Hurricane impact associated with transactional sex and moderated, but not mediated, by economic factors in Okay, Haiti

Introduction

Natural disasters, such as hurricanes, can have a wide-ranging deleterious impact on the wellbeing and quality of life of affected residents, with variation based on the severity of the particular disaster (Eitzinger, Eugene, Martínez, Navarrete-Frías, & Jarvis, 2019). They commonly include the destruction of local infrastructure, edifices, and agricultural lands, which can result in food shortages, the spread of infectious diseases, and sometimes even chaos and violence (Ferreira, 2016). In addition to the destruction of property, housing, and infrastructure, large numbers of people may be injured or killed as a result of a hurricane (Eitzinger et al., 2019). In particular, households may experience additional economic stress post-hurricane if economic providers for the household are injured or killed, and thus would be unable to work and bring in economic resources.

Haiti was hit by Category 4 storm, Hurricane Matthew, on October 4, 2016. The hurricane affected an estimated 2.1 to 2.4 million Haitians (Eitzinger et al., 2019; United Nations Office for the Coordination of Humanitarian Affairs, 2016), resulting in an estimated 500 to 1000 deaths (546 by official government counts and nearly 1000 by other estimates) (Delva, 2016; United Nations Office for the Coordination of Humanitarian Affairs, 2016) and causing $1.9 billion in damage according to the World Bank and Inter-American Development Bank (Eitzinger et al., 2019). In Okay, the site of this study, the vast majority of buildings (greater than 90%) were destroyed or majorly damaged from the hurricane (Lai & Peçanha, 2016).

Under such post-disaster conditions, research shows that women may engage in sexual risk behaviors, such as transactional sex (Daniel & Logie, 2017; Maclin, Kelly, Kabanga, & VanRooyen, 2015; United Nations High Commissioner for Refugees, 2011). Transactional sex is defined as an implicit exchange of money or goods for sex where the woman feels obligated to have sex because of financial or material gifts (Abels & Blignaut, 2011; Chatterji, Murray, London, & Anglewicz, 2005; Cluver, Orkin, Boyes, Gardner, & Meinck, 2011; R. Jewkes, Morrell, Sikweyiya, Dunkle, & Penn-Kekana, 2012; Maganja, Maman, Groves, & Mbwambo, 2007; Ranganathan et al., 2016) and can occur in any type of sexual relationship (one-time encounter, casual, committed relationships, etc.) (Maganja et al., 2007). Further, transactional sex has been repeatedly linked to negative health outcomes, such as intimate partner violence (K. Dunkle et al., 2006; K. L. Dunkle et al., 2007) and an increased incidence of sexually transmitted infections (STIs) such as human immunodeficiency virus (HIV) (Fielding-Miller, Dunkle, Cooper, Windle, & Hadley, 2016; R. Jewkes, Dunkle, Nduna, & Shai, 2012; R. Jewkes, Morrell, et al., 2012; J. Wamoyi, Stobeanau, Bobrova, Abramsky, & Watts, 2016). Despite the associated risks, transactional sex may be a means for women to compensate for lost income and provide for their households in post-disaster settings, especially if they have lost their livelihoods and/or family members who contributed financially to the household (Maclin et al., 2015; United Nations High Commissioner for Refugees, 2011).

Evidence shows that natural disasters, such as hurricanes, have a disproportionate effect on women, including reduced life expectancy compared to men under the same circumstances (Neumayer & Plümper, 2007; Smyrilli, Silva, Rosado, & Thompson, 2018). In the aftermath of a natural disaster, women often face an increased risk of sexual and gender-based violence (Anastario, Shehab, & Lawry, 2009; Fisher, 2010; Schumacher et al., 2010; Thornton & Voigt, 2007) as well as economic and food insecurity (Miller et al., 2011). Yet, little research has been done to investigate the role of natural disasters on engagement in transactional sex and to assess the possible influence of economic factors on this relationship. One study found that transactional sex was common in the post-disaster conditions in Port-au-Prince, Haiti following the devastating 2010 earthquake (United Nations High Commissioner for Refugees, 2011). Meanwhile, another study found that populations displaced by conflict and living in refugee camps in Tanzania had a much higher prevalence (3.5 times) of transactional sex compared to other local populations not residing in the camp (Rowley et al., 2008). The effects of conflict might be similar to the effects of natural disasters since both can cause wide-spread destruction as well as lack of opportunity for work (Rowley et al., 2008). These two studies provide preliminary evidence that women may be more inclined engage in transactional sexual behavior following a natural disaster than in normal times, but further research is needed.

Both gender inequity and gender-based violence are pervasive in Haiti (Centers for Disease Control and Prevention, Interuniversity Institute for Research and Development, & Comité de Coordination, 2014; Kianersi, Luetke, Jules, & Rosenberg, 2020; Richard, 2018; USAID, 2020). Despite the fact that almost of half of households in Haiti are headed by women, women face systematic exclusion from education, economic activities, and employment (USAID, 2020). Accordingly, women are often economically dependent on men and must choose male partner that will be able to provide for them and their children (Daniel & Logie, 2017). Women also lack legal protections and the existing laws that address gender-based violence are rarely enforced (Horton, 2012). In addition to these notable gender inequities in Haiti, specific post-disaster exacerbations of these issues have also occurred such as gender exclusion in relief efforts, violence, and transactional sex with aid workers and UN Peacekeepers (Bermudez et al., 2019; Horton, 2012; Kolbe, 2015; Vahedi, Bartels, & Lee, 2019). Widely-accepted, traditional gender norms reinforce the expectation that women should have sex with a partner in return for gifts or money that the partner provided them (Kolbe, 2015) and several studies conducted outside of post-disaster settings have found that transactional sex motivated by economic need is common (Daniel & Logie, 2017; Kolbe, 2015; Severe et al., 2014; Smith Fawzi et al., 2005). The effect of gender inequity and gender norms on engaging in transactional sex and the role that men play in demanding such behavior from women has motivated the development of some interventions that aim to erode male support for such dynamics, where women may lack agency to refuse sex, and reduce their perpetration of violence against women (Gibbs et al., 2020; Rowley et al., 2008).

As such, understanding the relationship between natural disasters and transactional sex is critical to improving public health, particularly in regions that are both impoverished and frequently effected by natural disasters. In this study in Okay, Haiti, we evaluate whether experiencing more severe impact from Hurricane Matthew in 2016 was associated with engaging in transactional sex and whether economic factors, such as economic stress, loss of income generating resources, food insecurity, and poverty, mediated or moderated this relationship. We hypothesize that (1) women who experienced more severe hurricane impact would be more likely to engage in transactional sex and (2) this effect would be significantly mediated or moderated though economic variables, including loss of income generating resources, food insecurity, and poverty. A finding of mediation would indicate that economic factors explain the association between the exposure to hurricane impact and the transactional sexual behavior. A finding of moderation of this relationship would imply that women’s baseline economic characteristics influenced the strength and direction of the relationship between hurricane impact and transactional sex. Economic mediation or moderation of the relationship would be important to inform intervention design, specifically the timing of disaster resilience interventions and when they might be most effective. Mediation would indicate that post-disaster interventions might be most effective at minimizing the effect of hurricane impact on transactional sex while moderation would imply that interventions should pre-emptively build disaster resilience before the hurricane hits.

Methods

Study population

This study was conducted in the southern peninsula of Haiti among female microfinance clients of the microfinance institution, Fonkoze. Fonkoze is Haiti’s largest microfinance organization, has branches across the country, targets women almost exclusively, and has been serving microfinance clients since 1994 (M. Tucker & Tellis, 2005). This study was conducted among their clients served by the branch office in Okay, Haiti. The Okay branch office serves over 6,500 clients in the city of Okay and surrounding municipalities. The study population was randomly sampled from the Fonkoze membership database of the Okay branch and the eligibility criteria for participants included that they were (1) a woman, (2) a microfinance client served by the Fonkoze Okay branch office, and (3) between the ages of 18 and 49 years old.

Data collection

Study data were collected using a cross-sectional, tablet-based survey design between December 2017 and February 2018. Before the data collection commenced, the entire survey instrument was translated into Haitian Creole and then back-translated into English to confirm accuracy. Trained local fieldworkers then collected survey data in the local language, Haitian Creole, among a random sample of 304 female microfinance clients. The surveys were completed between December 2017 and February 2018, approximately one year after the area Hurricane Matthew. Fieldworkers approached and recruited potential participants at their homes. Those who met the eligibility criteria, agreed to participate, and completed a written consent form were asked to self-report data on their socio-demographic characteristics, sexual behaviors, and experiences after Hurricane Matthew. This study was approved by the Human Subjects Office at authors’ institution (Protocol #1705661852).

Measures

Exposure.

Our primary exposure, hurricane impact, was measured in two ways: (1) self-reported injury or death in the household as a result of the hurricane (referred to as hurricane impact) and (2) household distance from the hurricane path. Hurricane-related death or injury has been used in previous studies to assess hurricane impact, particularly the effect of hurricane impact on psychological outcomes (Briere & Elliott, 2000; Goenjian et al., 2001). We used this measure to achieve an individualized operationalization of the severity of hurricane impact. We used the distance of the household from the hurricane’s path (for those who maintained their same households at the time of survey completion) to verify our self-reported measure of injury or death from the hurricane. Distance from the hurricane path was calculated using household geographical coordinates collected at the time of survey completion and the distance of these coordinates from the path of the hurricane (in miles) in ArcGIS software (ESRI, 2011). Proximity to the hurricane path or zone of destruction is also commonly employed as a measure of hurricane exposure and, importantly, does not rely on self-report (Furr, Comer, Edmunds, & Kendall, 2010; Goenjian et al., 2001). However, distance to hurricane path may not necessarily capture the variation in severity of devastation experienced by households. We conducted sensitivity analyses to assess whether our self-reported hurricane impact variable was associated with distance to the hurricane path. Our analyses demonstrated that hurricane impact (injury or death) was significantly associated with distance of household to the hurricane path [Prevalence ratio (PR) per 10 miles: 0.83; 95% Confidence Interval (CI): 0.73, 0.95] (Molly Rosenberg, 2019).

Outcome.

The primary outcome of interest in this study was transactional sex, which was defined as the self-reported receipt of money or gifts from the most recent sexual partner in the last year and feeling obligated to have sex with this partner in return for received money or goods. This broad assessment is inclusive of both formal and informal transactional sex, but aims to capture more informal transactional sex that occurs in both casual and committed relationships (Maganja et al., 2007). This measure has been previously employed in HIV prevention in various resource-limited settings (Abels & Blignaut, 2011; Chatterji et al., 2005; Cluver et al., 2011; R. Jewkes, Morrell, et al., 2012; Ranganathan et al., 2016; J Wamoyi, Stoebenau, Kyegombe, Heise, & Ranganathan, 2018). In the creation of this variable, individuals were who reported no sex partner in the last year were coded as not having transactional sex. This variable was based on three queries in the survey about transactional sex behavior (See Appendix A). The final variable was coded dichotomously.

Moderators.

Three economic moderators were assessed in regression models. The first potential moderator was food insecurity, which was measured using the Household Hunger Scale (HHS) (Ballard, Coates, Swindale, & Deitchler, 2011; Deitchler, Ballard, Swindale, & Coates, 2010, 2011), an internationally validated scale designed to measure household food insecurity across different cultures, and dichotomized low food insecurity (i.e. little to no hunger) versus moderate/severe food insecurity (i.e. moderate hunger to severe hunger) according to the scale’s scoring system. The second moderator was poverty and it was defined based on the value of the assets held by the household. This was measured by summing of the current value (in Haitian gourdes) of 20 possible household items and dichotomized at the median as low poverty versus high poverty. Lastly, the final moderator was loss of income generating resources, which was a binary variable derived from a question that queried whether income generating materials or goods were destroyed in the hurricane (yes/no).

Mediators.

Two economic mediators were assessed in the final structural equation model. The first was loss of income generating resources and was measured the same as described above. The second mediator was a latent variable, economic stress, and was measured by four observed factors: (1) household size (defined as number of people living in participant’s household), (2) food insecurity (measured the same as described previously), (3) lack of meat in diet, an indication of economic disadvantage (R.; Jewkes, Nduna, Jama-Shai, Chirwa, & Dunkle, 2016) (reverse coded from a query that asked whether participants had meat in their diet (yes/no)), and (4) poverty in assets (defined as the log-transformed sum of the current value (in Haitian gourdes) of 20 possible household items, reverse coded so that the resulting factor loading in our model was in the same direction as the other variables).

Covariates.

Other covariates queried in the survey and reported in the sociodemographic table included age (in years), marital status (married/living together, divorced/separated, never married), ever attend school (yes/no), education level (preschool, primary, secondary, more than secondary), literacy (yes/no; defined as ability to read a sentence in Creole), household size, and household asset quartiles (defined sum of the current value (in Haitian gourdes) of 20 possible household items and categorized in quartiles, from lowest (Q1) to highest (Q4)).

Statistical Analysis

First, we used log-binomial models to assess the relationship between hurricane impact and the binary outcomes of transactional sex, food insecurity, and poverty. Next, we used log-binomial models to evaluate the moderating effects of three economic variables by including interaction terms between the hurricane impact and (1) food insecurity, (2) poverty, and (3) destruction of income generating materials in three separate models. The objective was to assess whether the effect estimates for the two different resource strata in each model were statistically different from each other. Wald test p-values for the interaction terms, with an a priori cut point of p<0.10, were used to determine statistical significance (Durand, 2013).

Finally, we specified a structural equation model based on several models (Bollen, 1989). We began with a structural equation model with only observed variables including loss of income generating resources and food insecurity as mediators of the relationship between hurricane impact (distance from hurricane path and death or injury in family) and transactional sex. Next, in order to account for a greater number of economic stressors than just food security, we tested a possible latent variable to estimate the experience of economic stress. We ran a confirmatory factor analysis for this latent variable in an effort to evaluate the appropriateness of our indicators for the potential latent variable. For the economic stress latent variable, we eliminated non-significant indicators and then tested the latent variable in our structural model, replacing food security with the latent variable of economic stress (of which food security was an indicator). We used the following criteria to assess the fit of our specified model to our data: (1) a non-significant Chi-square p-value (p-value>0.05); (2) a comparative fit index (CFI) value greater than 0.90; (3) a Tucker-Lewis Index (TLI) value greater than 0.90; and (4) a root mean square error of approximation (RMSEA) value less than 0.050 (Brown, 2015; Steiger, 1990; L. R. Tucker & Lewis, 1973). This second model fit the data well and was chosen as our final model. Our path model for the final model is displayed in Figure 1. In order to scale the latent variable, we set its variance equal to 1. The two tested models were both recursive (i.e. did not contain correlated errors nor feedback loops) and, as a result, they were identified. In this study, we were primarily interested in the indirect, or mediated, relationships between hurricane impact (injury or death) and transactional sex. In the final model, we used bootstrapping to obtain appropriate standard errors for the indirect effects between these two variables (presented in Table 4). All statistical analyses were conducted using R version 3.4.1 (“Single Candle”) (R Core Team, 2020) and SAS 9.4 (Cary, NC) (SAS Institute Inc., 2014). For the structural equation modelling, we utilized the lavaan package in R (Rosseel, 2012).

Figure 1.

The association between hurricane impact and transactional sex, testing mediation through economic factors such as loss of income generating resources and the latent variable, economic stress, among female microfinance members in Okay, Haiti (N=304)

*Bolded estimates indicate significance at an a priori α=0.05.

Table 4.

Mediating pathways between hurricane impact and transactional sex among female microfinance clients in Okay, Haiti (N=304)

Mediating pathway	Estimate	p-value
Hurricane impact-> loss of income generating resources-> transactional sex	0.008	0.82
Hurricane impact-> loss of income generating resources-> economic stress-> transactional sex	−0.005	0.52
Hurricane impact-> economic stress-> transactional sex	−0.056	0.39

Results

Demographically, our sample ranged in age from 20 to 49 years old, with a mean age of 36 years (SD=7.7). Most of the participants (55%) were married or cohabiting with partner. Further, most (83%) had attended some schooling with 44% having completed secondary school or more, though few had obtained education past the secondary level. The average household size was 6 people (mean=5.9, SD=1.9). Nearly all of the participants had children (aged 17 or younger) in their household with the majority (57%) having one to two children living in the household. Almost 60% of women reported engaging in transactional sex in the last year (Table 1).

Table 1.

Characteristics of the study population of female microfinance clients in Okay, Haiti (N=304)

		Hurricane impact (injury or death)
	Total	Yes	No	p-value1
	N=304	N=117	N=184
Socio-demographic characteristics
	N (%)	N (%)	N (%)
Age				0.37
20–29	69 (22.7)	22 (18.8)	47 (25.5)
30–39	127 (41.8)	53 (45.3)	73 (39.7)
40–49	108 (35.5)	42 (35.9)	64 (34.8)
Missing	0	0	0
Marital status				0.19
Currently married or living as married	164 (55.0)	59 (51.8)	104 (57.5)
Divorced/separated	26 (8.7)	7 (6.1)	18 (9.9)
Never married	108 (36.2)	48 (42.1)	59 (32.6)
Missing	6	3	3
Education				0.89
None	51 (16.9)	21 (18.1)	29 (15.9)
Preschool/Primary	119 (39.5)	45 (38.8)	73 (40.1)
HS or more	131 (43.5)	50 (43.1)	80 (44.0)
Missing	3	1	2
Literacy				0.94
Yes	223 (74.1)	85 (73.9)	136 (74.3)
No	78 (25.9)	30 (26.1)	47 (25.7)
Missing	3	2	1
Household asset quartile2				<0.0001
Q1	75 (24.8)	50 (42.7)	25 (13.6)
Q2	76 (25.2)	26 (22.2)	50 (27.2)
Q3	75 (24.8)	24 (20.5)	51 (27.7)
Q4	76 (25.2)	17 (14.5)	58 (31.5)
Missing	2	0	0
Study variables
	N (%)	N (%)	N (%)
Food insecure
Yes	218 (72.2)	101 (86.3)	117 (63.6)	<0.0001
No	84 (27.8)	12 (13.7)	67 (36.4)
Missing	2	0	0
Income generating resources destroyed				<0.001
Yes	249 (82.7)	108 (92.3)	141 (76.6)
No	52 (17.3)	9 (7.7)	43 (23.4)
Missing	2	0	0
No meat in diet				0.55
Yes	111 (37.1	46 (39.7)	65 (35.5)
No	188 (62.9)	70 (60.3)	118 (64.5)
Missing	2	1	1
Transactional sex (n=271)3				<0.01
Yes	157 (58.4)	60 (52.6)	52 (33.5)
No	112 (41.6)	54 (47.4)	103 (66.5)
	Mean (SD)	Mean (SD)	Mean (SD)
Household size	5.9 (1.86)	6.1 (1.93)	5.9 (1.83)	0.36
Distance from hurricane	34.8 (5.5)	33.5 (5.0)	35.7 (5.7)	<0.001
Household assets (in Haitian gourdes)4	15828.9 (27056.3)	10707.5 (25769.5)	18682.4 (26983.5)	0.01

¹For categorical variables, we utilized Pearson’s chi-square test and Fisher’s exact test for frequencies with small cell sizes (i.e. those containing <5 observations) to assess differences in frequencies between the two strata of hurricane impact. For continuous variables, we utilized Student’s t-test was used to compare the differences in means between the two strata of hurricane impact.

²Measured by adding up the self-reported value (at time of purchase) of 20 key items in participant’s household.

³Those without a sex partner in the last year are coded as not having transactional sex.

⁴Measured by adding up the self-reported value (at current time) of 20 key items in participant’s household.

We found participants who experienced hurricane impact (family death or injury) were significantly more likely to have engaged in transactional sex, to have experienced recent food insecurity, and to be more impoverished compared to participants with lower hurricane impact. Specifically, women who experienced hurricane impact were 58% more likely to have engaged in transactional sex [prevalence ratio (PR) (95% confidence interval (95% CI)):1.58 (1.19–2.09)]; 36% more likely to have experienced recent food insecurity [PR (95% CI): 1.36 (1.19–1.55)]; and 59% more likely to report living in poverty compared to women who did not experience hurricane impact [PR (95% CI): 1.59 (1.28–1.98)] (Table 2).

Table 2.

Relationship between experiencing hurricane impact (injury or death) and transactional sex, food security, and poverty (in assets) among female microfinance clients in Okay, Haiti (N=304)

Outcomes	PR	(95% CI)	p-value
Transactional sex	1.58	(1.19, 2.09)	0.0015
Food insecurity	1.36	(1.19, 1.55)	<.0001
High poverty1	1.59	(1.28, 1.98)	<.0001

¹Poverty measured using the summed value of the participant’s current assets (in Haitian gourdes) and was dichotomized at the median as low poverty versus high poverty.

We identified significant moderation of the relationship between hurricane impact and transactional sex by both food insecurity and poverty (Wald p-values of 0.059 and 0.010, respectively). The relationship between hurricane impact and transactional sex was both significant and strong among women who reported food insecurity in the previous month [PR (95% CI): 1.91 (1.34–2.71)], but not among those who reported food security in the previous month [PR (95% CI): 0.84 (0.39–1.83)]. Similar moderating effects were observed by the variable of poverty. The relationship between hurricane impact and transactional sex was both significant and strong among women who reported poverty [PR (95% CI): 2.12 (1.14–3.21)], but not among those who reported low poverty [PR (95% CI): 0.87 (0.50–1.50)] (Table 3). We also ran a model assessing the moderating effects of the destruction of income generating materials but found that this variable did not significantly moderate the relationship between hurricane impact and transactional sex. However, these findings displayed similar trends as seen with food insecurity and poverty with the relationship between hurricane impact and transactional sex being strong and significant only among those who reported a loss of income generating resources.

Table 3.

Relationship between experiencing hurricane impact (injury or death) and transactional sex moderated by food insecurity, poverty (in assets), and destruction of income generating materials among female microfinance clients in Okay, Haiti (N=304)

I. Food insecurity
	Yes			No
	PR	95% CI	p-value	PR	95% CI	p-value
Transactional sex	1.91	(1.34, 2.71)	0.0003	0.84	(0.39, 1.83)	0.6668
Wald p-value for interaction term1						0.0593
II. High poverty2
	Yes			No
	PR	95% CI	p-value	PR	95% CI	p-value
Transactional sex	2.12	(1.41, 3.21)	0.0004	0.87	(0.50, 1.50)	0.6080
Wald p-value for interaction term1						0.0102
III. Destruction of income generating materials
	Yes			No
	PR	95% CI	p-value	PR	95% CI	p-value
Transactional sex	1.51	(1.13, 2.04)	0.0061	1.30	(0.44, 1.50)	0.6383
Wald p-value for interaction term1						0.7850

¹An a priori significance cut point of p<0.10 was used for Wald test p-values for the interaction terms.

²Poverty measured using the summed value of the participant’s current assets (in Haitian gourdes) and was dichotomized at the median as low poverty versus high poverty.

Our final structural model fit the data well (with fit: χ²=19.700, degrees of freedom=15; CFI=0.958; TLI=0.941; RMSEA=0.040). The results of the structural equation analysis (Figure 1 and Table 4), indicated that the indirect pathways between hurricane impact and transactional sex through the economic variables of loss of income generating resources and economic stress were not significant. Additionally, neither of our economic variables (income generating resources and economic stress) were significantly associated with transactional sex. However, the association between hurricane impact and transactional sex remained significant while controlling for these economic factors. Further, there were other significant relationships in our model. In terms of the measurement part of the model, all the indicators of economic stress positively and significantly loaded on to the latent variable (Figure 1, Table 4).

There were also significant relationships between all of the following variables: (a) distance to the hurricane and hurricane impact (household injury or death), (b) hurricane impact and transactional sex, and (c) hurricane impact and economic stress. As in our simple log-binomial regression models, the SEM model supported a strong and significant relationship between hurricane impact and transactional sex. Meanwhile, the relationship between distance from the hurricane and hurricane impact was both significant and strong. For every mile closer that the respondent’s house was to the hurricane path, women had 1.05 times the odds of experiencing a household death or injury, which corresponds to a 5% increase in odds. Experiencing hurricane impact also had a significant effect on economic stress, with those that had experienced hurricane impact having a 0.46 standard deviation (15.7%) increase in the latent variable of economic stress compared to those who had not experienced hurricane impact. However, the most important findings from the SEM model remain that (1) hurricane impact was significantly associated with transactional sex even while controlling for the economic factors in the study, (2) the mediated pathways between hurricane impact and transactional sex through key economic factors were not statistically significant, and, finally, (3) the economic variables in the study were not individually associated with transactional sex in our model (Figure 1).

Discussion

In this study in Okay, Haiti, we found that women who experienced hurricane impact (household death or injury) were significantly more likely to have engaged in transactional sex compared to those not impacted. This relationship was significantly moderated by two economic variables, food insecurity and poverty. However, the relationship was not mediated by economic variables—loss of income generating resources or economic stress, a latent variable composed of the following factors, household size, food insecurity, lack of meat in diet, and poverty (in assets). Further, these economic mediators were not significantly independently associated with transactional sex in the SEM model. However, the direct relationship between hurricane impact (household death or injury) and transactional sex remained significant even while controlling for possible mediation through the economic factors.

Our findings that food insecurity and poverty compounded the effect of the hurricane on engaging in transactional sexual behavior are consistent with existing research that demonstrates that individuals living in poverty are less able to absorb economic shocks associated natural disasters (Carter & Barrett, 2006; Carter, Little, Mogues, & Negatu, 2007; Dercon, 2004). In line with this body of research, those who have more economic resources, such as more wealth and more food security, are able to buffer themselves and their households from some of the negative effects of hurricane impact, such as sexual risk-taking and transactional sex (Dercon, 2004). In contrast, those who are more impoverished, informally employed, and lack financial savings and safety nets may feel intense economic pressure post-hurricane and, therefore, be more likely to engage in transactional sexual behaviors to provide financially for their households. Empirically, research has linked poverty to increased sexual risk-taking and transactional sexual behavior (Booysen & Summerton, 2002; Pascoe et al., 2015). This type of transactional sex has often been referred to as sex for survival (Hunter, 2002; Miller et al., 2011; United Nations High Commissioner for Refugees, 2011). Further, in addition to the link between poverty and transactional sex, food insecurity has also been consistently associated with risky sexual behaviors, including transactional sex, among women in low resource settings (Chop et al., 2017; K. Dunkle et al., 2004; Miller et al., 2011; Oyefara, 2007; Pascoe et al., 2015; Weiser et al., 2007). Transactional sex may be particularly risky when it occurs in more casual relationships. As such, we conducted a sensitivity analysis of the effect of hurricane impact on transactional sex among women who were not married or cohabitating with partner. This sensitivity analysis revealed that the relationship between hurricane impact and transactional sex was stronger among this group [PR (95% CI): 3.00 (1.64, 5.48)], indicating that interventions should be specifically focused on women who are engaged in more casual relationships compared to women in committed relationships. Finally, other studies have found that sexual risk-taking in the form of transactional sexual behavior may be especially likely to occur in times of economic desperation and, particularly, among the poorest women. Consistent with this research, our findings indicate that women with more resources may be protected from engaging in transactional sex in post-disaster settings in ways that those with fewer resources are not.

While engaging in transactional sex has historically been conceptualized as a decision borne out of economic desperation (i.e. for survival and subsistence needs) (Hunter, 2002; Miller et al., 2011; United Nations High Commissioner for Refugees, 2011), it is likely that the motivations for engaging in transactional sex are far more complex (Fielding-Miller, Dunkle, Cooper, et al., 2016; Fielding-Miller, Dunkle, Jama-Shai, et al., 2016; Masvawure, 2010; Ruark et al., 2014; Stoebenau et al., 2013). Our findings that the relationship between hurricane impact and transactional sex was not mediated by economic factors, loss of income generating resources and economic stress, seem to support the idea of a more complex relationship.

Emerging research also suggests that the motivations for engaging in transactional sex may not be simply economic, but that social or cultural factors may play a significant role in the decision-making process (Fielding-Miller, Dunkle, Cooper, et al., 2016; Fielding-Miller, Dunkle, Jama-Shai, et al., 2016; Masvawure, 2010; Ruark et al., 2014; Stoebenau et al., 2013). According to this body of research, transactional sex can act as a way for male partners to express their affection (Fielding-Miller, Dunkle, Cooper, et al., 2016; Ruark et al., 2014), demonstrate or solidify a long-term relationship (Fielding-Miller, Dunkle, Jama-Shai, et al., 2016), or for women to obtain luxury or other consumer goods or improve their social status (Masvawure, 2010; Stoebenau et al., 2013). Further, the perception of the risks of engaging in transactional sex may play an important role in women’s motivations for engaging in such behaviors. There are differing degrees of cultural acceptance of transactional sexual behaviors, with many cultures endorsing transactional sexual behaviors between romantic partners and, particularly so, in heterosexual coupling where the male partner is often expected to pay for things and give gifts as part of courtship. Where transactional sexual behaviors are widely accepted culturally, the perception of risk of transactional sex may be significantly attenuated and motivations for engaging in transactional sex may significantly diverge from more formalized sex work. Future studies should assess how culturally acceptable transactional sexual behaviors are within the context of sexual and romantic relationships, the influence of gender equity (or lack thereof), and how these two factors impact women’s decision-making regarding transactional sex.

While motivations for engaging in transactional sexual behavior are likely multifactorial and economic factors may not solely mediate the relationship between hurricane impact and transactional sex, the moderating effects described previously demonstrate that economic resources, or lack thereof, is an important component of this relationship. Specifically, these findings indicate that proactive economic interventions to build disaster resilience before natural disasters occur may be more efficacious than post-disaster, economic relief efforts in reducing and preventing engagement in transactional sex among women.

Many studies have studied transactional sex outside of disaster settings but few have tracked them in post-disaster settings (Abels & Blignaut, 2011; Chatterji et al., 2005; Cluver et al., 2011; R. Jewkes, Morrell, et al., 2012; Ranganathan et al., 2016). However, this study, and the only other study that examined transactional sex following a natural disaster, suggest that the post-disaster conditions may significantly increase the prevalence of such behaviors (United Nations High Commissioner for Refugees, 2011). Accordingly, women may be engaging in transactional sex in order to compensate for economic losses related to the hurricane, such as the destruction of their means of income generation, loss (injury or death) of an economic provider to the household, damage to food sources and crops, or other immediate and extreme economic devastation caused by the hurricane. While the lack of mediation effects by economic factors on the relationship between hurricane impact and transactional sex seems to dispute this, the moderation of the relationship indicates that there is some economic component. Second, the measure of hurricane impact utilized in this study (household death or injury) may affect women in ways that are not purely economic. For example, it may be that grief prevents women from engaging in formal employment and, thus, increases their likelihood of engaging in transactional sexual behavior. On the other hand, grief may also decrease individuals’ capacity to assess the potential consequences of behaviors and, thus, lead to increased engagement in risk-taking or risky sexual behaviors such as transactional sex. Little research has been done in this area, but a few studies suggest that traumatic life events and bereavement may increase the likelihood of engaging in sexual risk-taking behaviors and sexually transmitted infection risk (Bond et al., 2016; Mayne, Acree, Chesney, & Folkman, 1998; Thurman, Brown, Richter, Maharaj, & Magnani, 2006). These studies identify several possible reasons that bereaved individuals may be more likely to engage in sexual risk-taking including an increase in bereavement-related psychiatric conditions (specifically, depression and substance use) which might play a partially mediating role (Bond et al., 2016; Thurman et al., 2006), sexual behavioral changes to mitigate distress associated with bereavement (Mayne et al., 1998; Thurman et al., 2006), and survivor’s guilt leading to increases in risky sexual behaviors (Mayne et al., 1998). The significant association between our measure of hurricane impact, death or injury in household, and transactional sex are consistent with research suggesting that women may be specifically more vulnerable to transactional sex after the death of a male partner (Miller et al., 2011). However, while we can speculate, the mechanisms for this relationship have yet to be explained because economic factors were not found to be mediators in this analysis. This lingering knowledge gap is an opportunity for future research and investigation.

There are several limitations to this study that should be considered when interpreting the study results. First, the data is cross-sectional in nature and, therefore, we cannot fully disentangle the temporal ordering of the key variables of interest. However, transactional sex is queried within the context of the most recent partner in the last year and, thus, only relationships that began or continued after the hurricane contributed positive cases to the outcome of transactional sex in our analysis. This operationalization allows for more accurate temporal reference for the occurrence of the outcome. However, future studies with longitudinal designs should be conducted in order be able to distinctly separate exposure and outcome.

Additionally, our measure of hurricane impact—injury or death in household—may not have fully and accurately measured the experience of devastation from the hurricane. For example, households may have experienced significant damage to their house, property, or other belongings despite not having experienced a death or injury in the household. However, we accounted for this by the inclusion of both hurricane impact and distance of household to the hurricane path in the SEM model. Further, these two variables were highly correlated, thus validating our hurricane impact measure with an objective measure, distance of household to the hurricane path.

Similarly, we may have not captured the full gamut of economic variables at play in the relationship between hurricane impact and transactional sex. Though our results indicate that this pathway was not mediated by economic factors, it is possible that we accidentally excluded one or more economic variables that act as a significant mediators of the relationship. However, our collection of economic data from respondents was comprehensive and also included detailed post-hurricane assessment of economic losses.

The survey was also conducted only among female microfinance members in one region of Haiti and, thus, this sample may not be representative the overall female population living in Haiti. Therefore, our findings may not be generalized to other populations of Haitian women. Microfinance members may differ from non-members on a number of individual and demographic characteristics. Previous comparison studies have observed significant differences between microfinance members and non-members in marital status, age, gender, having children (aged 17 or less), but were not significantly different on household assets (Deyoe et al., 2020; M Rosenberg et al., 2020). Other studies have documented increases in income, expenditure and saving, assets, and property between pre- and post-membership in microfinance groups (Pati, 2011). However, we used a rigorous random sampling method with very high participation rates, maximizing the likelihood that the study population was representative of the target population of female Fonkoze clients served by the Okay branch office. Finally, the measures are all based on self-reported survey responses. Self-report, though a valid method of data collection, can affect results if under-reporting or over-reporting occur due to social desirability and recall bias. However, the surveys were collected in an interview setting by trained and experienced fieldworkers, which should contribute to the reliability and accuracy of survey responses. Future studies should consider evaluating more objective measures or use more self-controlled data collection procedures such as Audio Computer-Assisted Self-Interviewing (ACASI), which may reduce bias.

Conclusions and implications

Given the recognized risks associated with transactional sex, including intimate partner violence (K. Dunkle et al., 2006; K. L. Dunkle et al., 2007; R. Jewkes, Morrell, et al., 2012) and STIs such as HIV (Fielding-Miller, Dunkle, Cooper, et al., 2016; R. Jewkes, Dunkle, et al., 2012; J. Wamoyi et al., 2016), it is important to understand how hurricanes impact risk of engaging in transactional sexual behavior. Additionally, this research has important policy implications since these findings indicate that interventions to prevent transactional sex and the associated risks should be primarily directed toward efforts to build resilience before disaster strikes. To this end, economic development and empowerment programs, such as microfinance, may increase the capacity of women to weather the impact of hurricanes, thereby, potentially reducing economically-motivated transactional sex. Better understanding the effects of hurricanes, and more broadly natural disasters, on sexual health and behaviors among women in resource-poor settings and the interactions between individual experiences and contextual factors, will allow for the better design, development, and implementation of post-disaster humanitarian assistance programs and public health campaigns. Specifically, research focused on the effect of gender inequities, available economic opportunities for women before and after natural disasters, and how experiencing a natural disaster might intensify these dynamics are needed.

Highlights.

• Participants who experienced hurricane impact were 58% more likely to have engaged in transactional sex

• Food insecurity and poverty intensified the effect of hurricane impact on transactional sex

• Economic factors were not on the pathway between hurricane impact and transactional sex

• At-risk populations should be preemptively targeted with economic interventions to build resilience before disaster strikes

• Interventions following a disaster may not effectively reduce transactional sexual behavior in post-disaster settings