Can refugees improve native children's health?: evidence from Turkey

Abstract

Following the most dramatic migration episode of the 21st century, Turkey hosted the largest number of Syrian refugees in the world. This paper assesses the impact of the arrival of Syrian refugees on the Turkish children's health, with a focus on height – a standard nutritional outcome. Accounting for the endogenous choice of immigrant location, our results show that Turkish children residing in provinces with a large share of refugees exhibit a significant improvement in their height as compared to those living in provinces with fewer refugees. Against other potential channels, a refugee-induced increase in maternal unemployment and the associated increase in maternal care seem to explain the observed positive effect on children's health.

1. Introduction

Civil wars and natural disasters constitute immediate sources of forced migration, leading millions of people to leave their home country. The total number of refugees who have been displaced from their country of birth reached almost 89.3 million at the end of 2021, half of whom are reported to be children (United Nations High Commissioner for Refugees, 2021). Unprecedented immigration waves, such as those ignited by the Syrian civil war, can affect the well-being of local inhabitants through various channels, including their health conditions. On the one hand, threats include disease transmission due to refugee mobility, depletion of natural resources, reduced per capita availability of healthcare facilities and healthcare professionals, and environmental degradation (Maystadt et al., 2019). Particularly, in the case of children, the combined impact of such factors is likely to induce a deterioration in health outcomes during early stages of life. On the other hand, the refugee inflow may boost the welfare of the host population and accelerate local economies through increased demand, the flow of financial/humanitarian aid from intergovernmental organizations, and infrastructural improvements (Aygün et al., 2021; Erten et al., 2022). Delving into whether the benefits outweigh the costs is ultimately an empirical question, as these channels work in unique directions and the magnitude of their effects is largely unknown.

We aim to explore the effect of Syrian refugee inflow on Turkish native children's anthropometrics, focusing on the z-score of height-for-age (HAZ), using 2008, 2013, and 2018 rounds of the Turkish Demographic and Health Survey (TDHS). Using an Instrumental Variable approach, we show that the Turkish children residing in provinces with large refugee share improve their HAZ: the estimated quasi-elasticity ranges between 0.02 and 0.03 (in terms of z-score standard deviations). The positive effect persists even after a series of robustness and sensitivity checks. Placebo checks confirm the causal interpretation to be given to our results. The observed positive impact contrasts with previous findings in the related literature. For example, Baez (2011) and Dagnelie et al. (2023) report an adverse effect of the immigration shock on the native children's anthropometrics in Tanzania and in 29 African countries, respectively. This contradiction is fairly puzzling. We therefore examine several alternative channels to elucidate the observed positive effect.

The hypothesised mechanisms consist of changes in the supply of healthcare resources, mother's labor market outcomes, income, time allocation, quality–quantity trade-off, and investment in children's human capital through vaccination. The labor market and associated mothers’ time for child care channels are found to be the most plausible explanation. We show that Turkish mothers who reside in provinces with a large share of refugees are less likely to be in the labor force at the time of the survey. Similarly to Tumen (2018), we find that refugees reduce the odds of low-educated Turkish mothers being on the labor market. That is explained by strong labor substitution since refugees are – on average – low-skilled and are willing to take low-pay jobs. Paradoxically, improvements in anthropometric indicators are concentrated in households where low-educated mothers seem to have been replaced by Syrian immigrants in the workplace. Such a mechanism may seem surprising. However, it has been documented in other contexts that maternal unemployment – in particular among low-skilled mothers – may boost the quality of maternal time to spend with their offspring and have positive repercussions on their children's health (Mosca et al., 2017; Danzer Lavy, 2018; P. M. Anderson et al., 2019). For example, maternal unemployment can result in more frequent utilization of healthcare resources, better maternal supervision through the provision of rich-nutrition diets or through the longer duration or exclusivity of breastfeeding, each of which can positively contribute to the physical growth of children. Taken together, we attribute the positive health effect of the refugee inflow to migration-induced maternal unemployment among the least skilled mothers.

Our contribution is twofold. First, we provide evidence of the health effects of the largest migration movement of the 21st century. The existing evidence mainly focuses on refugee camps in Africa (Baez, 2011; Dagnelie et al., 2023). In the Middle East, most refugees are dispersed in their hosting countries (UNHCR, 2022). It is unclear how such a dispersion matters. On the one hand, the dispersion of refugees may reduce the risk of disease transmission and potential competition in accessing resources such as clean water, sanitation, or basic health services (Maystadt et al., 2019; Clemens Ginn, 2020). On the other hand, the concentration of refugees in camps may facilitate the provision of basic health services and facilitate the control of communicable diseases. In Turkey, two papers touch upon the health consequences of hosting refugees. Aygün et al. (2021) investigate the impact of refugees on healthcare system and mortality outcomes (i.e., infant, child, and elderly mortality). Their results suggest that Syrian refugees placed a considerable burden on healthcare system, despite the central government's investments in the physical and human capital indicators of healthcare resources in response to refugee shock. However, their estimates report no evidence of an impact on the Turkish natives’ mortality outcomes. Erten et al. (2022), in turn, mainly concentrate on the vaccination outcomes of native children in addition to the prevalence of infectious/non-infectious diseases among native children. Their benchmark results document a drastic reduction in the probability of being fully immunized against hepatitis B, tuberculosis, and measles. They attribute their findings to the refugee-induced supply constraints in the healthcare resources. Although we will discuss potential effects on the supply of healthcare services and mortality, our article focuses on the anthropometrics of children, that is, the HAZ z-score, a standard outcome in health economics¹ Improved early life health outcomes and sufficient nutritional intake are key to explain later life outcomes such as better cognitive abilities, higher wages in the labor markets, and reduced health risks (Case Paxson, 2010; Currie Vogl, 2013). A change in mortality and health care resources is difficult to interpret without knowing the impact on children's nutritional status, and inversely. Our paper also sheds light on the mechanisms that drive possible health impacts of refugees. Existing studies emphasize the transmission of diseases such as malaria in Africa (Dagnelie et al., 2023), or other vaccine-preventable diseases (e.g., chickenpox and tuberculosis) and sexually transmitted diseases (e.g., syphilis) in Latin America (Ibáñez et al., 2021), and/or overburdened healthcare resources (Tatah et al., 2016; Aygün et al., 2021; Erten et al., 2022). Against these channels, we highlight the role of another mechanism through which immigration can affect health outcomes: the labor market channel and the associated mother's time allocation.

The remainder of this paper is structured as follows. Section 2 provides information on the arrival of Syrian refugees in Turkey and their integration into the Turkish healthcare system. Section 3 provides a description of the data. Section 4 elaborates on the identification strategy. Empirical results are presented in Section 5 while potential mechanisms are explored in Section 6. Finally, the last section discusses the findings and the need for extended child care in Turkey.

2. Background

2.1. Syrian refugees

The Syrian civil war sparked in March 2011, resulting in a forced displacement of 7.5 million Syrian individuals. Due to the geographical proximity and the open-door policy of the Turkish government, Turkey has become the main destination for Syrians (Ferris Kirisci, 2016). The open-door policy is of paramount importance, as it eliminates any attempt to enter the country illegally. In the early wave of the refugee influx, Syrian refugees, under temporary protection, were placed in refugee camps. Although the Turkish government has established 26 refugee camps in 10 provinces that provided housing to more than 270,000 Syrians by the end of 2013 (Benner et al., 2015), the uninterrupted nature of refugee inflow put a heavy burden on the capacity of camps in 2014. The number of refugees reached 2.5 million in 2015, and 3.5 million in 2018 (Erdoğan, 2014; Akbulut-Yuksel et al., 2022). The massive increase resulted in the spread of refugees in 81 provinces of Turkey. As of 2019, the share of refugees residing in camps was 2.4%. In 2019, Istanbul, the most populous province in Turkey, hosted the largest number of refugees, amounting to 548,000 individuals. Istanbul was followed by provinces close to the Syrian border (e.g., 443,000 in Gaziantep and 431,000 in Sanliurfa, and 430,000 in Hatay) (Ministry of Interior (MOI), 2022). Unlike other countries that have a border with Syria (e.g., Jordan and Lebanon), Turkey has experienced a significant increase in the number of refugees overtime. Figure 1 depicts time series pattern of the number of officially registered Syrian refugees in Jordan, Lebanon, Egypt, and Turkey. It is obvious from Figure 1 that Turkey continues to host the largest number of immigrant populations among the other host countries.

Figure 1. Total number of refugees (in 100 K) by country,. Notes: The data on the registered number of refugees comes from the UNHCR.

2.2. Syrian refugees in Turkish health system

Refugees came to Turkey in need of healthcare. In addition to forced displacement, property damages, and economic destruction, the civil conflict in Syria resulted in a dramatic deterioration of healthcare services. Kherallah et al. (2012) report that both civilians and healthcare professionals have experienced severe physical injuries and mental health deteriorations including post-traumatic stress disorder. Furthermore, the war prevented healthcare facilities from operating effectively. According to WHO (2012), in the second year of the war, almost 40% of the hospitals were out of service and nearly 65% of the healthcare providers left their home country. Thereby, victims that were in urgent need of medical intervention were barely treated. WHO (2012) also state that the war left Syrians in an insanitary environment due to limited access to medical services, food, and drinking water. Resource constraints and unhygienic living conditions are known to increase the risk of epidemic diseases. WHO (2012) therefore warns that Syrians constitute a significant threat to the public health of host communities when considering the risk of spreading epidemic diseases such as measles and tuberculosis. The prevalence of measles, tuberculosis, and malaria have been reported to increase among countries providing temporary protection to Syrian refugees (WHO, 2012). For the Turkish case, the number of chickenpox, measles, tuberculosis, and hepatitis A cases increased both inside and outside of the camps. However, the incidence rate of these diseases among Syrian immigrants was found to be considerably larger as compared to the Turkish natives (Leblebicioglu, 2016; Ekmekci, 2017).

The Turkish government has adopted an ambitious response to the provision of healthcare services to forcibly displaced Syrian people who were in immediate need of medical treatment (Aygün et al., 2021). According to the existing policy, Syrians can benefit from any type of healthcare facility (e.g., preventive healthcare facilities, emergency departments, first-, second-, and third-step healthcare centers) in their province of registration with full coverage of medical and treatment costs. Syrian immigrants under temporary protections are recognized as the beneficiaries of universal health coverage just as the Turkish native population. In addition, the Turkish government began to set up Health Centers for Refugees (HCR) in 2017. In the 29 provinces of Turkey, the government recruited native Syrian healthcare professionals (i.e., more than 700 doctors and 900 nurses) and bilingual healthcare professionals (European Commission, 2006; Aygün et al., 2021). Nevertheless, the majority of HCRs were established after the implementation of the Supporting Immigrant Health Services Project (SIHSP), which was fully funded by the European Union (EU). Prior to the SIHSP, the number of HCRs located in provinces with a larger share of immigrants was limited to six.

Risks of health congestion were reported. For example, Savas et al. (2016) conducted a survey to investigate the effect of Syrian individuals on the workload of healthcare professionals in one of the provinces with a large share of refugees, Hatay. Their results indicate that waiting time at hospitals and healthcare providers’ hours worked significantly increased because of the Syrian patients. This study also highlights two crucial issues. First, intensive care units became insufficient in capacity, again due to Syrian patients. Second, the prevalence of infectious diseases, hospitalization rates, and intensive care receipts are found to be much higher among Syrian patients compared to Turkish patients. Nevertheless, Aygün et al. (2021) take this survey-design study one step further by combining several administrative data sources. They provide suggestive evidence indicating that adult intensive care beds decline in per-capita terms. Similarly, WHO (2019) reports that almost 40% of patients in hospitals in (Turkish) provinces that have a border with Syria are Syrian individuals. Therefore, provinces with higher refugee shares are more likely to encounter capacity problems, and the native population should be at higher risk of having infectious diseases. The breadth of the European Union (EU) financial assistance was nonetheless not negligible when considering the scope of the EU Facility for Refugees in Turkey (FRIT). The FRIT aims to ensure the socioeconomic well-being of Turkish natives as well as Syrian refugees. Knowing whether increased household income induced by external financial aid can compensate for potential negative effects on children's nutrition remains an empirical question.

3. Data and descriptive statistics

This paper combines two types of data for the empirical analysis: (1) individual level data on the Turkish native children's health outcomes as measured by anthropometric information, (2) 81-province level data on the Syrian refugee inflows as well as other province-level information such as the number of physical capacity indicators of healthcare (i.e., number of hospital beds and number of hospitals), public expenditure, GDP per capita, and a terrorism index as a measure for local violence intensity.

3.1. Data on children's anthropometric indicators

The primary source of data is the 2008, 2013, and 2018 waves of the Turkish Demographic and Health Survey (TDHS), each of which is a nationally representative cross-sectional survey. The TDHS provides comprehensive information on women of reproductive age (15–49) who gave birth in the five years preceding the survey year and their offsprings. Collected information includes, but is not limited to, household characteristics (e.g., information on household head and wealth index), demographics of mothers and their children (e.g., sex, age, birth year/month, and education), mothers’ marriage history (e.g., marital status and age at first marriage), fertility outcomes (e.g., current fertility, children ever born and living, and age at first birth), and utilization of healthcare services (e.g., antenatal/postnatal care visits, vaccination) and child health and nutrition (e.g., height and weight).

A vital dimension of early childhood welfare, proximate predictor of human capital formation, is investigated in our paper: nutrition. According to international standards (WHO, 2006), HAZ z-scores can be calculated for 5892 children, due to either rejection of height measurement and the absence of the child (ren) at the time of the survey. However, our operating sample consists of 5341 children under the age of five, as we exclude those who were born in year 2012. The reason why we exclude the year 2012 from our empirical analyses is the lack of refugee data (see Section 3.2). It is also of great importance to indicate that anthropometric information is only available for the last-born children, who are below the age of 5. The HAZ, given sex, is a widely used indicator of child health in the long term. A low HAZ reflects the cumulative impacts of chronic malnutrition, stunting, and infections before and after birth. Being stunted, more prominent under five years of age, is considered to originate from food intake at sub-sufficient levels over a prolonged period, vitamin and/or protein deficiencies, co-existence of multiple diseases, and the poor health status of mothers during gestation. Table 1 shows the descriptive statistics on the Turkish children's HAZ z-scores. On average, a child in Turkey is 0.334 standard deviation shorter during the sample period in refugee-dense provinces. Interestingly, the HAZ z-scores, on average, have improved after the immigration shock. The magnitude of this improvement is higher in refugee-dense provinces.

Table 1. Children's anthropometric indicators by native-to-refugee ratio density in provinces (excluding 2012)

Notes: The data come from Turkish Demographic and Health Survey (TDHS) 2008, 2013, and 2018. The entries are at means. Standard deviations are in parentheses. High intensity provinces are as follows: Kilis, Hatay, Sanliurfa, Gaziantep, Mardin, Sirnak, Osmaniye, Kahramanmaras, Mersin, Adana, and Adiyaman.

3.2. Data on refugees and time varying province attributes

Data on the number of refugees in Turkish provinces come from the Directorate General of Migration Management (DGGM). Before 2012, there were almost no Syrian refugees in Turkey, although there were officially registered refugees and asylum seekers from different countries. For example, UNHCR (2011) reports that there were 14,465 refugees in Turkey in 2011 whom of 6,600 were from Iraq. For the particular case of Syrian war victims, their arrival in Turkey began in July 2012 (Akbulut-Yuksel et al., 2022). However, refugees have been placed in refugee camps in Turkish provinces having a border to Syria in 2012. As a result, data on the number of refugees at the province level, and therefore treatment, start in 2013.² We also divide the number of refugees by the province-level 2010 population. Population data, at the 81 province-level, are obtained from the Turkish Statistical Institute (TSI). Figure 2 shows the spatial distribution of Syrian refugees per inhabitant across 81 provinces of Turkey in the years 2014, 2016, and 2018, respectively. As discussed in Beine et al. (2021), refugees progressively move to other regions, especially to industrialized and western provinces such as Istanbul. Besides, in 2018, a substantial portion of the dark-green shades still appear in provinces having either a border or close to Syria. This highlights the importance of distance as a significant factor in determining the location of Syrian refugees in Turkey (Beine et al., 2021).

Figure 2. Spatial distribution of Syrian refugees (in percent). Notes: Data on native population come from the Turkish Statistical Institute while the data on refugees come from Presidency of Migration Management. This figure shows the ratio of the Syrian refugee population to the province native population in Turkey for 2014, 2016, and 2018, respectively.

As indicated in Figure 2 and already documented by Aygün et al. (2021) and Aksu et al. (2022), refugees tend to locate in provinces that either have a border to Syria or are close to Syrian border (i.e., Southeastern provinces). With a few exceptions (i.e., Adana, Gaziantep, and Mersin, which are important epicenters of industrial economy), refugees tend to locate in provinces that are relatively less developed such as Kilis, Hatay, Sanliurfa, Osmaniye, and Kahramanmaras (i.e., mainly east part of Gaziantep province). Such a negative selection is common in refugee studies (Maystadt Verwimp, 2014; Wahba, 2014; Kadigo Maystadt, 2023). Therefore, we can anticipate that native children should have worse HAZ z-scores in refugee-dense provinces, at least before the refugees arrive. The descriptive statistics in Table 1 confirm that prior to 2012, HAZ z-scores are lower in provinces with larger share of refugees.

There exist several factors at the local level that could affect health status and be correlated with the presence of refugees. We use the following time-varying province controls: (1) the number of hospitals per 1,000 inhabitant, (2) public expenditure per 1,000 inhabitant, and (3) an index for terrorism. The first two variables are included to take into account public capacities in the provision of healthcare services and other local public goods. Given the high risk of terrorism in Turkey, the last index aims at capturing the later-life health consequences of early-life shocks related to terrorism. Terrorism threats have indeed been found to impair later life outcomes, including children's anthropometry (Camacho, 2008; Ekhator-Mobayode Abebe Asfaw, 2019; Grossman et al., 2019). Since these variables can also act as bad controls (J. D. Angrist Pischke, 2009), these variables will be added parsimoniously in our main specifications.

Data on the number of hospitals come from the Turkish Statistical Institute (TSI). Public expenditure data are obtained from the Turkish Republic Presidential Strategy and Budget department.³ Overall, the supply of health services or the level of public expenditures have improved over the period of investigation (see Appendix Table A.1). The observed improvement – irrespective of the arrival of refugees – can be attributed to the Health Transformation Program (HTP) implemented by the Turkish Ministry of Health (MoH) between 2003 and 2013, and the Health System Strengthening and Support Project supported by the World Bank since 2015.⁴ A terrorism index is constructed using data from the Global Terrorism Database (GTD). The GTD provides comprehensive information on the terror events that took place in each province: (1) the number of incidents, (2) the number of fatalities, (3) the number of injuries, and (4) the number of property damage. These four measures and the formulation provided by Vorsina et al. (2017) are utilized to construct the terrorism index.⁵ According to the Appendix Table A.1, the inflow of refugees has coincided with a rise in terrorist risk.⁶

4. Identification strategy

Our research design seeks to compare the average health outcomes of children exposed to the presence of refugees with children who are not (or less) exposed. More specifically, a naïve empirical specification can be presented as follows:

(1)$$Health_{icrt} = \beta Refugee \ Share_{ct} + \theta H_{ict}' + \gamma X_{ct}' + \alpha_{c} + \alpha_{t} + t_{r} + \epsilon_{icrt}$$

Health _icrt is the health outcome of the child i, residing in province c (and region r) in year t. Refugee Share _ct denotes a continuous treatment variable capturing the exposure to the presence of refugees for the child i leaving in province c. Exposure to refugee shock is determined by the number of refugees in the province of residence of the children at the time of birth, divided by the 2010 population of the province c. Since our main variable of interest has an abundant number of zeroes, in particular prior to 2013, we apply an Inverse Hyperbolic Sine (IHS) transformation.⁷ We control for individual and household observed characteristics (H _ict) including child's sex, child's month of birth, mother's age, mother's age square, mother's education, whether the mother resides in a rural area, the total number of older siblings, being a female headed household, and the wealth index. The vector of coefficients γ will capture the impact of time varying province attributes, namely (log) hospitals per 1,000 inhabitant, (log) public expenditure per 1,000 inhabitant, and the terrorism index. α _c stands for province fixed effects to control for any time-invariant heterogeneity across 81 provinces. α _t denotes survey year fixed effects to account for any macroeconomic shocks at the national level. The specification will be further augmented with region-specific time trends, t _r (and alternatively with region-year fixed effects).⁸ $\epsilon _{icrt}$ stands for the error terms. Standard errors are clustered at the 81-province level. Individual sampling weights are also used to render our results nationally representative.

Despite controlling for observed and unobserved characteristics, a challenge in comparing children with different levels of exposure to the presence of refugees at birth is that the location of refugees is not random. Refugees have indeed been found to move within Turkey, even if they appear to be particularly sensitive to variations of income at origin and distance (Beine et al., 2021). Refugees may therefore prefer to move to provinces with employment opportunities and better healthcare. The opposite is also possible. We cannot exclude that refugees are forced to settle in peripheral areas in which local hosts feature worst health conditions compared to the rest of the country.⁹ Syrian refugees might also prefer to locate in less populous and small-sized provinces to minimize their living costs. If labor market opportunities and government services are relatively limited in such provinces, then the coefficients obtained through the OLS model would be downward biased.

The endogenous choice of the refuge location makes it essential to instrument the exposure of Turkish children to the immigration shock. To account for the potential endogeneity of immigrant location decisions, we use a distance-based instrument following Del Carpio Wagner (2015). The instrumental variable approach is based on the rationale that the travel distance from the Syrian governorate from which Syrian immigrants take the road to each province in Turkey is central in predicting their settlement decisions. The gravity models of migration posit that the farther the two locations are from each other, the lower the spatial interaction, or namely migration, between them (J. E. Anderson, 2011; Beine et al., 2021).

The TDHS provides data on the province of residence of the respondents at the time of survey. It allows us to calculate the cross-border variation in travel distance between 81 provinces in Turkey and 13 governorates in Syria.¹⁰ Following Del Carpio Wagner (2015), the distance-based instrument for the Turkish children's exposure to Syrian refugees for each province in Turkey can be represented as follows:

(2)$$Distance \ Instrument_{ct} = \sum_{i = s}{{1\over \eta_{sc}}} \phi_{s}S_{t}$$

where η _sc denotes the shortest travel distance (measured in kilometers using Google Maps) from each origin Syrian governorate s to a Turkish province c. As noted by Erten et al. (2022), there are six border crossing points from Syria to Turkey, two in Hatay and the other four in Gaziantep, Kilis, Sanliurfa, and Mardin provinces. Therefore, choosing between these border crossing points are contingent upon the Syrian individuals’ governorate of residence and the intended province of arrival in Turkey. Due to the Turkish government's open-door policy toward Syrian refugees, there was no incentive for them to target one particular border crossing point while entering Turkey. Hence, the distance measure between 81 Turkish provinces and 13 Syrian origin governorates considers the shortest travel pathway. ϕ _s shows the share of Syrian population in each origin governorate s in the pre-immigration period (i.e., 2010)¹¹ , and S _t is the total number of officially registered Syrian refugees in Turkey in year t (measured in thousands). There are 1053 origin-arrival pairs to construct the Instrumental Variable (IV).¹² The instrument is also standardized in order to ease the interpretation of the first-stage results.

The IV method relies on the identifying assumption that the instrument has no correlation with the unobserved trends in the outcome variables. For instance, such assumption may be threatened when the instrument is correlated with the latent trends in macroeconomic indicators (e.g., (un)employment). Such trends could also affect the Turkish children's anthropometrics. Furthermore, Western region of Turkey is considerably more developed as compared to Eastern region, which is close to border crossing points. If the time trends in children's health outcomes in Western region of Turkey differ from that of Eastern region, then the identifying assumption will no longer hold. To minimize that threat, we follow Aygün et al. (2021) in using time–region interactions to weaken the independence assumption. We further discuss the validity of our instrumental variable approach in Section 5.3.

5. Empirical analysis

5.1. Main results

Table 2 presents the effect of refugees on children's nutritional status, as measured by the HAZ z-scores. Columns (1)–(3) show the OLS estimates. Columns (4)–(6) display the 2SLS results. In panel A, we only include survey year and province fixed effects. In panels B and C, we gradually add different sets of control variables. The first stage F-statistics (or, Kleibergen–Paap rk Wald F) are reported in columns (4)–(6), each of which is sufficiently large. This confirms that the distance-based instrument is strong enough for proper identification.

Table 2. Effect of refugees on children's HAZ (excluding 2012)

Notes: Data come from Turkish Demographic and Health Survey (TDHS). The full sample is for the 2003–2018 period, excluding 2012, at the 81-province level. The 2SLS model instruments the refugee share utilizing a distance-based instrument. Individual and household controls include child's sex, child's month of birth, mother's age, mother's age square, mother's education, whether the mother resides in rural area, total number of older siblings, being a female headed household, and wealth index. Time-varying province controls include (log) public expenditure per 1,000 inhabitant, number of hospitals per 1,000 inhabitant, and terrorism index. Individual survey weights are used in each specification. Standard errors, clustered at the 81-province level, are in parentheses. *** denotes statistical significance at the 1% level (p < 0.01), ** at the 5% level (p < 0.05), and * at the 10% level (p < 0.10), all for two-sided hypothesis tests.

Unlike previous findings in the African context (Baez, 2011; Dagnelie et al., 2023), the 2SLS estimates suggest a positive effect of refugee inflow at 1% significance level in each panel, regardless of the specification.¹³ Turkish children residing in provinces with larger refugee share are on average taller than others.¹⁴ The estimated quasi-elasticity stands at 0.0296 in our preferred specification with individual/household-level controls (see column (6) in panel B of Table 2). Increasing the presence of refugees by 10% (equivalent to an increase by about 2,900 refugees from the mean) boosts children's height by one-third of a standard deviation (0.0296 × 10 = 0.296). Although with the reverse sign, the magnitude is slightly higher than the quasi-elasticity of –0.02 found by Dagnelie et al. (2023). Although the magnitudes and the corresponding quasi-elasticities become slightly smaller in panels B and C, we still observe a notable improvement in HAZ when gradually adding (i) individual/household-level and (ii) province-, individual/household-level control variables. For presentation purposes, we relegate the estimated coefficients for the full set of control variables (i.e., panel C of Table 2) in Appendix Table A.3. Results are mostly as expected with positive and significant coefficients for proxies for healthcare supply, education, and wealth. The number of older siblings has a negative and significant effect on HAZ.

5.2. Robustness checks

The results of our estimation may be sensitive to the model's specification. Therefore, we check the robustness of our results to (a) alternative transformations of our dependent variable and our variable of interest, (b) the use of more requiring region-time trends and fixed effects; and (c) alternative samples.

Alternative transformations

We may be concerned by extreme values in our dependent variable. The literature defines biologically plausible values of HAZ z-scores, which lie between −6 and +6 (Mei Grummer-Strawn, 2007). We show that our main results are robust to the exclusion of biologically implausible values of the HAZ z-scores (see Appendix Table A.5). Another potential concern might be related to the utilization of IHS transformation in refugee share variable. To rule out the possibility that IHS might distort our results, we re-estimate our main results shown in Table 2 where IHS transformation is not applied to the refugee share variable. As can be seen from Appendix Table A.6, the positive effect on the HAZ remains the same without IHS transformation.

Alternative time trends

Having data at the 81-province level allows us to utilize more flexible trends than region trends, such as trends at the NUTS-1 (i.e., 12 regions) or even NUTS-2 level (i.e., 26 regions). We include these requiring time trends to make sure that our findings are not driven by differential pre-existing time trends across regions. The results are presented in Appendix Table A.7 and Appendix Table A.8, confirming the improvement on the HAZ.

Alternative samples.

The definition of our sample of interest may be questioned. Due to a lack of refugee data, we decided to exclude the year 2012. We re-estimate the main specifications including the year 2012, replacing missing information on refugees by zeroes. Results remain similar (Appendix Table A.9).

5.3. Plausibility of identification assumptions

Our identification strategy rests on key identification assumptions. First, we assume that there is no confounding trends in health outcomes between provinces with high and low values of refugee shares or the instrument, conditional on the use of fixed effects and control variables. Second, we assume that the composition of the treated and control provinces remains constant overtime. The stability of our 2SLS coefficients when controlling sequentially for province-level, individual, and household characteristics is reassuring but only deals with observed changes. The stability of our sample composition may be threatened by several selection issues. We discuss the plausibility of both assumptions below, distinguishing between selective migration, selective marriage, selective fertility, and selective mortality as threats to the second assumption.¹⁵

Confounding trends

Adding region-specific time trends may not be sufficient to assume that Turkish children residing in provinces with large (predicted) share of refugees follow a similar trend in their HAZ z-scores compared to children in provinces with lower share of refugees. Therefore, we assess the presence of pre-existing trends using two different exercises (i) quantifying the effect of the future exposure to refugee shock on the current HAZ z-score of Turkish children (i.e., using the forward values of the refugee share and the instrument) and (ii) following Aksu et al. (2022)'s Placebo exercises. First, we restrict the sample to children born before 2008 and assign the 10 lagged values of the refugee share and the distance-based instrument for each province. In Appendix Table A.10, we find a positive correlation with the OLS estimates but the estimated 2SLS coefficients are all statistically insignificant. Overall, such a placebo exercise rejects the possibility that the positive effect found in our main results may be explained by pre-existing trends in HAZ z-scores. Second, we follow Aksu et al. (2022) in estimating the slope coefficient from a regression of residual trends of the dependent variable (i.e., HAZ) on the value of the time-constant instrument (i.e., 2016). As suggested by Jaeger et al. (2020), the aim of this Placebo approach is to assess whether pre-shock trends in the outcome variable are independent of the instrument, after accounting for fixed effects, time trends, and/or control variables. The results are presented in Appendix Table A.11. In panel A, region-trends in columns (2) and (4) and region-year fixed effects in columns (3) and (5) are introduced using 5 regions of Turkey. In panels B and C, we introduce region-trends in columns (2) and (4) and region-year fixed effects in columns (3) and (5) using 12- and 26-regions of Turkey, respectively. Accros specifications, there is no correlation between the residual trends and the IV defined in 2016.¹⁶ In the rest of the paper, we focus on specifications with individual and household covariates, with or without region-specific trends and region-year fixed effects.¹⁷

Selective migration

Native displacement is a major concern in migration studies (Card DiNardo, 2000; Card, 2001; Borjas, 2006; Andersson et al., 2021). A massive outflow of natives may explain our results, in particular if those with a low socio-economic background (likely with poor health conditions) are leaving. To explore such a selective migration, we first note that Akgunduz et al. (2015) find lower in-migration rates and unchanged out-migrations in regions that hosted a large number of Syrian refugees in Turkey. Aksu et al. (2022) also find no impact on in-migration, with the exception of a positive impact for more-educated natives.¹⁸ With our data, we investigate whether the migration influx affects migration inflows and outflows based on the TSI's migration statistics at the 81-province level over the period of 2008-2018.¹⁹ Based on the results presented in Appendix Table A.12, we do not find evidence of native displacement as a result of the refugee inflows. Similarly, we do not report any evidence of outflows as a response to the inflows of refugees. However, we do find that low-educated natives are more likely to move in areas with a large share of refugees (see panel B, columns (2) and (3) of Appendix Table A.12).²⁰ If anything, the fact refugee-hosting areas act as an attraction force for the low-educated suggests that our main results would have been even more positive, if – as expected – low-educated natives are also associated with lower health outcomes for their children.

Selective marriage and selective fertility

Unprecedented population shocks, such as those triggered by the Syrian civil war, are known to potentially alter marriage market dynamics and fertility decisions of native women.²¹ The presence of refugees has the potential to impact patterns of family formation and fertility decision by modifying the employment opportunities and occupational status of native men and women (Carlana Tabellini, 2018). We cannot exclude the possibility of a selective change in our population of interest, since marriages and births at early ages have been linked with a number of adverse outcomes such as poorer maternal and child health, especially in developing countries [see Alam (2000) for India and Raj et al. (2010) for Bangladesh]. To test these possibilities, we explore the effect of refugee influx on Turkish women's age at marriage and age at first birth. In Appendix Table A.13, we report no evidence of a change in age at first marriage and age at first birth, indicating that our results are not driven by a change in family formation patterns.²² No evidence is also found when the sample is stratified between low- and high-educated mothers.

Selective mortality

The Fetal Origins hypothesis posits that in utero shocks (e.g., sub-optimal nutrition, wars, and weather shocks) impair fetal growth which in turn leads to a predisposition to have poorer health outcomes in later life (Almond Currie, 2011; Barker, 1990). Shocks in utero may also alter the composition of the population of interest (Dagnelie et al., 2018). In response to a detrimental shock during gestational age, a portion of fetuses lying at the bottom of the health distribution cannot survive since their initial health endowment does not exceed the survival cutoff (Dagnelie et al., 2018). We can therefore not exclude that in case of disproportionate mortality in refugee-hosting provinces, our positive effect would reflect the better health conditions of the surviving kids. We first note that such a selection is highly implausible since most of evidence has been found for large-scale shocks such as famine, conflict, or natural disasters (Almond Currie, 2011; Leon, 2012; Lavy et al., 2016; Dagnelie et al., 2018). We nonetheless test whether Syrian refugee shock affects the survival patterns during pregnancy, and thus the population composition. In Table A.14, we assess the likelihood of experiencing miscarriage and stillbirth among Turkish women. Our results do not yield a statistically significant effect on the probability of experiencing miscarriage and stillbirth (see columns (1) to (6) in Appendix Table A.14). Although not obvious in the studied context, induced abortion might also constitute another mechanism through which immigration can affect the composition of our sample and hence, native children's health. In Appendix Table A.14, we assess this possibility and report null results.²³

6. Investigating possible channels

Our findings contradict the existing research documenting either an adverse or a null impact of migration shocks on health outcomes (Montalvo Reynal-Querol, 2007; Baez, 2011; Ibáñez et al., 2021; Dagnelie et al., 2023). In the Turkish context, there exist two relevant studies exploring the health effects of forced migration. First, Aygün et al. (2021) examine the health effects of Syrian refugee shock in Turkey, as proxied by Turkish natives’ infant, child, and elderly mortality. Their OLS results provide suggestive evidence indicating that mortality outcomes exhibit an increasing pattern. Yet, once the endogenous location choice of Syrian refugees is accounted for, their IV estimates yield no evidence that forced migration affects mortality. The study indicates that Syrian individuals are more prone to locate in provinces for which, in the absence of the migration flow, mortality outcomes would follow a more negative trend over time. That being said, Erten et al. (2022) show that native children residing in provinces with larger refugee flows are less likely to be fully vaccinated against measles, hepatitis B, DTP, and tuberculosis. They also document that the native children living in refugee-intense regions are at higher risk of catching an infectious disease and upper/lower respiratory diseases. Taken together, the results displayed in this paper are sufficiently puzzling to require further investigation.

We explore a number of channels through which an unprecedented immigration crisis can affect health outcomes. These channels include (1) investment in healthcare resources, (2) Turkish mothers’ labor market outcomes and their time allocation at home, (3) income channel, (4) and the quality–quantity trade-off channel as measured by the total number of births (either still or live birth) and investment in children's human capital through vaccination. To explore these channels, we keep similar OLS and 2SLS specifications and samples, than the ones explained in Section 4. The only differences are that some outcomes (healthcare) require to aggregate our data at the province level. Time-varying province attributes are omitted since they are likely to be endogeneous to the outcome of interest related to healthcare resources, labor markets, income, the number of births, and investment in children's human capital.²⁴

6.1. Investment in healthcare resources

The most obvious explanation relates to changes in healthcare resources. We therefore explore the response of the central government to the refugee wave in terms of investment in human and physical healthcare resources in the refugee-hosting provinces. Using data from the TSI, we quantify the effect of refugees on the supply of healthcare resources at the 81-province level. The province-level healthcare indicators cover: (1) the number of doctors, (2) the number of nurses, (3) the number of midwives, and (4) the number of hospital beds. Descriptive statistics are shown in Appendix Table A.1.

We present our detailed results in Appendix Tables A.16 but a summary of the 2SLS results is provided in Table 3. The 2SLS estimates indicate that there is a significant increase in the number of doctors, nurses, midviwes, and hospital beds. Overall, the results confirm that the central government responded to this migration shock by increasing the supply of healthcare professionals and hospital beds (see panel A of Table 3). However, one should be cautious when interpreting these results as the outcome variables are not in per capita terms. To proxy for the capacity to maintain sufficient healthcare services per inhabitant, we divide the healthcare outcomes by the total population, including both Turkish natives and Syrian refugees, in each province. When transformed in per capita terms, the effect is reversed for the supply of healthcare professionals and not significant for the number of hospital beds (see panel B of Table 3).²⁵ Similar to Aygün et al. (2021), we do report a negative effect for doctors and no effect for hospital beds, both are in per capita terms.

Table 3. Summary table I: investment in healthcare resources (2SLS results , with region-year FEs)

Notes: Information on the healthcare resources variables is obtained from Turkish Statistical Institute (TSI). The full sample is for the 2003–18 period, excluding 2012, at the 81-province level. Each cell shows the estimates for the ratio of migrants to natives, with year, province fixed effects, and region-year fixed effects. The 2SLS model instruments the refugee share utilizing a distance-based instrument. Standard errors, given in parentheses, are clustered at the province level. *** denotes statistical significance at the 1% level (p < 0.01), ** at the 5% level (p < 0.05), and * at the 10% level (p < 0.10), all for two-sided hypothesis tests.

Overall, our results suggest that the central government has made significant investments in healthcare resources in refugee-receiving provinces. Yet, the magnitude of this investment did not sufficiently compensate for the reduction in per capita availability of healthcare professionals in provinces with more refugees. Provinces that experience a disproportionate amount of refugee inflows encounter a significant shortage of medical professionals relative to provinces with less refugees. The supply of healthcare services does not seem to explain our positive effect on children's health. We should nonetheless acknowledge that we cannot exclude the possibility that the overall health system has been affected by the refugee inflows. Although it does not jeopardize our main results, our identification strategy cannot deal with general equilibrium effects. Furthermore, our proxies for healthcare services only account for the number of healthcare staff or hospital beds, but not for the quality of services.

6.2. Labor market, income, and time allocation

Labor market competition

Two potential pathways through which migration flows can deteriorate children's health outcomes are fiercer competition in the labor market and related income shocks (Baez, 2011; Maystadt et al., 2019). In Turkey, recent research reports a notable job loss for natives in informal sector due to Turkish workers’ being replaced with Syrian workers (Ceritoglu et al., 2017; Aksu et al., 2022).²⁶ The adverse employment effect is found to be more pronounced among disadvantaged groups, namely those who are less-educated, temporary-waged, and women in self-employment, and young workers in agricultural sector (Aksu et al., 2022). The job losses induced by migrant influx are likely to produce two opposite effects. The first effect is a reduction of income following possible job losses. Limited access to healthy nutrients, stemming from the households’ reduced purchasing power, may prevent children from receiving an appropriate diet and thereby cause them to experience growth retardation. The second effect is less obvious. Unemployment, in particular among women, may translate into more time spent with offsprings. Women might allocate more time to invest in their children's health capital through more frequent healthcare visits and/or provision of healthier nutrition. That second mechanism is likely to be particularly binding in countries where child care is poorly developed.

To test these possibilities, we first explore the effect of refugees on Turkish mother's labor market outcomes. Based on the sample of Turkish women's being in the labor force at the time of survey, the 2SLS model indicates a negative and a statistically significant coefficient (panel A, column (1) of Table 4).²⁷ Mothers living in provinces that received a large number of refugees are less likely to be employed as compared to mothers residing in less affected provinces. The effect is quite sizable. A 10% increase in refugees correspond to a fall by 5 percentage points (pp) in the probability to be employed. Given the mean value of employment (0.15), that is equivalent to a one-third reduction in employment. The negative employment effect differs between better-educated and less-educated women. Syrian refugees, being on average less skilled, might not be plausible substitutes for better-educated natives in the labor market (Tumen, 2018). If this is the case, then one can expect to find null or even positive employment effects on better-educated mothers and to find negative employment effects on less-educated mothers. We divide our sample based on the completion of compulsory years of schooling.²⁸ Our results support this conjecture (panel A, columns (2) and (3) of Table 4).²⁹ Mothers with less than 12 years of education are more likely to be out of the labor market, while mothers with 12 or more years of education are not affected by the presence of refugees. The negative employment effect is found to be more prevalent for the native mothers with less education, who can be assumed to work in low-pay jobs. It is therefore natural to revisit our main results for children with low-educated mothers. We re-estimate equation 1 by stratifying our sample by mother's education. Children, whose mother did not complete compulsory years of education, experience an increase in their HAZ: the estimated quasi-elasticity is 0.021 (see panel A, column (5) of Table 4).³⁰ Taken together, job loss experienced by less-educated mothers is associated with a puzzling improvement in their children's health. A plausible explanation is that the likely negative income effect is compensated by time reallocation toward their offsprings.

Table 4. Summary table II: labor markets, income and time allocation (2SLS results , with region-year FEs)

Notes: Data come from Turkish Demographic and Health Survey (TDHS). The full sample is for the 2003–2018 period, excluding 2012, at the 81-province level. Individual and household controls include child's sex, child's month of birth, mother's age, mother's age square, mother's education, whether the mother resides in rural area, total number of older siblings, being a female headed household, and wealth index (not in wealth index regression). Year, province, and region-year fixed effects are included. Individual survey weights are used in each specification. Standard errors, clustered at the 81-province level, are in parentheses. *** denotes statistical significance at the 1% level (p < 0.01), ** at the 5% level (p < 0.05), and * at the 10% level (p < 0.10), all for two-sided hypothesis tests.

Income effect

Health improvements are compatible with positive income induced by the presence of refugees. In other contexts, the presence of refugees has boosted the local economies (Taylor et al., 2016; Maystadt et al., 2019; Maystadt Duranton, 2019). For Turkey, it is less obvious (Aksu et al., 2022). To assess that possible explanation, we run equation 1 using the household wealth index. The estimated coefficients are far from being statistically significant (panel B, column (1) of Table 4).³¹ We cannot exclude the possibility that the null effect is driven by the importance of financial aid – either provided by central government or international organization. In any case, it does not seem sufficient to explain the observed health improvements.

Maternal time allocation

The crowding-out effect of refugees that left Turkish native women unemployed can increase the amount of time to be spent between mother and their offspring. This can have positive repercussions on children's health outcomes because mothers would be able to invest more in their children's health capital through several channels. Examples include, but are not limited to, increasing the frequency of received pre- or postnatal care and exerting more effort to provide a healthier diet, both of which are crucial for children's nutrition. To test this hypothesis, we explore the impact of refugees on the following outcomes: (1) mothers’ time spent with their children, (2) the number of antenatal care visits, and (3) receipt of postnatal care within two months.³² Overall, our findings suggest that the presence of refugees increases (i) the time spent by mothers with their offsprings, (ii) the number of antenatal care visits, and (iii) the likelihood of receiving postnatal care two months after birth. The effects summarized in Table 4 are sizeable (see panel B, column (4); panel C, columns (1) and (4)).³³ A 10% increase in the share of refugees leads to a 5 pp rise in the likelihood that the mother is spending time with her offspring, and a 9 pp increase in the number of antenatal care visits. These effects represent (i) a 6% increase in the maternal time spent with offspring and (ii) a 4% rise in the antenatal care visits at the mean values (0.75 for time spent and 2.02 for antenatal checks). For the probability of receiving postnatal care, our baseline specification, which allows for region-year FEs, reports a positive but statistically insignificant effect while the specification without region-specific trends shows a positive effect at 1% significance level (see panel B, column (4) of Appendix Table A.19). Therefore, we argue that antenatal and postnatal care increase because women have more free time due to maternal unemployment. Furthermore, we expect our results to be sensitive across educational level for these variables. We present our results in Appendix Table A.18 for maternal time spent with offspring at home (see panel C, column (12)), Appendix Table A.20 for antenatal care visits and receipt of postnatal care. Our findings on maternal time spent suggest that less educated women are more likely to spend time with their children at home (see panel C, column (12) of Appendix Table A.18). Nevertheless, we find no evidence of the refugee influx increases better educated mothers’ time spent with their offspring. Such an increase in mothers’ time allocation is likely to induce a rise in healthcare utilization which in turn can positively contribute to their children's growth patterns. As can be seen from panel A, column (12) of Appendix Table A.20, the effect is stronger for less educated mothers for antenatal care visits, while we report null results for better educated mothers. Regarding the likelihood of receiving postnatal care two months within birth, the positive effect is captured for less educated mother in the specification where we do not introduce region-trends and region-year fixed effects (see Panel B, column (10) of Appendix Table A.20). The positive effect, however, disappears once we add region-year FEs (see panel B, column (12)).

6.3. The quantity–quality trade-off

Conventional theories on family economics, especially on the intra-household resource allocation, suggest that parental decisions regarding the number of children and investments in children's human capital are interdependent. The theory posits that there exists a negative association between the number of children (i.e., quantity) and children-specific outcomes such as health and schooling (i.e., quality) (Becker Lewis, 1973; Becker Tomes, 1986).³⁴ Therefore, it is theoretically possible that the favorable effect of migration on children's growth indicators can be attributed to the quality–quantity trade-off: a decrease in the total number of children in the household can bring about an increase in the investment in health capital per child. To test whether such a trade-off exists, we run equation 1 where the dependent variable is the number of children per woman. Our investigation rejects that possible explanation (panel A, column (1) of Table 5).³⁵ Furthermore, Aygün et al. (2021) find no evidence of a significant effect on child mortality. Second, possible changes in household size or fertility decisions do not seem to have been followed by an improvement in early childhood investment. We indeed explore children's probability of being fully vaccinated against hepatitis B, measles, and tuberculosis.³⁶ The 2SLS estimates are all negative and statistically significant at any conventional level (see Table 5, panel A, column (3) and panel B, columns (1) and (4)). That is, the likelihood of being fully immunized decreases in the provinces receiving a greater share of Syrian refugees as compared to provinces with a lower share of refugees. The results confirm the findings of Erten et al. (2022) who show a notable decline in the vaccination outcomes induced by the immigration shock. Nevertheless, the relationship between vaccination and child height is a complex one.³⁷ In our analysis, we consider vaccination as a way to measure early child investment in health. It is possible the reduction in vaccination mitigates the positive effect of refugees on the height of natives’ children but at least, it cannot be a valid explanation for this positive effect on height. Our results might reflect a lower-bound estimate. If there was no detrimental effect on vaccination, our results may have been even stronger.

Table 5. Summary table III: quantity–quality trade-off (2SLS results , with region-year FEs)

Notes: Data come from Turkish Demographic and Health Survey (TDHS). The full sample is for the 2003–2018 period, excluding 2012, at the 81-province level. Individual and household controls include child's sex, child's month of birth, mother's age, mother's age square, mother's education, whether the mother resides in rural area, total number of older (not in total births per woman regressions), being a female headed household, and wealth index. Individual survey weights are used in each specification. Year, province, and region-year fixed effects are included. Individual survey weights are used in each specification. Standard errors, clustered at the 81-province level, are in parentheses. *** denotes statistical significance at the 1% level (p < 0.01), ** at the 5% level (p < 0.05), and * at the 10% level (p < 0.10), all for two-sided hypothesis tests.

7. Conclusions

We investigate the effect of massive immigration shock induced by the Syrian Civil War on Turkish children's anthropometric indicators. Dealing with the endogeneous settlement of Syrian refugees across 81 Turkish provinces, we find that the Turkish children's HAZ z-score is significantly higher in the provinces with a large share of refugees. This finding contradicts existing evidence which report a substantial decline in native children's growth indicators in the African contexts. Our additional results suggest that a plausible explanation is the increase in mothers’ time spent with their offsprings, paradoxically as a result of job losses mostly experienced by low-skilled women. We indeed confirm that children's health improvement is concentrated among households with low-educated mothers. We do not find evidence for alternative explanations related to the investment in healthcare resources, an income shock, or the quality–quantity trade-off in children's human capital investment. On the contrary, we confirm previous results such as a decrease in the per capita availability of healthcare resources (Aygün et al., 2021) and a lower probability of being fully vaccinated among children (Erten et al., 2022) in Turkey. These confirmed results do not explain the observed positive impact on child growth. If anything, it suggests some lower-bound estimates. If healthcare system and vaccination outcomes had not been negatively affected, the magnitude of the positive effect could have been even stronger.

It is important to be careful when interpreting the health improvements associated with the job loss experienced by mothers with low levels of education. While the loss of job among low-educated mothers is a source of concern, our paper sheds light on a crucial trade-off associated with parental participation into the labor markets in absence of strong child care services. From a policy perspective, maternal unemployment is certainly not a desirable channel to improve child health. However, it highlights the importance of child care provision, especially in a middle-income country like Turkey. When the provision of child care remains at sub-optimal levels, women are likely to encounter a trade-off between engaging in their career via formal employment and staying at home. Accessible and affordable child care is a potent tool to boost the equality of opportunity by promoting women's participation to labor market and child development in later stages of life (Blau Currie, 2006; Currie Almond, 2011; Ruhm Waldfogel, 2012). Despite a notable progress in the availability of preschool and child care facilities, enrollment rates remained at low levels (World Bank, 2015).³⁸ One possible reason for low enrollment can be the fact that public preschools target children aged 4–5. Population shock induced by the continuous arrival of refugees, on the other hand, is likely to reduce the enrollment rate when considering the negative relationship observed between childcare supply and population density in Turkish provinces (World Bank, 2015). Altogether, our research indirectly sheds light on the importance of extending child care at younger age. It is critical because extending child care can foster women's employment (by freeing up women's time) and hence economic development (Duflo, 2012).

Finally, this paper is not without limitations. Our analysis consists in assessing the change in health outcomes in refugee-hosting provinces compared to other provinces. We are therefore not able to estimate general equilibrium effects. For instance, we are not able to capture the consequences of a general decrease (due to budget re-allocations) or increase (due to international aid) of the central government budget dedicated to healthcare.There also exist a number of unexplored channels due to lack of data. First, improvements in the anthropometrics are parallel to the supply of satisfactory nutrition intake. Unfortunately, the TDHS does not ask sufficiently detailed nutritional intakes from the time of birth. Instead, it asks whether the child(ren) has been given some particular foods/beverages (e.g., milk, juice, eggs, bread, meat, and fish) within the last 24 h. One alternative outcome could be breastfeeding duration or exclusivity of breastfeeding to measure nutrition. However, there are a number of contributors to duration and exclusivity of nursing (e.g., mother's fertility preferences, sibling sex composition, and size) (Jayachandran Kuziemko, 2011; Chakravarty, 2015). Therefore, it is not straightforward to distinguish such effects from refugees’ arrival and it is beyond our paper's scope. Second, we do not investigate possible changes in healthcare provider. The information is available in the TDHS but cannot be exploited. Beyond the endogeneous nature of the provider choice, there is only a limited number of mothers with a private healthcare insurance. For instance, in 2013, less than 2% of mothers reported to have access to a private health insurance. Third, the time allocation (at home) question in the TDHS is far from providing an ideal measure when quantifying the changes in the time spent between mother and children.