal (< 2%) for all outcomes except for STI infection, where data were missing by design (described above).

Results

Sample Characteristics

The mean age of participants in the sample was 27 years (range 15–59) and the majority (73%) reported never having been married. Table 1 provides descriptive statistics for the latent class indicators and outcomes. Approximately 21% of participants reported engaging in any (at least one act) psychological IPV perpetration in the past year; 14% reported engaging in any physical IPV perpetration, and 7% reported sexual IPV perpetration. Half (50.1%) of participants reported not engaging in any condom use during sexual intercourse during the past month; 10% reported having had three or more sexual partners in the past year; and 20% reported partner concurrency in the past year. Approximately 28% of participants perceived their chances of contracting HIV as “moderate” or “great”; 16% reported having been intoxicated in the past month; 8% reported past year marijuana use. Of the participants who were tested for STIs (n = 340), 10% tested positive for an STI (Chlamydia and/or NG).

Table 1.

Descriptive Statistics for Latent Class Indicators and Outcomes (n = 965)

	n (%)
Latent Class Indicators
Past year psychological IPV perpetration
Never	765 (79.3)
Once	44 (4.6)
Multiple (≥ 2 acts or times)	156 (16.2)
Past year physical IPV perpetration
Never	826 (85.6)
Once	44 (4.6)
Multiple (≥ 2 acts or times)	95 (9.8)
Past year sexual IPV perpetration
Never	896 (92.8)
One or more times	69 (7.2)
Past month condom use
Consistent use	342 (35.5)
Some use	134 (13.9)
No use	488 (50.6)
Past year number of sexual partners
One	749 (77.6)
Two	117 (12.1)
Three or more	99 (10.3)
Past year partner concurrency
No	773 (80.1)
Yes	192 (19.9)
Outcomes
STI infection^
No	307 (90.3)
Yes	33 (9.7)
Perceived HIV risk
Low (small or no risk)	682 (72.0)
High (moderate or great risk)	265 (28.0)
Past month alcohol intoxication
No	810 (83.9)
Yes	155 (16.1)
Past year marijuana use
No	783 (81.6)
Yes	177 (18.4)

^{^}Among those tested; the number of participants tested for STI infection was n = 340

Identification and Characterization of Latent Classes of IPV/Sexual Risk

A series of latent class models with one to five classes were estimated and compared. The AIC, ss-BIC and the LMR-LRT all suggested that the four-class model provided the best fit to the data (see Table 2). Further, the four-class model provided a clearly interpretable solution with good classification accuracy (entropy = 0.83). We thus selected the four-class solution based on a combination of model fit, parsimony, and interpretability.

Table 2.

Fit Indices for Latent Class Analysis of IPV/Sexual Risk Indicators

N of classes	Free parameters	LL	AIC	ss-BIC	LMR-LRT	Entropy
1	10	−3425.5	6871.0	6888.0	–	–
2	21	−3228.6	6499.1	6534.8	< .001	.77
3	32	−3163.8	6391.5	6445.9	< .001	.79
4	43	-3138.5	6363.1	6436.0	.004	.83
5	54	−3127.6	6363.1	6454.7	.71	.86

The bolded row corresponds to the class solution that was selected

LL = log likelihood; AIC = Akaike information criteria; ssBIC = sample size adjusted Bayesian information criteria; LMR-LRT = Lo-Mendell-Rubin adjusted likelihood ratio test

The parameter estimates for the four-class model are shown in Table 3 and graphically depicted in Fig. 1. Overall, patterns for the four classes were consistent with expectations. The two largest classes were labeled normative (64% of the sample) and IPV only (14% of the sample). These classes were both characterized by a very high likelihood of having had only one sexual partner (item response probabilities were greater than 0.95 for both classes), but were distinguished by the fact that members of the IPV only class were highly likely to have engaged in two or more acts of psychological IPV perpetration (0.58), whereas members of the normative class had a very low likelihood having engaged in any (1 or more acts) psychological (0.05), physical (0.03), or sexual (0.03) IPV perpetration.

Table 3.

Parameter estimates for model of four IPV/sexual risk latent classes

	Latent Class
	Normative	IPV only	Sexual risk only	Comorbid IPV/sexual risk
Latent class prevalences	.64	.14	.13	.05
Item response probabilities
Psychological IPV
Never	.95	.34	.86	.00
One time	.03	.07	.03	.16
Multiple (≥ 2 times/acts)	.02	.58	.11	.84
Physical IPV
Never	.97	.50	.90	.32
One time	.02	.13	.03	.13
Multiple (≥ 2 times/acts)	.01	.36	.07	.55
Sexual IPV
Never	.97	.81	.96	.72
One or more times	.03	.19	.05	.28
Condom use
Consistent condom use	.43	.24	.24	.09
Some condom use	.06	.11	.38	.58
No condom use	.52	.65	.38	.33
# of sexual partners
One partner	.95	1.00	.00	.00
Two partners	.05	.00	.47	.47
Three or more partners	.00	.00	.53	.53
Concurrent partnerships
None	.92	.84	.44	.21
One or more	.08	.16	.56	.79

Item response probabilities greater than .50 bolded to facilitate interpretation. All indicators have a past year reference period

Fig. 1.

Latent class prevalences and item-response probabilities for the four-class model. Note Psychological and physical IPV were coded as: never (reference), once, or two or more times in the past year. Sexual IPV was coded as: never (reference) or ever in the past 12 months. Condom use was coded as: consistent (reference), some use, or no use. Past year number of sexual partners was coded as: 1 (reference), 2, or 3 or more. Past year concurrency was coded as none (reference) or any. Response probabilities for each multi-categorical indicator add to one. Response probabilities for reference categories are not depicted

The third and fourth classes were labeled sexual risk only (13% of the sample) and comorbid IPV and sexual risk (5% of the sample). These classes were both characterized by high probabilities of endorsing having three or more sexual partnerships (0.53 for both classes) and partner concurrency (0.56 and 0.79, respectively), but were distinguished by endorsement of the IPV indicators, with those in the comorbid class being the most likely of any of the classes to endorse frequent psychological (0.84) and physical (0.55) IPV perpetration. Notably, item response probabilities for condom use followed a pattern in which both groups characterized by having only one sexual partner (normative and IPV only) were characterized by a high likelihood of endorsing “no condom use” (0.52 and 0.65, respectively) and relatively low probabilities of endorsing “some condom use” (0.06 and 0.11, respectively). In contrast, the groups characterized by having multiple sexual partnerships (sexual risk only and comorbid) were characterized by somewhat lower probabilities of endorsing “no condom use” (0.38 and 0.33, respectively) and higher probabilities of endorsing “some condom use” (0.38 and 0.58, respectively).

Associations Between Class Membership and STI Infection, Perceived HIV risk, and Substance Use

Table 4 presents findings from models examining the association between class membership and STI infection, perceived HIV risk, and substance use. Consistent with expectations, compared to those in the normative group, those in the comorbid group were more likely to be diagnosed with an STI infection (Wald χ² = 5.77, df = 1, p = 0.016), have high perceived HIV risk (Wald χ² = 16.14, df = 1, p < 0.001), report alcohol intoxication (Wald χ² = 9.20, df = 1, p = 0.002), and report marijuana use (Wald χ² = 22.80, df = 1, p < 0.001). Members of the comorbid group were also significantly more likely to report marijuana use than those in both the IPV (Wald χ² = 4.28, df = 1, p = 0.038) and sexual risk only (Wald χ² = 3.98, df = 1, p = 0.046) groups and more likely to report high perceived HIV risk compared to those in the IPV only group (Wald χ² = 8.34, df = 1, p = 0.004), but not compared to the sexual risk only group (Wald χ² = 1.03, df = 1, p = 0.31).

Table 4.

Predicted probabilities of STI infection, perceived HIV risk, and substance use by latent class membership

Outcome	Class 1 Normative	Class 2 IPV only	Class 3 Sexual risk only	Class 4 Comorbid IPV/sexual risk	Post-hoc pair-wise comparisons
					Comparison	Wald χ2	p
STI infection	.04	.07	.02	.14	4 > 1	5.77	.016
					4 > 3	3.72	.054
High perceived HIV risk	.17	.21	.36	.47	4 > 1	16.14	< .001
					4 > 2	8.34	.004
					3 > 1	17.51	< .001
					3 > 2	5.90	.015
Alcohol intoxication (past month)	.05	.14	.12	.15	4 > 1	9.20	.002
					3 > 1	9.77	.002
					2 > 1	14.03	< .001
Marijuana use (past 12 months)	.09	.21	.20	.38	4 > 1	22.80	< .001
					4 > 2	4.28	.038
					4 > 3	3.98	.046
					3 > 1	9.81	.002
					2 > 1	9.87	.002

Only those pairwise comparisons with p-values < .10 are reported in the table. STI infection was coded as 1 = positive for Chlamydia and/or Gonorrhea, 0 = tested negative for both Chlamydia and Gonorrhea. High perceived HIV risk was coded as 1 = moderate or great perceived HIV risk, 0 = small or no perceived HIV risk. Alcohol intoxication in past month was coded as 1 = one or more times, 0 = never. Marijuana use in the past 12 months was coded as 1 = ever, 0 = never. Models for each outcome included age, SES, and marital status as covariates

Several comparisons among the other classes were also statistically significant. Members of the sexual risk only group were more likely to report high perceived HIV risk compared to those in the IPV only (Wald χ² = 5.90, df = 1, p = 0.015) and normative (Wald χ² = 17.51, df = 1, p < 0.001) groups. Further, compared to those in the normative group, members of both the IPV and sexual risk only groups were more likely to report alcohol intoxication (Wald χ² = 14.03, df = 1, p < 0.001 for IPV only; Wald χ² = 9.77, df = 1, p = 0.002 for sexual risk only) and marijuana use (Wald χ² = 9.87, df = 1, p = 0.002 for IPV only; Wald χ² = 9.81, df = 1, p = 0.002 for sexual risk only).

Discussion

An emerging body of literature in SSA and globally has found a link between IPV perpetration and risky sexual behavior among men. Yet, to date, few studies have examined how multiple dimensions of these behaviors intersect in ways that may uniquely elevate health risks. Research on outcomes of IPV perpetration and sexual risk often focus on single variables to measure these constructs, an approach that ignores multidimensional nature of these behaviors and does not account for the interaction of multiple behaviors that may to be key to understanding differential risk for negative health outcomes. This is of key importance given that typological perspectives suggest that there may be different subgroups of men in the population marked by different patterns of IPV and sexual risk behavior involvement that have different causal origins and sequelae. The current study draws from these perspectives to address these gaps in the literature by using LCA to identify classes characterized by different patterns of IPV perpetration and sexual risk behavior among young men in Tanzania and determine whether and how these patterns are associated with STI infection, perceived HIV risk and substance use indicators.

The study’s major findings are twofold. First, we identified four subgroups of men characterized by distinct profiles of IPV perpetration and risky sexual behavior: normative, IPV only, sexual risk only, and comorbid IPV and sexual risk. That a comorbid pattern emerged, characterized by a high likelihood of involvement in multiple acts of both physical and psychological IPV, as well as multiple sexual partnerships, is consistent with the findings of two studies—one in the US and one in South Africa—that identified a subgroup of men who engage in high levels of IPV and risky sexual behavior. Specifically, Casey et al. (2016) identified a small (8% of the sample) subgroup of men in the US, labeled “misogynistic,” characterized by high risk of engaging in IPV and sexual assault as well as a high number of lifetime sexual partners and one-night stands [29]. Jewkes et al. (2018) identified a subgroup of South African men (25% of the sample), labeled “high violence,” characterized by a high likelihood of involvement in IPV, transactional sex, and having had a concurrent sexual partner [46]. Typological perspectives on IPV and masculinity suggest that this comorbid pattern of behavior may be rooted in patriarchal privilege and resulting hypermasculine social norms that prescribe male dominance in heterosexual relationships as well as sexual prowess and performance [29, 46, 47].

That an IPV only profile emerged, characterized by a relatively high likelihood of involvement in IPV, particularly frequent psychological IPV, but a low likelihood of engaging in multiple sexual partnerships or concurrency, is a novel finding. While little, if any, previous research with men in SSA has examined how different forms of IPV intersect among men, research on IPV victimization among women in SSA suggests that there is heterogeneity in the forms and severity of violence they experience [31, 48, 49]. For example, Reyes et al. (2018) identified two patterns of IPV victimization among South African women; a pattern characterized by the experience of multiple forms of severe IPV and controlling behavior and a pattern characterized by experiencing only moderate forms of IPV [31]. The finding of this profile is also consistent with the notion, put forth by some violence typologies, that there may be a subgroup of men whose IPV behavior is not driven by harmful masculine gender norms (which would also manifest in risky sexual behavior), but rather by “situational factors” that lead to relationship conflict [26, 28].

The finding of a sexual risk only profile also matches to some extent with subgroups identified in the aforementioned studies of men in South Africa and the US [29, 46]. Specifically, Jewkes et al. (2018) found a subgroup (30% of the sample) characterized by high probabilities of partner concurrency and transactional sex but, relative to the “high violence” group that was identified, somewhat lower probabilities of endorsing any physical or sexual IPV perpetration or of having a weapon [46]. Similarly, Casey et al. (2016) study found a small group (4% of the sample), labeled “sex focused,” that had high average numbers of lifetime sexual partners and one-night stands but low risk of involvement in IPV or sexual assault [29]. Men in this group also endorsed relatively low levels of hostility toward women and traditional masculinity compared to other groups. In interpreting this finding, Casey et al. note that “…behaviors [like having multiple sexual partnerships] associated with traditional masculinity may not be equally problematic or hold the same risk across all men” [29].

We also note that condom use did not “travel together” with the other sexual risk indicators (number of sexual partners and concurrency) in terms of its patterning across classes. Specifically, members of the classes characterized by involvement with only one sexual partner (normative and IPV only) were highly likely to endorse not using condoms. Conversely, men in both classes characterized by multiple sexual partnerships (sexual risk only and comorbid) had the highest likelihood of reporting “some condom use.” This is consistent with other research that condom use may decrease with increased partner intimacy and underscores the importance of examining the intersection of multiple dimensions of “risky” sexual behavior for understanding and predicting sexual health outcomes [30, 41].

Our second major finding is that profile membership distinguished risk for negative outcomes. Men in the comorbid group were more likely to have an STI infection, engage in heavy alcohol and marijuana use, and reported high levels of perceived HIV risk than those in the normative group. Marijuana use and high perceived HIV risk were also more likely among men in this group compared to those in the IPV only group. Finally, both marijuana use and STI risk were elevated in this group compared to the sexual risk only group, though the latter finding was only marginally significant. Taken together, these results provide some evidence that being involved in both IPV and multiple sexual partnerships (i.e., a comorbid pattern) denotes elevated health risks for men across a range of indicators. This finding is consistent with the notion, put forth by syndemics theory, that harmful behaviors such as IPV perpetration and sexual risk tend to co-occur in particular contexts or subgroups due to social conditions (e.g., poverty, inequitable gender norms) and exert conjoint negative influences on health, suggesting the importance of targeted and comprehensive treatment and prevention efforts for men in this subgroup as well as their partners. Although more research is needed, we speculate that such prevention and treatment efforts may need to be tailored to address the specific etiological pathways and contextual factors that contribute to this particular behavior pattern. We also note, however, men in the psychological IPV and sexual risk-only groups also reported significantly higher levels of perceived HIV risk and substance use involvement than the normative group, suggesting that men engaging in these patterns of behavior also need targeted treatment and prevention.

Limitations and recommendations for future research

We note several limitations. First, data are cross-sectional, precluding our ability to establish temporality between the behavior patterns and the outcomes examined. Second, because the study was observational and exploration of associations between class membership and health outcomes was exploratory it is plausible that associations detected were spurious. Third, except for STI infection, data were self-reported and thus potentially susceptible to social desirability bias. Fourth, there are other indicators of IPV perpetration and risky sexual behavior that might have further helped to distinguish the patterns identified. For example, theory and empirical research suggest that involvement in transactional sex, controlling behavior, and involvement in other types of violence behaviors, such as economic violence, are key indicators that should be considered [12, 26, 29, 46, 50].

These limitations point to several directions for future research that build on the findings of the current study. Longitudinal research across different populations and settings is needed to replicate the patterns identified and examine whether profile membership predicts future health outcomes among both men and their partners. Such research could include additional indicators that tap into additional dimensions of IPV and sexual risk. In addition, future research is needed to identify the structural, social, and individual determinants that distinguish involvement in different behavioral profiles. Such research would inform whether and how prevention approaches should be tailored to particular subgroups by increasing our understanding of the causal origins that underly the patterns identified. Finally, more research is needed to determine whether the impacts of extant sexual risk and/or violence prevention programs for men depend on their behavioral profiles (indicating the needed for tailored approaches). For example, Gibbs et al. (2020) found that the effects of an IPV prevention intervention differed depending on baseline violence profiles such that effects were found for men in the “high violence” profile, but not for men in “medium” or “low” violence profiles [51].

Conclusion

We provide a powerful and nuanced assessment of the nature of the association between IPV and sexual risk in a sample of Tanzanian men by using a person-centered approach to identify distinct constellations of risk across dimensions of IPV perpetration and sexual behavior. Findings suggest that men who engage in a comorbid pattern of behavior, characterized by involvement in physical and psychological IPV perpetration as well as multiple sexual partnerships, are at particularly high risk for negative health outcomes, including STI infection and substance use, suggesting the need for targeted prevention and treatment efforts for men with this behavioral profile.

Availability of Data

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.