Associations between hurricane exposure, food insecurity, and microfinance; a cross-sectional study in Haiti

Abstract

Natural disaster and food insecurity are prevalent in Haiti. Natural disasters may cause long-term food insecurity. Microfinance programs may provide resilience against this outcome. The objectives of this study were 1) to assess the association between the impact of Hurricane Matthew and long-term food insecurity and 2) to understand whether this association varies by participants’ membership in a microfinance program. In 2017–2018, we interviewed 304 Haitian female microfinance clients. We used log-binomial regression to evaluate the association between hurricane Matthew impact and long-term food insecurity, with evaluation of effect modification by timing of microfinance exposure. We found that one year after the hurricane, participants who were severely impacted by the hurricane were more likely to report poor dietary diversity and moderate to severe household hunger, compared to the less severely impacted participants. Both associations became insignificant among those who received their first microfinance loan before the hurricane. Natural disasters like hurricanes are associated with long-term food insecurity at individual and household levels. Microfinance programs might improve post-hurricane long-term food security.

1. Introduction

Food insecurity, described as limited access to sufficient, nutritious, and safe food (Anderson, 1990; Hoddinott & Yohannes, 2002), contributes to many forms of malnutrition, including undernutrition (Food and Agriculture Organization of the United Nations, 2018) and can adversely affect public health. It weakens the immune system and indirectly increases the incidence of negative health outcomes causing an increment in different diseases and public health burden (Nelson & Williams, 2014). On a slightly upward trajectory since 2015, hunger and undernourishment influenced more than 821 million people in 2018, globally (Food and Agriculture Organization of the United Nations, 2018). Haiti has faced the challenge of food insecurity for decades, which results in the country’s high ranking in the 2015 Global Hunger Index (USAID, 2017). Food insecurity affects more than 50% of Haitian population (USAID, 2017).

Poverty is the strongest associative factor in the development of undernourishment and nutrient deficiency in Haiti. According to the World Bank, 59% of the population live under $2.42 per day (The World Bank, 2019), contributing to chronic hunger. Natural disasters such as hurricanes have added another dimension to this issue. In impoverished environments, individuals often reside in areas with greater risk for natural disasters, which can reinforce their vulnerability when a disaster strikes. Over the past two decades, earthquakes, floods, droughts, and hurricanes have intensified Haiti’s food insecurity by limiting food production and increasing food prices (L. H. Marcelin, Cela, & Shultz, 2016; USAID, 2019). Hurricane Matthew struck Haiti as a Category 4 hurricane on October 4, 2016, affecting over 2 million people (The World Bank, 2017), with a death toll of more than 500 (Louis Herns Marcelin & Cela, 2017). The disaster had long-term consequences; Six months after the hurricane, more than 90% of affected communities in Sud departments did not have a suitable place for living and more than 70% of them lacked access to drinkable water (Louis Herns Marcelin & Cela, 2017). The destruction of the country’s agricultural system due to Hurricane Matthew was widespread, and in some areas, up to 90% of crops were damaged (The World Bank, 2017). At individual level, deleterious consequences of severe natural disasters might restrict a household’s access to food resources because of increased food prices, damage to income-generating assets, and injury or death of household’s primary wage earner (USAID, 2019).

Despite their seemingly intuitive relationship, few studies have quantified the direction and magnitude of the association between natural disasters, specifically hurricanes, and long-lasting post disaster food insecurity. To remain consistent with the previous literature, we use the phrase “long-term” to refer to continued (more than a year) post-disaster food insecurity (Clay, Papas, Gill, & Abramson, 2018). Previous descriptive studies have reported the existence of short-term food insecurity prevalence following natural disasters in Haiti. A longitudinal household survey in Haiti’s national capital, Port-au-Prince, found that, a few weeks after an earthquake in 2010, 19% of participants suffered from severe food insecurity and more than 50% of participants reported moderate to severe food insecurity (Hutson, Trzcinski, & Kolbe, 2014; Kolbe, et al., 2010). Similarly, in an inferential study, Rukundo et al. evaluated food insecurity after a landslide disaster in Eastern Uganda, finding that food insecurity intensified among the affected individuals (Rukundo, et al., 2016). However, there is a gap in the literature regarding long-term effects of natural disasters on food security (Cartwright, Hall, & Lee, 2017). To our knowledge, no previous research has studied the long-term associations between hurricane exposure and food insecurity in the context of Haiti. In addition, to alleviate the hunger issue in Haiti, it is of great importance to find sustainable remedies that can strengthen Haiti’s rehabilitation capabilities in response to natural disasters.

Participation in microfinance programs may provide resilience against the effects of natural disasters (Meyer, 2010). Microfinance programs are designed to provide small amounts of financial support to low-income individuals to invest in income-generating projects (M. S. Rosenberg, Seavey, Jules, & Kershaw, 2011). Through these income-generating projects, participants may be able to save income beyond what is needed immediately to support their households. As a result, these savings could be drawn on to maintain food and nutrition under unfavorable situations. In other terms, during household emergencies such as following natural disasters or when a household member has an acute disease, households that are enrolled in microfinance programs are less likely to manage the situation using extreme strategies, such as selling the crops and food (Meyer, 2010; Townsend, 1995), perhaps underpricing the products in order to raise the required funds quickly. Moreover, microfinance programs usually incorporate and deliver educational activities. Improvements in local people’s knowledge and perceptions of their landscape, through educational activities, may also provide resilience against the effects of natural disasters (Beilin & Reid, 2015).

An assessment of an older savings and lending program funded by the United States Agency for International Development (USAID) in Haiti showed that the program’s members used their savings and loans to pay for fundamental needs, including food (Parker, Francois, Desinor, Cela, & Fleischman Foreit, 2017). Generally, women are more likely than men to spend their income on food and nutrition (Namayengo, Antonides, & Cecchi, 2018). Additionally, results from a systematic review on micro-credit and micro-saving programs in sub-Saharan Africa suggest that these interventions might improve food security (Stewart, van Rooyen, Dickson, Majoro, & de Wet, 2010). Hence, given the central role of women regarding food acquisition and distribution in households, we hypothesize that the participation of Haitian women in a microfinance program may provide resilience against the post-hurricane long-term food insecurity.

The objectives of this study were 1) to assess the association between the impact of Hurricane Matthew and long-term food insecurity and 2) to understand whether this association varies by women’s membership in a microfinance program.

2. Methods

2.1. Study setting and study population

We conducted this study in Haiti, a low-income country, where approximately 60% of the population of nearly 11 million lives below the poverty line (Central Intelligence Agency, 2016). The largest microfinance organization in Haiti, Fonkoze, has been serving Haitians, mainly women, for more than two decades (Tucker & Tellis, 2005). With 44 branch offices across the country, the organization provides services to a wide geographic area. One of the most commonly used microfinance programs of Fonkoze is the solidarity lending program. In this program, groups of 5 clients are organized together to obtain and repay loans. The solidarity lending program’s loans range between US$100 and US$1300 and clients who have been in the program for a longer time may benefit from larger loan sizes as well as other opportunities.

From December 2017 to February 2018, we interviewed female clients who were members of the Fonkoze branch office in Okay, a small port city with a population of around 70,000 inhabitants in Haiti’s southern peninsula. This branch office was established in 1996 and currently serves over 6,200 Fonkoze clients. The sampling frame for the current study comes from the 2017 client database of the Okay Fonkoze branch office. We used random sampling to identify our potential participants. The inclusion criteria were 1) female, 2) ages of 18–49 years, and 3) current client of Fonkoze’s Okay branch office.

Fieldworkers visited the potential participants’ households up to three times. If they were at home, fieldworkers provided them with the details of our study and disclosed the voluntary nature of participation. Participants provided written informed consent. Indiana University’s Human Subjects Office approved the study (Protocol #1705661852). Further details about Fonkoze microfinance program, study population, and data collection in this study have been published previously (M. Rosenberg, et al., 2019)

2.2. Data collection

Local, trained fieldworkers collected the data in face-to-face, tablet-based interviews, in the local language of Haitian Creole. The survey was designed, and data were collected using the REDCap electronic data capture tools hosted at Indiana University (Harris, et al., 2019; Harris, et al., 2009). To ascertain proper translation of the survey questions, they were translated into Haitian Creole and then verified with back-translation. A variety of questions were included in the survey, including questions about sociodemographic characteristics, food security, and microfinance experience. One of the survey sections was specifically targeted to elicit information about Hurricane Matthew and its impact on the participants and their households (Appendix 1).

2.3. Key variables

2.3.1. Independent variable

The independent variable of interest (exposure) was the impact of Hurricane Matthew. To measure hurricane impact, we built a binary variable based on self-reports of injury or death in the household or damage to income-generating assets due to the hurricane. Hence, the hurricane impact variable equals 1 if there were any injury or death or damage to income-generating assets in the household (severe impact) and 0 otherwise (less severe impact).

To validate our hurricane impact exposure, we evaluated the association of this variable with two objective measure of hurricane impact: (1) the hurricane distance from households and (2) the windspeed categories that reached the household. We were expecting that there would be a positive association between these variables. We used data from the National Hurricane Center (National Hurricane Center and Central Pacific Hurricane Center, 2016) and the geocoordinates of participant households to estimate household distance from the hurricane path and the approximate windspeed of hurricane at the location of their residence. Hurricane distance from the household was a continuous variable and hurricane windspeed was a binomial variable [slow (70–80 mph) vs. fast (90–100 mph)].

2.3.2. Dependent variable

The dependent variable (outcome variable) in this study was long-term food insecurity. To evaluate food insecurity, we used two different measures. Household Hunger Scale (HHS) is a cross-cultural validated indicator for household food insecurity (Ballard, Coates, Swindale, & Deitchler, 2011; Deitchler, Ballard, Swindale, & Coates, 2010). We used the HHS variable to categorize participant households into those who experienced “Moderate to Severe hunger” vs. “Little to no hunger” in the last 30 days. We used the Women Dietary Diversity Score (WDDS) as a second measure of food insecurity, with focus on the micronutrient intake of the individual participant (Kennedy, Ballard, & Dop, 2011; Ruel, Deitchler, & Arimond, 2010). The WDDS asks about the consumption of the following nine different food groups; starchy staples, dark green leafy vegetables, other vitamin A rich fruits and vegetables, meat and fish, other fruits and vegetables, organ meat, milk and milk products, legumes, nuts and seeds, and eggs (Kennedy, et al., 2011). In the guidelines for measuring WDDS, it is recommended to categorize the WDDS and use its mean score as the cut-off point for analytic purposes (Kennedy, et al., 2011). Hence, after calculating the WDDS score, according to the guidelines, we dichotomized the scale variable with mean as the cut-off point. The WDDS was measured with reference to a 24-hour time period. Both WDDS and HHS were queried just over one year after exposure to Hurricane Matthew.

2.3.3. Effect modifier

We examined microfinance as a potential effect modifier of the relationship between hurricane exposure and food insecurity outcome. We measured microfinance in two different ways. First, we made a binomial microfinance variable based on the date of first microfinance loan receipt (first loan before hurricane vs. first loan after hurricane). Using this approach all participants (n=301) were included in the analysis. However, hurricane exposure might influence individuals’ subsequent decision-making regarding enrollment to a microfinance program. Those who were severely impacted by a natural disaster might be less interested in or able to enroll to a microfinance program. For this reason, we made an alternative microfinance variable and restricted our sample size to those who had received their first loan before the hurricane (n=142).

This second variable was also binomial and was based on the duration between the first loan receipt date (before hurricane) and the hurricane date in months (Loan duration >12 months before hurricane vs. Loan duration ≤12 months before hurricane). The assumption here is that those with a longer duration of microfinance exposure would have had sufficient time to see the positive effects from participation by the time of the hurricane, but those with a shorter duration would not (M. S. Rosenberg, et al., 2011).

2.3.4. Confounding covariates

We examined potential confounding effects of several sociodemographic and economic variables, including; age (continuous); marital status (currently married, divorced or separated, and never married); education (none, preschool or primary, and secondary and higher); literacy (yes vs. no); household size (a continuous variable defined as the number of people in a household); household assets (determined as the sum of the self-reported value of 20 potential items in the participants’ households, such as oven, television, radio, and refrigerator); and interviewee’s last month income. The asset index and interviewee’s last month income variables were categorized in quartiles, from lowest (Q1) to highest (Q4). We further created a location covariate based on arrondissement of residence. Haiti has 10 departments and 42 arrondissements. Participants in this study came from 5 arrondissements. Particularly, participants from Les Cayes arrondissement (an urban area) might have more wealth and better built houses compared to those from other arrondissements.

2.4. Data analysis

We used t-test and Fisher’s exact test to evaluate the association between the hurricane impact exposure (the independent variable) with the hurricane distance from households and the windspeed categories that reached the households, respectively. We used log-binomial regression to obtain prevalence ratios (PR) for the associations between hurricane impact and food insecurity outcomes (McNutt, Wu, Xue, & Hafner, 2003). An a priori p-value of 0.05 was determined to be statistically significant. If log-binomial models failed to converge, we ran Poisson regression models with a robust error variance; these models also produce PRs and are less sensitive to convergence issues (McNutt, et al., 2003; Zou, 2004). To evaluate effect modification by microfinance exposure, we included an interaction term between microfinance and hurricane impact to the regression models. An interaction term with a p-value of less than 0.2 was considered significant (Selvin, 2004). In our multivariate (sensitivity) analysis, we adjusted for covariates that were significantly different between the two hurricane exposure groups (Eq. 1). We used SAS 9.4 for data analysis (Cary, NC) (Hoddinott & Yohannes, 2002). Moreover, to assess how robust the exposure-outcome association is to unmeasured confounding, we calculated an e-value (VanderWeele & Ding, 2017) using “EValue” package in RStudio (Mathur, Ding, Riddell, & VanderWeele, 2018; RStudio Team, 2018).

$$\mathit{\text{log}}\left(\mathit{\text{Food}}\mathit{\text{Ins}}\mathit{\text{Prev}}\right)=\alpha +{\beta }_{1}Hurr+{\beta }_2{x}_2\dots +{\beta }_6{x}_6+{\beta }_{7}MF+{\beta }_{8}Hurr\ast MF$$ Equation 1:

Where the prevalence of food insecurity (‘Food Ins Prev’) is predicted by the impact of Hurricane Matthew (‘Hurr’), our covariate set (β₂x₂ to β₆x₆), and the interaction between microfinance experience (‘MF’) and hurricane impact.

3. Results

Overall, 349 households were selected as potential participants, of which one refused to participate, 41 were excluded since they did not match the inclusion criteria, and three could not be located. Moreover, in the current study, data on independent/dependent variables were not available for three participants and they were removed from the analysis. In total, 301 observations were included in the analyses of this study.

Ages ranged from 20 to 49 with a median of 36 years (Table 1). More than half (55%) of the participants were married at the time of the study. Educational attainment was low (17% had no education and 39% had only a preschool/primary school level of education). Most participants were from Les Cayes arrondissement (82%). Compared to less severely impacted group, participants reporting severe hurricane impact were more likely to report no literacy, lower household asset but higher last month income (at individual level), living in Aquin or other arrondissements, and a larger household size. Most women reported a low WDDS (60%). Similarly, most households suffered from moderate to severe hunger, 73%.

Table 1.

Sociodemographic characteristics of the study population (301 female Haitian microfinance clients)

	Hurricane Impacta
	Total N=301	Severe impact N=252	Less severe impact N=49	p
Socio-demographic characteristics
	N (%)	N (%)	N (%)
Age				0.3276
20–29	69 (22.9)	54 (21.4)	15 (30.6)
30–39	126 (41.9)	109 (43.3)	17 (34.7)
40–49	106 (35.3)	89 (35.3)	17 (34.7)
Marital status				0.6855
Currently married	163 (55.3)	138 (55.9)	25 (52.1)
Divorced/separated	24 (8.1)	21 (8.5)	3 (6.3)
Never married	108 (36.6)	88 (35.6)	20 (41.7)
Missing	6	5	1
Education				0.1139
None	50 (16.8)	46 (18.5)	4 (8.2)
Preschool/Primary	117 (39.3)	99 (39.8)	18 (36.7)
HS or more	131 (44.0)	104 (41.8)	27 (55.1)
Missing	3	3	0
Literacy				0.0212
Yes	221 (74.2)	179 (71.6)	42 (87.5)
No	77 (25.8)	71 (28.4)	6 (12.5)
Missing	3	2	1
Household asset quartilesb,c				<.0001
Q1	76 (25.2)	73 (29.0)	3 (6.1)
Q2	74 (24.6)	64 (25.4)	10 (20.4)
Q3	76 (25.2)	62 (24.6)	14 (28.6)
Q4	75 (24.9)	53 (21.0)	22 (44.9)
Interviewee’s last month income quartilesc				0.0362
Q1	125 (41.5)	98 (38.9)	27 (55.1)
Q2	45 (15.0)	36 (14.3)	9 (18.4)
Q3	86 (28.6)	80 (31.7)	6 (12.2)
Q4	45 (15.0)	38 (15.1)	7 (14.3)
Location (arrondissements)
Les Cayes	237 (82.0)	196 (81.0)	41 (87.2)	0.0233
Aquin	33 (11.4)	28 (11.6)	5 (10.6)
Other (Port-Salut, Baradères, and Anse-à-Veau)	19 (6.6)	18 (7.4)	1 (2.1)
Missing	12	10	2
	Mean (SD)	Mean (SD)	Mean (SD)
Household size	4.95±1.87	5.08±1.85	4.27±1.84	0.007
Food Insecurity Outcomes
	N (%)	N (%)	N (%)
Women Dietary Diversity				<.0001
High	120 (39.9)	85 (33.7)	35 (71.4)
Low	181 (60.1)	167 (66.3)	14 (28.6)
Household Hunger				0.0002
Little to no hunger in the household	82 (27.5)	58 (23.3)	24 (49.0)
Moderate to severe hunger in the household	216 (72.5)	191 (76.7)	25 (51.0)
Missing	3

^a.Main exposure: Household injury, death, and/or damage to income-generating assets due to hurricane Matthew (severe impact) vs. otherwise (less severe impact).

^b.To create the Household asset quartile variable, we created a score based on the participants’ responses to questions about the value of 20 potential items in the participants’ households, such as oven and television. We categorized the variable into four quartiles.

^c.Q1 has the lowest values and Q4 has the highest values. Quartiles do not have the same amount of data because of tied values.

Boldface indicates significant finding, p-value<0.05

More than 84% of participants were severely impacted by the hurricane. The hurricane exposure variable was significantly and positively associated with the distance from hurricane (p= 0.0082) and hurricane windspeed (p= 0.0228) variables. Households that were closer to the hurricane and that were in the fast (90–100 mph) windspeed zones were more likely to report injury or death or damage to income-generating assets.

Severe hurricane impact was significantly associated with low WDDS and moderate to severe HHS (Table 2). Women with severe hurricane impact were more than two times as likely to report low WDDS compared to women who were less severely impacted [PR (95% CI): 2.32 (1.48, 3.64)]. Likewise, compared to households that were less severely impacted, severely hurricane impacted households were 50% more likely to have a moderate to severe HHS [PR (95% CI): 1.50 (1.13, 1.99)]. In general, similar results were obtained when we adjusted for the potential confounders including, literacy, household size, household assets quartiles, interviewee’s last month income quartiles, and location (arrondissements) in the sensitivity analysis.

Table 2.

Relationship between hurricane impact and food insecurity outcomes.

Exposures	Outcomesa
	Low WDDS		Moderate to severe HHS
	PR (95% CI)	aPR (95% CI)b	PR (95% CI)	aPR (95% CI)b
	n= 301	n= 285	n=298	n=282
Hurricane impact (severe impact vs. less severe impact)	2.32 (1.48, 3.64)	1.95 (1.21, 3.12)	1.50 (1.13, 1.99)	1.34 (0.99, 1.84)
Estimated E-value (lower limit of the CI)	4.07 (2.32)	3.31 (1.71)	2.37 (1.51)	E-value is 1c

^a.We used log-binomial regression for statistical analysis. In cases where the log-binomial model did not converge, we used Poisson regression models with a robust error variance.

^b.Models adjusted for literacy (Yes vs. No), household size (continuous), household assets quartiles, interviewee’s last month income quartiles, and location (arrondissements).

^c.Confidence interval crosses the null value, so E-value is 1.

- WDDS: Women Dietary Diversity Score (Low vs. High); HHS: Household Hunger Scale (Moderate to severe vs. Little to no hunger); aPR: adjuster prevalence ratio

- Boldface indicates significant finding (Applicable only to first row), p-value<0.05

The E-value for the association between severe hurricane impact and low WDDS outcome was 4.07, meaning that the observed PR of 2.32 could be explained away by an unmeasured confounder that was associated with both the hurricane exposure and the low WDDS outcome by a PR of 4.07 (Table 2) (VanderWeele & Ding, 2017). It is unlikely to find a confounding factor that is associated with both hurricane impact and food insecurity with such a strong magnitude of association. The E-value for the association between severe hurricane impact and moderate to severe HHS outcome was 2.37. Hence, a potentially unmeasured confounder would need to have a PR of 2.37 with both hurricane exposure and HHS outcome to turn the reported PR to null. It might be possible to find such a confounder. In fact, when we adjusted for the potential abovementioned confounders the association turned insignificant. However, the magnitude of the point estimate remained notable [PR (95% CI): 1.34 (0.99, 1.84)].

For the effect of hurricane on low WDDS outcome, we found effect modification by the date of first loan (first loan before hurricane vs. after hurricane) (Table 3). Among participants who received their first loan after the hurricane, those who were in the severe impact were 3 times as likely to have a low WDDS [PR (95% CI): 2.99 (1.53, 5.85)]. This significant association was weaker and not statistically significant among those who had received their first loan before the hurricane (p-value for interaction was 0.1534). A similar significance trend was observed for the HHS outcome, though the p-value for interaction was not significant. Lastly, we restricted the effect modification analyses to those who received their first loan before hurricane (n=142). Here, we used loan duration as the effect modifier (Loan duration >12 months before the hurricane vs. ≤12 months). The p-values for interaction terms were not significant in these analyses. However, the PRs for the effect of hurricane on food insecurity outcomes in this restricted subsample were generally smaller than that in the whole sample. Similar results were found when we adjusted for the potential confounders in sensitivity analysis (Appendix 2)

Table 3.

Prevalence ratios for the associations between hurricane impact and food insecurity outcomes, stratified by different microfinance variables

	Low WDDSa PR (95% CI)		p for interaction term	Moderate to severe HHSa PR (95% CI)		p for interaction term
Stratified by loan duration, among the total study population (n=301)
	First loan before hurricane n=142	First loan after hurricane n=159		First loan before hurricane n=140b	First loan after hurricane n=158
Hurricane impact (severe impact vs. less severe impact)	1.58 (0.89, 2.78)	2.99 (1.53, 5.85)	0.1534	1.73 (0.99, 3.05)	1.42 (1.03, 1.97)	0.5531
Stratified by loan duration and restricted to those who received their first loan before the hurricane (n=142)
	Loan duration >12 months before hurricane n=83b	Loan duration ≤12 months before hurricane n=59		Loan duration >12 months before hurricane n=82	Loan duration ≤12 months before hurricane n=58
Hurricane impact (severe impact vs. less severe impact)	1.55 (0.78, 3.10)	1.49 (0.55, 4.02)	0.9457	1.75 (0.88, 3.47)	1.59 (0.59, 4.28)	0.8790

^a.We used log-binomial regression for statistical analysis.

^b.The numbers might not sum up to 301 or 144 because of missing values

WDDS: Women Dietary Diversity Score (Low vs. High); HHS: Household Hunger Scale (Moderate to severe vs. Little to no hunger)

Boldface indicates significant finding; p<0.05 and p for interaction term<0.2

4. Discussion

In this study, we found that the impact of a recent hurricane was associated with long-term food insecurity in Haiti. Women who reported severe impact from the hurricane were more likely to report low WDDS and experience moderate to severe hunger compared to those who were less severely impacted. We further found that microfinance program might help participants to improve their long-term food security after a hurricane. Specifically, the magnitude of food insecurity one year following hurricane exposure was weakened among those who had received their first microfinance loan before the hurricane.

Additionally, we found a high magnitude of food insecurity in our sample. Descriptive studies have reported varying long-term food insecurity prevalence after different natural disasters. Similar to our findings, in a longitudinal survey following Hurricane Katrina, the prevalence of food insecurity remained high after two years, but started to improve in the third year, though it was still higher than the average state prevalence (Clay, et al., 2018). However, another study in Nepal reported a much faster food security restoring period following an earthquake in 2015 (Thorne-Lyman, et al., 2018). One year after the earthquake, the food insecurity prevalence in this study was close to that from one year before the earthquake (Thorne-Lyman, et al., 2018). This heterogeneity of long-term post-disaster food insecurity prevalence, and recovery time, might be due to the different types of natural disaster, populations, study settings, and infrastructures. It may also be that the impact is different based on the poverty levels pre-disaster (disaster has greater effect on food insecurity among poorer people than among better off people). Last but not least, the measurement of food insecurity differed in the aforementioned studies.

We found that hurricane impact was associated with increased food insecurity even one year after the disaster. Few research studies have been conducted previously to measure the magnitude of the association between natural disaster exposure and long-term food insecurity. A simulation study in two governorates of Yemen showed that a 2008 tropical storm and subsequent flooding had a long-term negative impact on food security of these regions (Breisinger, Ecker, Thiele, & Wiebelt, 2016). A cross-sectional study in Eastern Uganda also found that households affected by a natural disaster, landslide, had higher food insecurity and lower dietary diversity compared to those not affected (Rukundo, et al., 2016). Moreover, a descriptive and ethnographic study in Ethiopia found that in destitute settings, where there is no adequate systematic support or social safety nets, indigenous mutual support practices can help to build resilience capacities and to lessen the negative impacts of livelihood shocks (Endris, Kibwika, Hassan, & Obaa, 2017). Lastly, in Africa, youth savings groups have been promoted to instill saving habits in youth, provide access to financial resources and services to them, and help them to start income-generating activities (Flynn & Sumberg, 2018). These saving groups might help young people to cope in uncertain economic situations (Flynn & Sumberg, 2018).

Multiple studies in Kenya (Weiser, et al., 2017), rural Malawi (Weinhardt, et al., 2017), and other low to middle income countries (Boccia, et al., 2011) have shown that microfinance programs can improve food security. We found that microfinance favorably modified the association between hurricane exposure and food insecurity outcome by weakening their association. Two similar studies on drought established that a set of interventions including microfinance could reduce hunger following exposure to drought (Coppock, Desta, Tezera, & Gebru, 2011; Doocy, Teferra, Norell, & Burnham, 2005). The current study was one of the first studies to quantitively evaluate the effect modification of microfinance, with evidence supporting the hypothesis that microfinance can provide resilience against post-hurricane long-term food insecurity. Through the mechanism of providing capital to clients, microfinance may help to reduce vulnerability, prepare for future risk, and cope with post-disaster economic losses and prolonged insecurity. (Sebstad & Cohen, 2000)

4.1. Limitations

This study has some limitations. Because of our cross-sectional study design, and unavailability of data prior to the implementation of Fonkoze program, the temporal relationship between hurricane impact and long-term food insecurity outcome is not clear. In fact, some participants might had been suffering from food insecurity even before the hurricane and they may have been the ones most likely to be severely impacted by the hurricane. Evaluating the associations between microfinance membership, hurricane exposure, and long-term food insecurity can be challenging in the context of Haiti due to chronic food insecurity along with sociopolitical instabilities. Because of these or some other unmeasured factors, households that were severely impacted by hurricane might differ from those that were less severely impacted by the hurricane, although natural disasters tend to be at least somewhat random in their exposure distribution. Providing support to this, our self-reported exposure variable was positively and strongly correlated with hurricane distance from the household, a fairly exogenous exposure. Moreover, our consistent results from the sensitivity analysis along with the large E-value suggest that the magnitude of unmeasured and uncontrolled confounding needed to explain our findings is improbably large. Recall bias might also be a source of measurement error in our study, specifically for the food insecurity outcome. However, the recall period of the validated questionnaires are short, 30 days for HHS and 24 hours for the WDDS (Ballard, et al., 2011; Kennedy, et al., 2011). Lastly, the data were collected in an interviewer administered quantitative survey. Hence, different types of interviewer bias, such as interviewer gender bias, are also potential sources of error in this study (Kianersi, Luetke, Jules, & Rosenberg, 2019).

4.2. Summary

Natural disasters, like hurricanes, might have a positive association with long-term food insecurity at individual and household levels. Microfinance programs might help to improve long-term food security, particularly through improvements in dietary diversity. Moreover, in countries like Haiti, the frequency of natural disasters is high, which necessitates the development of reliable interventions to prevent long-term food insecurity. We suggest future studies further evaluate the effect modification of microfinance programs on this relationship and the possible pathways through which microfinance programs could augment post-disaster food security with longitudinal study designs, and complemented with qualitative and ethnographic data collection.

• Natural disasters, like hurricanes, may cause long-term food insecurity

• Severe hurricane impact was associated with moderate to severe household hunger

• Severe hurricane impact was also associated with low dietary diversity

• Both associations were attenuated among pre-hurricane microfinance clients

• Microfinance may help to reduce post-disaster prolonged food insecurity

Appendix 1.

Survey instruments (Questions about the independent (exposure) and dependent (outcome) variables)

Question	Answer options (separated by comma)
Food Security and Dietary Diversity instrument
1. In the past month was there ever no food to eat of any kind in your house because of lack of resources to get food?	Yes, No, Don’t know, Refuse
2. How often did this happen in the past month?	RARELY (1–2 TIMES), SOMETIMES (3–10 TIMES), OFTEN (MORE THAN 10), Don’t know, Refuse
3. In the past month did you or any household member go to sleep at night hungry because there was not enough food?	Yes, No, Don’t know, Refuse
4. How often did this happen in the past month?	RARELY (1–2 TIMES), SOMETIMES (3–10 TIMES), OFTEN (MORE THAN 10), Don’t know, Refuse
5. In the past month did you or any household member go a whole day and night without eating anything at all because there was not enough food?	Yes, No, Don’t know, Refuse
6. How often did this happen in the past month?	RARELY (1–2 TIMES), SOMETIMES (3–10 TIMES), OFTEN (MORE THAN 10), Don’t know, Refuse
(For questions 7 to 22) Yesterday during the day or night did you drink/eat any
7. Bread, biscuits, pastries, buns, pasta, noodles, crackers, breadfruit or other foods made from grains such as corn, wheat, millet, rice?	Yes, No, Don’t know, Refuse
8. Pumpkin, carrots, sweet potatoes or other tubers and vegetables that are yellow or orange inside?	Yes, No, Don’t know, Refuse
9. White sweet potatoes, white yams, manioc, cassava, plantains or any other foods made from roots?	Yes, No, Don’t know, Refuse
10. Any dark green leafy vegetables such as spinach, lettuce, other dark green leafy vegetables or okra?	Yes, No, Don’t know, Refuse
11. Ripe mangoes, ripe papaya, apricots, cantaloupe melons or other fruits that are yellow or orange inside?	Yes, No, Don’t know, Refuse
12. Other fruits or vegetables, like bananas, pomegranates, tomatoes, green beans, avocado, etc.?	Yes, No, Don’t know, Refuse
13. Liver, kidney, heart, or other organ meats?	Yes, No, Don’t know, Refuse
14. Any meat such as beef, pork, lamb, goat, chicken, duck or other meat?	Yes, No, Don’t know, Refuse
15. Eggs	Yes, No, Don’t know, Refuse
16. Fresh or dried fish, shellfish, or seafood?	Yes, No, Don’t know, Refuse
17. Any foods made from beans, peas, pistachios, walnuts, mamba, or other seeds?	Yes, No, Don’t know, Refuse
18. Cheese, yogurt/curd milk, or other milk products?	Yes, No, Don’t know, Refuse
19. Any other oils, fats, or butter, or foods made with any of those products?	Yes, No, Don’t know, Refuse
20. Any sugary foods such as chocolates, sweets, candies, pastries, cakes, or biscuits?	Yes, No, Don’t know, Refuse
21. Condiments for flavor such as chilies, spices, parsley?	Yes, No, Don’t know, Refuse
22. Foods made with red palm oil, red palm nut, or red palm nut pulp sauce?	Yes, No, Don’t know, Refuse
Natural Disaster Experience instrument
1. Did you live in this zone during Hurricane Matthew?	Yes, No, Don’t know, Refuse
2. Did you live in this same house during Hurricane Matthew?	Yes, No, Don’t know, Refuse
3. Was anyone in your household injured during Hurricane Matthew?	Yes, No, Don’t know, Refuse
4. Was anyone in your household killed during Hurricane Matthew?	Yes, No, Don’t know, Refuse
5. Was your house damaged in Hurricane Matthew?	Yes, No, Don’t know, Refuse
6. Were you able to repair your house after Hurricane Matthew?	Yes, No, Don’t know, Refuse
7. Were any of your belongings damaged in Hurricane Matthew?	Yes, No, Don’t know, Refuse
8. Did Hurricane Matthew damage or destroy any of your belongings that you rely on to make money?	Yes, No, Don’t know, Refuse
9. During Hurricane Matthew, did you take any precautions to protect important documents or valuable household items?	Yes, No, Don’t know, Refuse
10. If yes, were you successful in protecting your documents and valuables?	All documents and valuables were protected, Some documents and valuables were protected but others destroyed or damaged, All documents and valuables were destroyed or damaged, Don’t know, Refuse

Appendix 2.

Sensitivity analysis: Prevalence ratios for the associations between hurricane impact and food insecurity outcomes, stratified by different microfinance variables and adjusted for potential confounders

	Low WDDSa aPR (95% CI)		p for interaction term	Moderate to severe HHSa aPR (95% CI)		p for interaction term
Stratified by loan duration, among the total study population (n=301)
	First loan before hurricane n=133b	First loan after hurricane n=152		First loan before hurricane n=131	First loan after hurricane n=151
Hurricane impact (severe impact vs. less severe impact)	1.32 (0.74, 2.35)	2.61 (1.27, 5.35)	0.1343	1.42 (0.84, 2.38)	1. 30 (0.88, 1.92)	0.7860
Stratified by loan duration and restricted to those who received their first loan before the hurricane (n=142)
	Loan duration >12 months before hurricane n=74b	Loan duration ≤12 months before hurricane n=59		Loan duration >12 months before hurricane n=73	Loan duration ≤12 months before hurricane n=58
Hurricane impact (severe impact vs. less severe impact)	1.28 (0.62, 2.60)	1.36 (0.53, 3.50)	0.9499	1.49 (0.79, 2.79)	1.30 (0.54, 3.15)	0.6602

^a.We used log-binomial regression for statistical analysis. In cases where the log-binomial model did not converge, we used Poisson regression models with a robust error variance.

^b.The numbers might not sum up to 301 or 144 because of missing values

- All models in this table are adjusted for literacy (Yes vs. No), household size (continuous), and household assets quartiles.

WDDS: Women Dietary Diversity Score (Low vs. High); HHS: Household Hunger Scale (Moderate to severe vs. Little to no hunger);

aPR: adjuster prevalence ratio

Boldface indicates significant finding, p<0.05, p for interaction term<0.2