Might Climate Change the “Healthy Migrant” Effect?

1.0 Introduction

International migration tends to be positively selective on health, meaning that migrants are typically healthier than non-migrants in sending regions. This association exists because, in general, international migration is difficult—it entails uprooting from a known cultural, economic and political context. Migration also results in shifting social relationships and the establishment of new ones – all taxing processes.

Yet climate change is altering international migration by changing the context of livelihoods across the globe. In some places, environmental scarcity is acting as a “push” factor – intensifying migration as a means of diversifying livelihoods or pursuing new ones (e.g., Nawrotzki, Riosmena et al. 2015). In other places, increased availability of environmental resources provides the capital through which households are able to send migrants elsewhere to earn income (e.g., Gray 2010; Hunter et al. 2013). In both cases -- “scarcity” or “surplus” -- environmental factors influence migration and many scholars anticipate climate will increasingly play a role in population movement (McLeman 2013). Of course, other forms of environmental pressures yield migration including natural disasters, sea level rise, and resource conflict. In addition, economic, political and other contextual factors shape the ways in which migration-environment linkages manifest (Black et al. 2011; Hugo 2011).

What remains unknown is the potential for climate change to alter demographic processes such as the phenomenon of the “healthy migrant.” Climate stress may reduce health selectivity if individuals in both good and poor health experience heightened pressure to relocate. On the other hand, migration’s health selectivity may be intensified if migration becomes an even more challenging endeavor, one fit only for those in particularly good health. Of course, alterations to the healthy migrant phenomenon may also vary by context since different locales will face different climate strain.

We investigate this question by bringing together the literatures linking migration-health and migration-environment and by presenting an empirical examination of health selectivity in the context of Mexico-US migration. This migration stream represents one of the longest and most sustained international flows of people in the world, and recent research has found that climate strain is associated with heightened likelihood of households sending a migrant to the US, particularly in the short-term (Nawrotzki and DeWaard 2016). Mexico-US migration also exhibits the “healthy migrant” phenomenon (e.g. Crimmins et al. 2007), making it a logical choice for a study that examines the intersection of climate-migration-health.

2.0 Material and Methods

2.1 Migration and Health Data

We tap into the strength of Mexican Migration Project (MMP) data to explore the climate-migration-health intersection in the context of Mexico-US migration. The MMP is a collaborative effort between Princeton University (USA) and the University of Guadalajara (Mexico) that provides detailed migration information from a sample of Mexican communities. Since 1982, three to five different communities have been surveyed annually, totaling 161 communities across the years 1982–2016. The communities are chosen based on fieldwork by the principal investigators, with decisions informed by several factors including community sex ratio and level of urbanization. Within each community, approximately 200 households are randomly sampled to participate in data collection which includes both qualitative and quantitative demographic, social and economic data. MMP-provided weights are applied such that the samples are representative of the communities surveyed.

The full MMP sample consists of 169,945 individuals, but migration and health information are only collected for the head of each household (MMP). Thus, our sample is restricted to household heads. Additionally, females are excluded from our analysis since there were so few listed as head of the household. Prior research in Mexico reveals that men and women, especially as household heads, tend to make different decisions regarding livelihood strategies (e.g., Buechler 2016; Radel et al. 2016). In this way, there are likely gender differences in the climate-migration-health connections due to the differential likelihood of engaging migration as a household strategy. Yet, the MMP does not offer a sufficiently large sample of female household heads to examine these differentials, and as such, due to concern with confounding head-of-gender influences with our focus on climate and health, our analyses are restricted to male household heads.¹

2.1.a. Migration Data

Within the MMP, migration is defined as a move for work, job search, or to establish a new residence – thereby excluding short-term visits to family or friends. In this study, migration is represented as a binary variable indicating no migration (0) or a household head’s first international migration (1). After the first move, important migration correlates such as social networks, will likely be altered. As such, to best isolate climate and health factors, we constrain our estimation to only first moves. Further, only migration after age 14 is modeled to exclude movement by juvenile household members.

2.1.b. Health Data

Health data collection began in 2007 within the MMP, so our study focuses on households in communities surveyed since that time. All health measures are self-reported and include responses for hypertension, diabetes, heart attack and heart problems, lung conditions, cancer, stroke, and psychological disorders. In addition, respondents were asked to self-assess health at age 14, while also being asked about their adult height. These latter two measures are used in this study since they best represent early life health, therefore useful in the examination of the overall health selectivity of migration in this population. The other health measures ask if any of the conditions have ever been experienced, thereby not allowing for characterization of health prior to migration event.²

The MMP asks respondents their perceived health at age 14 on a scale of poor, regular, good, or excellent. The second category “regular” is a translation from Spanish; however, in Spanish the word “regular” has a slight negative connotation. In line with the coding and interpretation in other studies (e.g., Ullman, Goldman & Massey 2011), we attribute a response of “regular” to reflect “fair” health. In addition, since only 0.20% of individuals in the sample reported being in poor health at age 14, we cluster “poor” and “fair” health responses into one category that, combined, represent 2.91% of respondents. Table 1 offers weighted descriptive statistics including health differences between migrants and non-migrants.

Table 1.

Weighted descriptive statistics by U.S. migration experience

	Sample	Migrant	Non-Migrant	p-Value
Health & Climate Characterististics
Height(cm)	166.62	167.70	166.34	<0.001
Self-rated health, age 14
Poor-Regular	2.91%	2.36%	3.56%
Good	65.34%	53.62%	68.43%	<0.001
Excellent	31.75%	44.01%	28.51%	<0.001
30 year precip. norm(mm)	970.55	861.92	999.20	<0.001
Demographic Characteristics
Age	47.97	44.76	48.82	<0.001
Married	82.66%	83.58%	82.42%
Socioeconomic Characteristics
Mean years of education	7.51	7.16	7.60	<0.001
Mean assets score (0–14)	9.45	9.74	9.38	<0.001
Contextual Characteristics
Rural	54.51%	66.10%	51.45%	<0.001
Irrigation in municipality	88.93%	88.02%	89.17%

Mexican Migration Project, MMP161

N=6,435

Adult height represents our second health measure, and although the measure has limitations, two bodies of research boost our confidence in its utility as an overall health measure. First, substantial research reveals that early life conditions exert a hard-to-reverse effect on adult height (Alacevich and Tarozzi 2016: 65). Indeed, based on a review of research over the past 25 years, Perkins and colleagues (2016:149) conclude, “evidence across studies indicates that short adult height (reflecting growth retardation) in low- and middle-income countries is driven by environmental conditions, especially net nutrition during early years.” In addition to nutrition, broader socio-economic mechanisms also link childhood health context to adult height. These include parental social class, including maternal education, in that these factors shape access to resources, risk exposure (including disease), and parental health behaviors (Perkins et al. 2106).

Still, a logical concern is that while adult height may be useful as a collective representation of population health, a different lens is necessary for individual-level analyses. At this scale, the influence of genetics may complicate the use of height as an indicator of overall health. Research has suggested that nutrition and other environmental factors may be particularly impactful on adult height during the first two years of development, with genetics exerting a stronger influence from age 2 through puberty (Silventoinen et al. 2008).

Although height is not fully determined by early life conditions, several studies do identify a correlation. Case and Paxon (2008:4), for example, make use of a sample of over 70,000 individuals over age 50 in the Health and Retirement Study and report, “results are consistent with height providing a marker of a healthier and financially more comfortable early life environment.” HRS respondents who report childhood health as excellent or very good were, on average, 0.25 inches taller than other cohort members. Earlier studies in Finland and the UK find similar connections between adult height and childhood living conditions (Kuh and Wadsworth 1989; Silventoninen, Lahelma and Rahkonen 1999). And as a final example, evidence for this correlation at the individual-scale has been provided through comparison of height among adults and children of Indian ethnicity in England relative to those in India, finding that Indian children in England are 6–8% taller relative to those in India, even controlling for age, gender, and parental height (Alacevich and Tarozzi 2016).

With this foundational knowledge of early life conditions linking to adult height, we reach to a second body of literature that identifies a connection between early life conditions and adult health – a paradigm termed the “developmental origins of health and disease” (e.g. Gluckman et al. 2008; Heindel and Vandenberg 2015). The specific etiologies of these connections remain under examination, but adult cancer, lung disease, cardiovascular illness and arthritis/rheumatism have all been linked to aspects of the environment in early life. Negative conditions could include poor nutrition in early life, exposure to disease or toxins, as well as more general socioeconomic deprivation. As one research example, using the U.S. Health and Retirement Survey, Blackwell and colleagues (2001) found a four-fold increase in lung conditions in middle age for those that had experienced a major bout of infectious disease in childhood. Those who had childhood experience of a non-infectious disease reported a three-fold increase in cancer, plus a doubling in reports of arthritis/rheumatism.

In utero context also influences later life health – small size at birth, for example, has been linked to later life high blood pressure (Law et al. 2005). Based on a cohort study in the U.K., Syddall et al. (2005) find that, by age 75, a one-standard-deviation increase in birth weight reduced mortality risk by 0.85 percent. The strongest connections have been identified with risk of death due to circulatory problems including cardiovascular disease (Syddall et al. 2005).

Although there are documented links between early life conditions and adult height, as well as early life conditions and adult health, a certain degree of caution is necessary in the use of height as a general proxy for health; adult height is certainly not perfectly correlated with adult health. Limitations include the fact that height has genetic components (Case and Paxson 2010), and also individuals might exaggerate their height in self-reports leading to systematic differences by sex, age, and socioeconomic status (Osuna-Ramirez et al. 2006). We address these limitations in a few ways. To account for broader temporal variation in height, we use a mean-centered year variable. To further account for potential regional differences in height, we stratify all models according to ecological zones discussed below in section 2.4.

As an example of height’s usefulness in migration-health scholarship, recent research does find evidence of positive health selectivity making use of height across migrants and non-migrants in populations as varied as China, India, Mexico, and the Philippines (Riosmena, Kuhn and Jochem 2017). In fact, Mexican migrants report a 3 cm positive height advantage in one recent study (Riosmena, Wong and Palloni 2013), while another finds that Mexican migrant men are 0.44 standard deviations taller than non-migrants (Riosmena, Kuhn and Jochem 2017). This finding is in line with our sample, as in the MMP, migrants have an average height of 167.70 cm, compared to 166.34 cm for non-migrants (Table 1).

2.2 Climate Data

Monthly municipal-level rainfall data, 1961–2010, provide the foundation for indicators of climate strain in this study. The rainfall data, originally raster files compiled by the University of East Anglia’s Climate Research Unit, have been pre-processed into a municipal level dataset by the Terra Populus division at the Minnesota Population Center (Harris et al. 2014; Kugler et al. 2015; MPC 2013). Originally, these data represented monthly rainfall (in millimeters) by municipality. However, for our purposes, annual measures were created to evaluate longer term trends.

2.3 Control Variables

To best isolate the association between climate-migration-health, other known correlates of migration must be included as controls. Key are age and education as a large body of research documents patterns of migration as related to age and education generally and with regard to Mexico-US migration. As with migration trends more broadly, younger individuals are more likely to move although this negative age selectivity has declined following the economic recession in the US beginning in early 2008. Even so, middle-aged men (aged 26–45) remain significantly less likely to migrate than men aged 15–24 (Villareal 2014).

The pattern with regard to education is somewhat distinct for Mexico-US streams as compared to broader understanding regarding migration-education linkages. In most contexts, individuals with higher levels of education tend to have higher migration probabilities, in part due to engagement with broader labor markets. Yet Mexico-US migrants tend to have relatively lower levels of education, although this negative selection on education has declined since the recession (Villareal 2014). Still, the past influence of this long-standing negative selectivity is evidenced by contemporary patterns. In 2014, only 6 percent of Mexican immigrants aged 25 and over had a bachelor’s degree or higher, contrasted with 29 percent of foreign-born overall (Zong and Batalova 2016).

Marital status has also been shown, in many settings, to be associated with migration probabilities. From Mexico, pioneer migrants -- those forging new migration streams from areas with lower levels of historic migration -- are less likely to be married than those that follow (Lindstrom and Ramirez 2010). As such, we also include marital status as a control. In addition, to account for household socio-economic status, we incorporate an asset score based on household amenities such as running water, sewage, electricity and internet access.

And finally, we integrate community and municipal-level characteristics of relevance to understanding migration patterns. These include a binary indicator of access to irrigation in the municipality as well as urban/rural status of the community. Communities listed as “rancho” or “town” are coded as rural, while “smaller urban area” or “metropolitan area” are coded as urban. We also indicate the level of overall migration within the climate categories in order to reflect migration prevalence and possible intensity of social networks. Taken together, these controls reflect factors that shape local opportunity and/or economic and social well-being.

2.4 Analytical Strategy

We first present bivariate associations between migrants and non-migrants with regard to health status (Table 1) and as related to environmental stress (Table 2).³ Following, within a multivariate framework, we use survival analyses to examine migration probabilities as related to both health and recent environmental conditions while controlling for the factors outlined above that are also linked with migration.

Table 2.

Weighted descriptive statistics of migrants by municipal climate classification

	Dry	Mod-Dry	Mod-Wet	Wet	p-Value
Health & Climate Characterististics
Height(cm)	167.39	169.72	167.95	164.51	<0.1
Self-rated health, age 14
Poor-Regular	2.97%	2.62%	2.96%	0.83%
Good	59.26%	39.70%	68.21%	58.53%	<0.001
Excellent	37.78%	57.69%	28.83%	40.63%	<0.001
30 year precip. norm(mm)	613.58	775.48	927.02	1205.86	<0.001
Demographic Characteristics
Age	43.85	48.08	42.97	41.75	<0.01
Married	82.05%	85.40%	86.52%	79.80%	<0.01
Agricultural Occupation	38.46%	30.94%	35.93%	29.82%	<0.01
Socioeconomic Characteristics
Mean years of education	6.53	6.63	8.58	7.51	<0.001
Mean assets score (0–14)	8.71	10.48	9.96	9.43
Contextual Characteristics
Rurala	71.01%	71.32%	67.27%	51.59%	<0.05
Irrigation in municipality	96.61%	87.82%	95.44%	73.38%	<0.001
Migration experienceb	28.60%	33.13%	11.88%	16.54%	<0.001

Mexican Migration Project, MMP161

N = 6,435

Notes:

^aRural is measured at the community level

^bMigration experience represents the % of HH-heads in the climate category that migrated to the US in the study window

Survival analyses estimate probabilities of an event occurring – in our case, migration – across time. Time is represented in survival data by multiple observations for each unit of analysis (household head), each observation representing a period during which they were at “risk” of the central event (migration) occurring. Since the MMP is a moving sample, each year new household heads are added to the overall dataset. For our survival analysis, a dataset is created such that each household head contributes a record for each year they were at “risk” of migration. This represents their individual survey year back to the year in which they were 14 years old, the beginning of our risk window. A household head is not represented in the dataset for years after their survey year (since migration data for them would not be available), and they also leave the dataset if they experience a US migration. Our analytical window ends in 2010 with the end point being determined by the availability of climate data.

Our climate measures reflect a comparison of rainfall in the year prior to the observation year relative to average rainfall over the 10 years prior for each included municipality. In this way, the measure reflects very recent conditions as compared to what might be considered a decadal “normal.”⁴ To illustrate, the “year prior” variable for 1971 represents 1970 rainfall as a percentage of the average annual rainfall 1961–1970 (the relevant 10-year “normal”). Thus, a “year prior” value for 1971 of 0.90 would indicate that municipality received 90% of the annual average rainfall in the prior year as compared to typical rainfall over the previous 10 years.

In addition to comparing recent years to a 10-year “normal”, this project reports on consideration of the climate-migration-health association across Mexico’s regions as characterized by longer-term climate trends. Therefore, based on the 30-year rainfall “normal” for each municipality, and using a quartile approach, municipalities are classified as dry, moderately-dry, moderately-wet, and wet and these classifications allow for investigation of variation of the underlying relationship between climate-migration-health across climate zones. Stratifying analyses using this quartile approach also serves to better isolate the influence of climate-health on migration within each climate zone, particularly important since there are regional differences in height and self-assessed health across zones (Table 2).

Finally, for each observation year, the individual and municipal-level socioeconomic control variables are included, with some representing time-varying data (e.g. age) and some representing static data (e.g. self-assessed health at age 14).

3.0 Theoretical Background and Prior Research

3.1 Migration’s Health Selectivity

Considerable evidence suggests that international migrants have better health than non-migrants in their origin countries as well as compared to native-born populations in destination. In fact, immigrants often exhibit lower rates of mortality, heart disease, and smoking than the native-born population in their new country (Cunningham et al. 2008; Kennedy et al. 2015; Singh and Hiatt 2006). This finding is especially striking given immigrants relatively low socioeconomic standing and less access to healthcare, conditions that are generally associated with poorer health (Derose et al. 2009; Park and Myers 2010).

The self-selection of healthier individuals into migration streams is one possible mechanism for the immigrant health advantage. Historically, support for this healthy migrant phenomenon has been most commonly found among populations from Puerto Rico (Landale et al. 2000, 2006) and Mexico (Martinez et al. 2015; Riosmena et al. 2013). As an example, Crimmins et al. (2005) used nationally representative data for older adults living in the US and found US-based Mexican immigrants over age 50 were taller than non-migrants of the same age living in Mexico. In other research, Mexican immigrants between the ages of 25–64 and living in the US reported less hypertension than their non-migrant Mexican counterparts (Barquera et al. 2008). More recently, using Mexican Migration Project (MMP) data reflecting migration between 2007–2009, Ullman et al. (2011) found Mexican migrants to the US had better health in early life using the same two proxies used in this study – self-rated health at age 14 and adult height– as compared to men remaining in Mexico.

Yet the degree of support for the healthy migrant effect often depends on the measures analyzed. For example, using longitudinal data for Mexican migrants aged 15–29 just prior to migration, Rubaclava et al. (2008) found “weak support” for the healthy migrant effect when testing six different indicators of pre-migration health. For example, rural males were positively selected on BMI and blood pressure, yet measures of self-rated health were negatively associated with migration for urban males. Among women, those from rural areas that did not report good self-rated health were more likely to migrate to the US, while urban women were positively selected both on height and self-rated health. These complicated findings suggest that context may shape the manifestation of the healthy migrant effect – a process we examine and combine with the further addition of environmental conditions.

Overall, positive selectivity has been found to be strongest when migration’s costs and risks are high (Docquier and Rapoport 2008; Feliciano 2005). When applied to health, it follows that only the fittest, healthiest individuals might attempt migration, and find success, when migration is especially risky -- such as illegally crossing the heavily militarized US border. Conversely, health selectivity is often weak or not observed in migration flows which require less risk taking, such as chain migration or family reunification—common among Dominican immigrants to the United States (Nwosu and Batalova 2014; Riosmena, Kuhn and Jochem 2017).

Among Mexican immigrants to the US, the focus of this study, positive health selection is also one explanation for the Hispanic Health Paradox (HHP). The HHP describes the surprising finding that Hispanics in the US have better than expected health given their relatively low socioeconomic status (Hummer et al. 2007; Jimenez 2011). The HHP is most commonly found in studies on adult mortality (Markides and Eschbach 2005) and the findings are especially strong for immigrants as opposed to native-born Hispanics (Hummer et al. 2000). As the ultimate indicator of health, Mexican immigrants exhibit lower mortality than non-Hispanic whites (Palloni and Arias 2004).

Some researchers critique findings that support the Hispanic Health Paradox, claiming that the data used may overestimate positive aspects of migrant health. For example, since Mexican immigrants often lack access to healthcare in Mexico (Pagan, Puig & Soldo 2007) and the US (Derose et al. 2009), negative health diagnoses may simply not occur. As a result, health may be more positively reported. Yet such data artifacts cannot fully account for the HHP. Studies using biomarkers -- such as blood pressure and metabolic and inflammatory risk -- have also found that Mexican-born US immigrants exhibit more positive health profiles than US-born individuals of Mexican heritage. In fact, Mexican-born US immigrants had similar health risk profiles to Whites in the US (Crimmins et al. 2007).

Although healthy migrants may partially explain the Hispanic Health Paradox, selective return migration to Mexico may also play a role. Often referred to as the “salmon bias” – individuals in poor health may be more likely to return to Mexico. This negative health selection could result in overestimation of good health among Hispanic immigrants in the US at any particular point in time (Abraido-Lanza et al. 1999). Palloni and Arias (2004) compared Mexican immigrants in the US to return migrants living in Mexico and did find that those residing in the US reported better self-rated health. Yet in a more recent study, Riosmena et al. (2013) combined return migrants and Mexican immigrants living in the US and still found support for the health selectivity of migration as immigrants were 3 cm taller than non-migrants, and had lower odds of reported hypertension and poor self-rated health compared to non-migrants – further bolstering the healthy migrant phenomenon and the Hispanic Health Paradox.

Sociocultural protection is the third mechanism that could potentially explain, in part, the Hispanic Health Paradox. Suggesting positive culturally-influenced health behaviors, Latinos living in neighborhoods with high concentrations of fellow Latinos exhibit better health on measures of self-rated health, mental health, cancer, and mortality, compared to Latinos living in neighborhoods with lower concentrations (Eschbach et al. 2004, 2005; Ostir et al. 2003; Patel et al. 2003). Yet, these same effects diminish with time in the US and the resulting acculturation toward less healthy lifestyles such as reduced intake of healthy foods (Akresh 2007) and increased smoking and alcohol use (Abraido-Lanza et al. 2005).

But an important gap exists in research on the health-migration connection, especially as it pertains to the positive health selection and the healthy migrant effect. Health selection processes might differ across environmentally distinct areas, variation not yet empirically explored. As reviewed below, research documents that climate change is influencing livelihoods across the globe, including household use of migration as a livelihood strategy. We posit that climate change may interact with migration’s health selection such that it could increase positive health selectivity in some regions -- where only the healthiest can migrate -- or might decrease selectivity elsewhere such that there is a general “push” toward movement away from stressed regions.

3.2 Climate and Migration

Focusing now on climate-migration scholarship, work on the environmental aspects of migration has burgeoned over the past several years while also demonstrating increasing methodological and theoretical sophistication. Overall, findings indicate that local environmental conditions interact with socioeconomic, cultural, and political circumstances to influence migration decision-making and patterns (Hunter, Luna and Norton 2015; McLeman 2013).

The climate-migration connection can usefully be considered a continuum. Disaster-induced displacement may be compelled by sudden-onset environmental shocks, while migration of a more voluntary nature has also been documented with regard to environmental factors such as shorter-term rainfall shortage and/or heat stress. Clearly trends in rainfall and temperature influence agricultural productivity and can impact mobility within rural regions, particularly those dependent on agriculture or other natural resources for sustenance and income generation (Hunter et al. 2014). Evidence of these connections has emerged from a wide variety of settings (e.g. Gray and Mueller 2012; Gray and Wise 2016), including Pakistan, Sub-Saharan Africa, and Cambodia. In rural Pakistan, for example, heat stress increases male migration as a result of its negative impacts on both farm and non-farm income (Mueller, Gray and Kosec 2014).

In the context of Mexico, rural livelihoods are extremely vulnerable to climate strain such as drought since 78 percent of households engage in farming and 82 percent of cultivated land is rain fed (Wiggins et al. 2002). Connections with migration have been demonstrated in that rainfall shortage is associated with international migration, particularly from rural areas with strong migration networks (Hunter, Murray and Riosmena 2013; Nawrotzki, Hunter et al. 2015; Nawrotzki, Riosmena and Hunter 2013; Nawrotzki, Riosmena et al. 2015). In addition, migration from drought-stricken regions is highest 2–3 years following substantial decline in rainfall, suggesting migration as shorter-term adaptation to environmental stress (Nawrotzki and DeWaard 2016).

With regard to the rural climate-migration connection, we situate our understanding in the New Economics of Labor Migration (NELM) framework which allows for the integration of environmental factors into household decisions regarding migration. NELM asserts that individuals do not make migration decisions alone; instead, social units such as families and households engage migration to maximize household income and minimize risks associated with market fluctuations and failures (Stark and Bloom 1985; Massey et al. 1993). Through the selective migration of one or more of its members, households can diversify income portfolios, particularly vital for poor rural families who may have little access to formal insurance markets and other risk management institutions (Lucas and Stark 1985; Stark and Levhari 1982). Migrant remittances offer income diversification (Rosenzweig and Stark 1989) and can serve as an ex ante risk mitigation strategy or an ex post means of weathering challenging environmental and/or economic conditions (Gray 2010; Halliday 2006).

Yet rural regions are not the only areas vulnerable to climate-related stressors. In fact, history clearly demonstrates such connections through displacement of urban residents due to environmental shocks such as hurricanes and earthquakes. Hurricane Katrina’s impact on New Orleans, Louisiana (USA) illustrates this association. The hurricane made landfall in August 2005, killing over 1200 people and causing over $100 billion (USD) in property damage. At least 1 million residents were displaced and New Orleans’ population has experienced an overall decline of approximately 25 percent since the storm (Zaninetti and Colten 2012).

In addition to extreme events, there are at least two pathways through which urban residents may be impacted by rainfall shortage and/or heat stress. First, climate change may stress urban infrastructures including transportation, energy, and water systems, thereby negatively impacting households and economies (e.g., Broto and Bulkeley 2013). The built environment’s heat island effect may intensify these challenges. Second, food production in urban areas has become an important survival strategy especially among the urban poor (e.g. Ngome and Foeken 2012). Urban agriculture increases household food security while also providing an opportunity for income generation. As an example, in Toluca Metropolitan Area, a peri-urban Mexican setting, maize production is often a buffer against economic shocks to non-farm income (Lerner, Eakin and Sweeney 2013). As in the case of rural regions, migration may represent an adaptive strategy in face of such stressors acting on urban households.

3.3 Bringing together climate, migration and health

While substantial research has linked climate-migration, as well as migration-health, as noted, there is scant research linking the triad of climate-migration-health. One area that has received research and policy attention is the health implications of forced migration in response to extreme weather events. Such displacement disproportionately affects vulnerable groups such as children, the elderly, and those with pre-existing conditions, demonstrating increased risk for a host of poor health outcomes (Toole 2005). Pre-existing conditions that shape impact include prior trauma or childhood adversities which increase the risk of disaster-related post-traumatic stress disorder (PTSD) (Bromet et al. 2016; Fritze et al. 2008). PTSD has also been shown to impact broader disaster-impacted populations, even individuals without prior mental health challenges (Tracy, Norris & Galea 2011). Mental health consequences are related not only to disaster-related trauma, but also broader social and economic issues related to climate change, particularly in already disadvantaged communities. Decreased income security, access to care, and increases in exposure to conflict and violence, anxiety related to the future, can all combine to yield mental health challenges among displaced populations in general (Fritze et al. 2008; Reuveny 2007). As another health consequence related to migration-climate, infectious disease is also common in refugee and resettlement communities, often overcrowded and lacking adequate health care facilities (Rajabali et al. 2009).

Migration may also result from actual or perceived health impacts due to climate, such as the consequences of heat waves, food and water shortages, or air quality (McMichael, Barnett and McMichael 2012). In this case, environmental factors may “push” the relocation of complete households away from hazardous settings.

Yet, distinct from these bodies of research on post-migration health, our analytical focus is on the health selectivity inherent in the process of migration itself – does climate change the “healthy migrant” effect? This question represents a different aspect of the climate-migration-health connection in that it is not focused on post-migration health consequences nor the health drivers of full-household migration. Instead, the focus is on selectivity within a household as related to livelihood migration. Although the prospect of climate to alter migrant health selectivity has been noted elsewhere (Bowles, Reuveny and Butler 2014), the analyses presented here represent one of the first empirical examinations of this association, particularly making use of longitudinal, individual-level data. Such analyses are essential for addressing health needs in both sending and receiving regions. For example, in many low-income settings, rural drought has already played a role in migration to urban areas, boosting population in informal settlements already challenged with health and hygiene issues (Costello et al. 2009).

4.0 Results

We begin by presenting two descriptive tables that reveal basic patterns in socio-demographic, health and community characteristics, including rainfall trends, with a focus on contrasts between household heads who migrated to the US during the study period and those that did not.

As demonstrated in Table 1, our key outcomes – measuring health – demonstrate patterns anticipated by the “healthy migrant” perspective in that migrants are taller than non-migrants while also being more likely to report “excellent” health at age fourteen. Both of these associations achieve statistical significance.

But key in isolating any connection between climate-migration-health is controlling for other factors known to be predictive of migration. Descriptively, and as would be anticipated based on understanding of the composition of Mexico-US migration streams, we find US-bound migrants to be younger, less educated, as well as generally more likely to be married and from rural communities as compared to non-migrants. And likely due to remittances, US migrants also tend to report slightly higher levels of household assets. Finally, migrants are also more likely to reside in drier regions of Mexico, based on a 30-year precipitation normal. In all, Table 1 reveals important socioeconomic and health differences between US-bound migrants and non-migrants that remain in Mexico.

Table 2 brings in environmental dimensions through descriptive exploration of sample characteristics across the four climate zones, representing quartiles based on the 30-year rainfall “normal.” As suggested in the table, migrant health selectivity appears to differ across climate zones. In regions characterized by relatively more rainfall, US-bound migrants are shorter in height and less likely to report “excellent” health in favor of “good” health. Alternatively, this indicates that US-bound migrants from particularly arid regions tend to report better health at age 14 – potentially indicating some positive health selectivity from regions in which rural livelihoods may be more challenged by long-term rainfall trends. This trend is also observed for adult height, as migrants from more arid regions are taller than those from more humid regions. Indeed, positive health selection is strongest in the moderately-dry climate zone, as these migrants are taller and report the highest percentage of excellent health at age 14. However, US migrants from arid regions also tend to be older and less educated as contrasted with those from regions with more rainfall, demonstrating less positive selectivity. Taken together, Table 2 preliminarily suggests that dry regions are more likely to send healthy migrants, but they are socioeconomically less advantaged.

Making use of these descriptive foundations, Table 3 moves to multivariate models using logistic regression in an event history format with random intercepts for municipality. The first model -- including only the two health measures -- does indeed reveal a “healthy migrant” effect with Mexico-US migrants being taller and more likely to report “excellent” health at age 14 relative to Mexican non-migrants. This positive health effect holds with the addition of socio-demographic controls in model 2. Thus, even with the consideration of age, marital status, educational level and assets, both adult height and self-rated adolescent health are statistically significant in predicting the likelihood of US migration among male Mexican household heads, thereby offering support for the healthy migrant phenomenon.

Table 3.

Odds ratios from random intercept logistic regression models predicting U.S. migration

	Model 1		Model 2		Model 3
Health Characteristics
Height (cm)	1.03	***	1.01	***	1.01	**
	(0.003)		(0.004)		(0.004)
Excellent health, age 14	1.59	***	1.62	***	1.64	***
	(0.09)		(0.09)		(0.09)
Demographic & SES Characteristics
Age			0.94	***	0.93	***
			(0.003)		(0.003)
Married			1.07		1.06
			(0.07)		(0.07)
Years of Education			0.98	**	0.95	***
			(0.01)		(0.01)
HH-Assets			1.05	***	1.09	***
			(0.01)		(0.01)
Contextual Characteristics
Rural					2.13	***
					(0.55)
Irrigation available					0.71
					(0.21)
Year					1.04	***
					(0.004)
Model Characteristics
Log Likelihood	−7973.98		−7669.75		−7586.58
Person Year Observations	163,773		163,773		163,773
SD of Constant	0.84	***	0.82	***	0.76	***
ICC	0.18		0.17		0.15

S.E. in parentheses

^***p<0.01

^**p<0.05

^*p<0.1

Further, this statistically significant migration-health connection remains after model 3’s inclusion of contextual control variables indicating rurality and irrigation. On these contextual controls, Mexico-US migrants are far more likely to come from rural communities as compared to their non-migrant Mexican counterparts. Migrants are also more likely to be younger and less educated.

To get to the focus of this analysis -- to explore the environmental dimensions of the migration-health connection -- Table 4 presents results across ecological zones; Intriguing connections are revealed particularly for the indicators of height, although analyses using self-rated adolescent health as a predictor were not statistically significant (self-rated adolescent health available on request).⁵

Table 4.

Odds ratios from random intercept logistic regression predicting U.S. migration using height and precip. in previous year

	Dry		Mod-Dry		Mod-Wet		Wet
Health & Climate Characteristics
Height (cm)	1.01		0.93	**	1.08		1.00
	(0.03)		(0.03)		(0.05)		(0.05)
Precip. Year Priora	0.92		0.29	**	3.03		0.77
	(0.46)		(0.15)		(2.51)		(0.61)
Height* Year Prior	1.00		1.01	**	0.99		1.00
	(0.003)		(0.003)		(0.01)		(0.005)
Demographic & SES Characteristics
Age	0.92	***	0.93	***	0.94	***	0.93	***
	(0.01)		(0.01)		(0.01)		(0.01)
Married	1.20		1.00		1.20		0.94
	(0.16)		(0.12)		(0.19)		(0.15)
Years of Education	0.92	***	0.96	***	0.99		0.92	***
	(0.02)		(0.01)		(0.02)		(0.02)
HH-Assets	1.10	***	1.05	**	1.11	***	1.18	***
	(0.03)		(0.02)		(0.04)		(0.04)
Contextual Characteristics
Rural	2.36	***	1.13		6.28	***	0.80
	(0.68)		(0.32)		(3.00)		(0.40)
Irrigation Available	1.40		0.90		1.02		0.29	***
	(0.63)		(0.30)		(0.86)		(0.13)
Year	1.05	***	1.01		1.06	***	1.08	***
	(0.01)		(0.01)		(0.01)		(0.01)
Model Characteristics
Log Likelihood	−2102.94		−2685.53		−1412.49		−1335.50
Person Year Observations	33,794		38,259		47,990		43,730
SD of Constant	0.37	***	0.38	***	0.68	***	0.67	***
ICC	0.04		0.04		0.12		0.12

S.E. in parentheses

^***p<0.01

^**p<0.05

^*p<0.1

Notes:

^aPrecip. Year Prior variable is modeled using 10-percent increments for ease of interpretation

Controlling for individual socio-demographic and community characteristics, the main effect suggests US-bound migrants from moderately dry Mexican regions are actually shorter than non-migrants, although the statistically significant interaction with year prior precipitation suggests this association is tempered by recent rainfall. Note that we include here a measure of rainfall in the prior year as a percentage of the 10-year “normal” rainfall for a potential migrant’s municipality. Specifically, in moderately dry zones, US-bound migrants tend to be taller when last year’s rainfall was relatively abundant as compared to the climate normal. In this way, the results suggest that in moderately dry regions, periods of less rain are associated with lesser health selection. In other words, health selectivity is lower during times of strain, suggesting both healthy and unhealthy household heads engage in international migration (or stay home).

While just above the 0.10 statistical significant threshold, the opposite finding emerges in moderately wet regions where the main effect suggests that US migrants are taller than non-migrants, although again the interaction with rainfall is important. Here, periods of relatively abundant rainfall reduces positive health selectivity. In this way, the results suggest that periods of more rain are associated with negative health selection. Figure 1 graphically displays these results.⁶

Figure 1.

Effects of Height on Probability of Migration by Recent Precipitation and Climate Zone

These interaction results are intriguing. To begin to make sense of them, consider first the differential meaning of recent rainfall in the different zones. In moderately-dry regions, periods of rainfall shortage are likely particularly stressful – adding challenge to livelihoods already working within a relatively dry zone. In these challenging dry times, there is no apparent “healthy migrant” effect – health is not associated with the likelihood of migration. This could either suggest healthy household heads are needed at home, or that less healthy heads are more likely to join the migration stream in order to diversify livelihoods. The MMP does not allow for disentangling these possibilities, which remains a task for future research. Moreover, in these moderately dry zones, positive health selectivity is most pronounced in times of relative rain abundance -- periods likely characterized by greater livelihood security and periods during which healthier heads are more likely to migrate. But the key suggestive finding is the fact that in moderately dry regions, health selectivity does vary by recent rainfall patterns.

On the other hand, in regions characterized by moderately wet climates, the “healthy migrant” effect is most pronounced during periods of rainfall shortage – the opposite association as exhibited in moderately dry regions (see Figure 1, although not achieving statistical significance). In these areas, livelihood strategies may be compromised if low levels of rainfall occur, challenging livelihoods with some reliance on rain-fed agriculture, for instance. During these challenging times, healthy household heads may be encouraged to leave.

Also interesting is the fact that health does not predict migration probability in particularly dry zones, nor in particularly wet zones. In these more extreme ecological regions, migration processes are not at all selective on health – US-bound migration appears to be a more universal process with negative selectivity for age and education.

5.0 Discussion and Conclusions

The results presented here examine the climate-migration-health association within the longstanding international migration stream between Mexico and the US. Intriguing distinctions in the migration-health connection are revealed when considered across ecological zones defined by aridity. Specifically, in Mexican regions characterized by consistently very wet or very dry conditions, migration does not exhibit any health selectivity, either positive or negative. Yet in some regions with more variable environmental conditions, migration does exhibit a connection with health. In moderately dry regions, such as association exists in that health is not associated with migration during times of rainfall shortage, although there is positive health selectivity in periods of higher rainfall. This could either suggest that in times of rainfall shortage healthy household heads are needed at home, or that less healthy heads are more likely to join the migration stream in order to diversify livelihoods. On the other hand, in times of rainfall surplus, healthier heads are more likely to migrate.

To interpret these distinctions, it is useful to consider the “scarcity” vs. “surplus” lens which distinguishes between conditions where “environmental scarcity” (such as low rainfall, high temperature) acts as a “push” factor – intensifying migration as a means of diversifying rural livelihoods or relocation from challenging places. In other cases, “environmental surplus” (associated with more rainfall, lack of extreme heat) may enable livelihood diversification through migration as a result of security in local, livelihood conditions.

Recall from the literature reviewed above that health selectivity is often weak or not observed in migration flows which require less risk taking. Linked to this and by way of preliminary explanation, we posit that migration networks may play a role in the low health selectivity within very wet and very dry regions. More specifically, in regions receiving relatively more rainfall, perhaps migration is less necessary as a livelihood diversification strategy – since rain fed agriculture and its related industries may be more secure. In this way, migration from these regions would be low risk since livelihoods may be otherwise fairly secure.

On the other hand, in paticularly dry areas, households may have longstanding strategies, including migration, to deal with climate strain. The social networks formed within migration corridors reduce the challenge and the “cost” of migration by providing information on employment, housing, and other important aspects of relocation. If migration, and therefore corridors, have a substantial history within very dry regions, international movement holds less risk as well and may exhibit less health selectivity. Descriptive exploration of migration processes across climate zones provides some preliminary support for these explanations in that only 14% of household heads in very wet regions have migration experience, while nearly twice that percentage (30%) in very dry regions migrated internationally during our study window. In this way, positive health selection is not necessary in times of strain, since migrants from moderately-dry regions may have more social networks on which to draw. However, in periods of relative “surplus” (more rainfall), health is linked to migration suggesting the healthy migrant effect operates when migration is perhaps less essential to immediate household well-being.

On the other hand, like households in “very wet” areas, perhaps communities in relatively wet regions have developed fewer migration-related social networks through the years due to their location in regions posing less environmental challenge (indeed, only 11.88% of household heads in relatively wet regions have international migration experience). In this case, healthier household heads are more likely to move in periods characterized by lower rainfall – perhaps forging the way for future migrants. This association would be in line with prior research suggesting positive health selectivity is strongest when migration’s costs and risks are high (Docquier and Rapoport 2008; Feliciano 2005). During periods of rainfall abundance, less healthy household heads might move perhaps due to less pressure – and therefore less livelihood risk – during times of “environmental surplus”.

In all, the key significant finding is that the healthy migrant effect manifests in times of less strain in moderately dry regions – perhaps related to the presence of social networks and the relatively different “costs” associated with international movement from these zones. Future research should continue to explore these potential intersections with social networks and the climate-migration-health triad, as well as exploring migratory responses to other forms of environmental strain such as natural disasters, sea level rise and resource-related conflict.

Of course, the present study is hampered by some limitations, mostly related to available data. First, the current analyses focus on only male heads of households, thereby not including other migration that may take place within the household. In this way, the analysis examines only a portion of the potential migrant selectivity within the household. MMP data limitations suggest that additional data sources with more detail on non-head household members, should be used to expand this analysis.

Second, we examine only 2 self-reported measures and, as such, caution must be exercised when interpreting our results. The self-reported nature of the health measures results in several forms of potential bias. As an example, there is the potential for age-height confounding in that perhaps younger men are more likely to exaggerate height. Indeed, the correlation between age-height is −0.13 (p<0.00). Such a correlation could yield inflated estimates of healthy migration if younger respondents also report higher levels of migration. Given this concern, respondent age represents an important control variable, incorporated here to respond to this potential bias. Also of concern, is the potential for confounding between socioeconomic status and self-reported health, particularly in measuring health at age 14. Respondents with lower socioeconomic status may be less likely to report poor health given their relatively lower access to health care, and therefore identification of health challenges. To address this concern, we incorporate measures of both education and household assets.

Of course, there are also genetic components to height and, as such, even accurate self-reported measures may reflect individual differences not associated with health per se. While we cannot control for genetic composition, we stratify our models by region in an effort to control for the actual height differentials across regional categories – potentially associated with local genetic expressions. These steps – including age, education, assets, and region -- allow better isolation of the relationship between health and climate on migration, albeit not perfect, nor causal.

The exclusion of other dimensions of health selection – notably mental health – would be an important extension to this work. Research has linked migration, especially international migration, to negative mental health outcomes often due to the disruption of family and friendship networks and the resulting loss of social support systems (e.g., Sluzski 1992; Lu 2010). In addition, the use of other indicators of physical health, particularly immediately prior to migration, would enhance confidence in findings. Although research has linked adult height to early life conditions, existing data do not allow for control of other important influences such as genetics. As such, the inclusion of additional pre-migration health indicators would greatly facilitate research on migrant health selectivity, while linking to environmental context also requires geographical coding.

And finally, future research should also explore health outcomes post-migration. Despite more favorable early-life health, migrants returned from Mexico to the US had higher prevalence of heart disease, obesity, smoking, and mental health disorders than non-migrants (Ullman et al. 2011).

Despite these limitations, the research presented here is suggestive and represents important progress. It calls attention to an association with scarce empirical examination in part due to lack of data allowing for research on the climate-migration-health intersection. There are virtually no data sources that provide detailed information on health, including timing of health conditions, plus detailed information on migration, including timing of movement, while also providing geographic information sufficient to append climate-related information. Such data are essential for questions such as that posed here. Future integration of these dimensions within both migration and health-focused surveys would go a long way toward allowing researchers to more fully examine these important connections.

And such examination is important since these intersections have implications for households as well as population health in both sending and receiving areas (Bowles et al. 2014). Indeed, migration provides many Mexican households with supplemental income, reducing vulnerability to climate variation and dependence on agriculture. Patterns of Mexico-US migration have already been linked to climate change – one study estimates that by 2080, 1.4 to 6.7 million Mexicans will migrate to the United States due to the negative impacts of climate change on agricultural productivity (Feng et al. 2012). Other research finds greater international migration during times of low rainfall for rural communities in historically dry parts of Mexico (Nawrotzki et al. 2013) and greater environmentally-associated migration from regions with well-established migration-networks (Hunter et al. 2013). If climate change yields pressure on less healthy individuals to migrate, the need for migrant-sensitive health systems and services may be intensified in destination regions. Further, there may be increased potential for households to be “trapped” by lack of mobility options in the event of severe livelihood strain (e.g. Adger et al. 2015). As such, we argue that given the likelihood of future climate change and livelihood response through migration, the potential for migration-health connections to be altered by climate deserves additional research and policy attention.

Highlights.

• Migration can be risky and, as such, migrants tend to be healthier than non-migrants that remain in origin communities – a pattern termed the “healthy migrant effect”.

• The possibility for climate change to alter the healthy migrant effect is examined for male household heads in Mexico, within the Mexico-US migration stream.

• In moderately dry zones of Mexico, health selectivity is less in times of drought meaning that both health and unhealthy households heads may emigrate.

• By way of explanation, since these regions experience frequent dry periods, they may have strong migrant networks since others have migrated to diversify household incomes.

• These networks would reduce the risk associated with Mexico-US migration from relatively dry zones, thereby reducing the need for healthiest migrants during times of drought. Particularly healthy individuals may also stay home to assist in local livelihood activities.

• The potential for environmental factors to reshape the healthy migrant effect deserves more research attention in order to inform health policies and programs.