Syrian Refugee Inflows, Health-Care Access, and Childhood Vaccination in Turkey

Abstract

1. Introduction

Armed conflicts and climate change forcibly displace millions of people worldwide every year. According to the UNHCR (2020), 82.4 million people experienced forced displacement in 2020, 26.4 million of whom live as refugees abroad. Developing countries host 86 percent of these refugees despite facing numerous resource constraints. Previous studies have documented that forced migration aggravates the disease burden in host countries because of the arrival of refugees with poor health (Montalvo and Reynal-Querol 2007; Baez 2011; Ibáñez, Rozo, and Urbina 2021). However, the increase in infectious diseases might also be driven by a decline in the availability of health-care resources, including health-care professionals or hospital beds per capita, which might induce a reduction in the utilization of preventative health-care services such as the vaccination of children and thereby increase the spread of infectious diseases.

This paper examines the impact of Syrian refugee inflows on the supply of health-care resources, the subsequent changes in childhood vaccination behavior, and the prevalence of infectious diseases among children in Turkey.¹UNHCR (2021) reports that Turkey currently hosts the highest number of refugees in the world, as 3.7 million Syrian refugees are settled in Turkey.² Turkey has made significant improvements in its health-care system since the early 2000s, leading to an increase in the ratio of fully vaccinated children between ages 0 and 2 from 54 percent in 2003 to 81 percent in 2008 (Hacettepe University Institute of Population Studies 2004, 2009). However, the rapid population increase in refugee-receiving areas may have reduced access to health-care services, including vaccination of children, in the absence of adequate investments in the supply of health-care professionals and infrastructure.

We exploit the differential arrival of refugees after the outbreak of the Syrian civil war in March 2011 across Turkish provinces as an exogenous shock to native health outcomes in host regions to investigate this question. However, it is possible that the settlement of refugees across regions was not random. Following Del Carpio and Wagner (2016) and Erten and Keskin (2021), we use a weighted average of the travel distance between Syrian governorates and Turkish regions/provinces as an instrument to predict the location choice of refugees in the first stage of an instrumental variable (IV) model. This empirical strategy allows us to account for potential endogeneity in the timing and level of refugee arrivals across provinces in Turkey. The inclusion of region-specific time trends further ensures that the results are not driven by pre-existing trends in outcomes observed in a particular region, including southeastern Turkey, which received large refugee inflows.

Our results show that native children living in regions that received large inflows of Syrian refugees experienced an increase in their risk of catching an infectious disease compared to children in less affected regions.³ In contrast, there is no evidence of significant changes in the incidences of noninfectious diseases, such as diabetes, cancer, or anemia. At the same time, we find that the number of health-care professionals and hospital beds per capita decreased in provinces that received large refugee inflows compared to provinces that received fewer refugees. Although the Turkish government invested heavily in the supply of doctors, nurses, and midwives in host regions, these investments did not fully offset the decline in per-person availability of health-care resources driven by the rapid increase in the local population.

We then examine whether the reductions in access to health-care providers affect the use of preventative health-care services such as the vaccination of children. Our findings reveal a significant decline in the vaccination of children in provinces that received a higher share of refugees compared to less affected provinces. These reductions in childhood vaccination might have further contributed to increases in the spread of vaccine-preventable diseases in affected regions.

Examining alternative channels through which refugee inflows might impact native children’s health outcomes, we find no evidence of a significant impact of refugee inflows on mothers’ time spent with children. Altogether, we conclude that although contact with potentially infected refugees might increase disease spread among natives, the migration-induced supply constraints in health-care access may have also worsened health outcomes in Turkey.

This paper makes several contributions to the existing literature. First, it is closely related to studies examining the public-health consequences of hosting large refugee populations in communities. For example, Ibáñez, Rozo, and Urbina (2021) find that higher refugee inflows from Venezuela to Colombia are associated with an increase in the incidence of vaccine-preventable diseases, including chickenpox and tuberculosis. Similarly, Baez (2011) reports increasing incidences of diarrhea and fever as well as increased childhood mortality in northwestern Tanzania after it was flooded by more than 500,000 refugees from Burundi and Rwanda. Both of these papers attribute the increase in the incidence of infectious diseases to direct transmission from the refugee population that arrives in the host country with a high disease prevalence from their war-torn countries.

This paper differs from these studies by examining the potential mechanisms driving the increased prevalence of infectious diseases in host regions, with a particular focus on the availability of health-care services. Specifically, it highlights the role of supply constraints stemming from overburdened health clinics, hospitals, and health-care providers as an unanticipated flow of migrants seek health-care services in the host country. Indeed, the pandemic-related service closures have led to substantial reductions in utilization of health-care services by children and young people even in developed countries (Bell et al. 2020; Chanchlani et al. 2020). UNICEF (2021) reports that the pandemic interrupted vaccination campaigns worldwide, resulting in 23 million children not receiving basic childhood vaccines.

One study closely related to ours is Aygün, Kirdar, and Tuncay (2021), who focus on the effects of refugee inflows on mortality rates for different age groups, but did not find any evidence of a significant impact. Although an examination of potential effects on these extreme outcomes is important, investigating the effects on the prevalence of infectious diseases in early childhood—regardless of whether they result in childhood mortality—is crucial for several reasons. As shown in previous studies, improved early-life health care and the completion of childhood vaccines have significant later life impacts on cognition, earnings, and health risks during adulthood. For instance, Bleakley (2003) examines the impact of eradicating malaria in the United States, Bharadwaj, Løken, and Neilson (2013) and Bütikofer, Løken, and Salvanes (2019) focus on the impacts of early-life health interventions, and Bloom, Canning, and Shenoy (2011) and Driessen et al. (2015) examine the effects of childhood vaccinations on school attendance and cognition. Thus, possible changes in vaccination rates and infectious disease prevalence in early childhood in response to refugee inflows may exert a wide range of indirect impacts on the host country in the long run.

Finally, this paper relates to a growing body of empirical work on the effects of Syrian refugee inflows on the labor markets (Del Carpio and Wagner 2016; Ceritoglu et al. 2017; Aksu, Erzan, and Kırdar 2022), housing markets (Tumen 2016; Balkan et al. 2018), educational systems (Tumen 2019), political outcomes (Altındağ and Kaushal 2021), domestic violence (Erten and Keskin 2021), and other outcomes in Turkey.⁴ Despite using a similar identification strategy, this paper is the first to document the effects of refugee inflows on infectious disease prevalence and childhood vaccination in Turkey.

Overall, our findings have several policy implications as we explain in more detail in the conclusion. First, increased investments in health-care resources in areas with greater concentration of refugees in host countries are crucial. Second, a related policy implication is the implementation of widespread vaccination campaigns that cover all vaccine-preventable diseases for refugee and native children. These campaigns should also include investments in new vaccination sites in host communities with large refugee populations. Designing early action plans on these fronts may help combat the potential spread of infectious diseases, which might have long-lasting effects on human capital accumulation of children.

In the Background section, we provide information on the Syrian refugee inflows to Turkey. Our Data section discusses the data and the identification strategy we employ in our analysis. The following section presents the empirical results, and the last section concludes the paper.

2. Background

Starting in March 2011, protests in Syria inspired by Arab Spring events called for the removal of President Bashar al-Assad. These protests were violently suppressed by Syrian troops (Slackman 2011). Following this suppression, civil unrest grew and expanded into an international conflict including various armed groups and countries. As a result, 6.7 million Syrians were internally displaced, and approximately 6.6 million people have fled Syria since 2011 (UNHCR 2021). Among those who have fled, 3.6 million currently live in Turkey and compose approximately 5 percent of Turkey’s population (UNHCR 2021). This significant increase in population took place over a short period of time: the number of refugees rose steadily from 145,000 at the end of 2012 to 1.5 million by the end of 2014 (see fig. S2.1 in the supplementary online appendix). Furthermore, the spatial distribution of refugees varied across provinces. As shown in fig. 1, in some provinces bordering Syria, refugee-to-native population ratio reached 61 percent by the end of 2014, while many other provinces experienced a far smaller inflow.

Figure 1.

Share of Syrian Refugees in the Turkish Population (in Percent), 2014

Source: Data on native population come from the Turkish Statistical Institute. Estimates on province-level refugee populations in 2014 come from the Ministry of Interior. The aggregate counts of the number of refugees are obtained from the United Nations High Commissioner of Refugees (UNHCR).

Note: The figure plots the ratio of the Syrian refugee population to the province native population in Turkey in 2014.

Open in new tabDownload slide

Syrian refugees predominantly originate from northwestern Syria, where the conflict began. The largest shares of refugee outflows originated from Aleppo (36 percent), Idleb (21 percent), al-Raqqah (11 percent), Lattakia (9 percent), and Hamah (8 percent) (DGMM 2013). In a survey implemented by the Directorate General of Migration Management of Turkey (DGMM), the majority of refugees (close to 80 percent) reported lower transportation costs as the main reason they chose to migrate into Turkey instead of other countries (DGMM 2013).

In October 2011, the Turkish government enacted a temporary protection regime, which provided a range of rights and services to Syrian refugees in Turkey, including free access to education, health services, social assistance, and freedom of movement within Turkey. In addition, Syrian refugees were assured of no forced return, which allowed the refugees to legally cross the Turkish border without the fear of deportation.

Since the beginning of their arrival, the Turkish government offered generous health-care coverage to the Syrian refugees upon their registration with the local authorities in each province. Registered Syrian refugees were immediately given temporary protection status, which allowed them to receive medical care as needed (ORSAM 2015). Refugees who lived in camps had free access to health-care facilities in the camp, but if more comprehensive treatment was needed, they were sent to nearby public hospitals.⁵ Refugees living outside the camps also had free access to health-care services.⁶ The hospitals in border provinces were reported to offer approximately 30 percent to 40 percent of their services to Syrian refugees, which creates capacity constraints and leads to delays in service delivery (ORSAM 2015).

The Turkish Ministry of Health in collaboration with UNICEF and WHO carried out several health campaigns targeting the refugee population. One of these campaigns was a polio mop-up vaccination campaign from 2013 to 2015.⁷ In the five cities with the highest refugee concentrations—namely, Mardin, Hatay, Kilis, Sirnak, and Sanliurfa—both Syrian and Turkish children aged 0–59 months were vaccinated against polio. In six other cities with Syrian refugee camps, only Syrian children were vaccinated against polio. In 2014, the campaign was extended to Syrian children in Istanbul, targeting all children in districts with large refugee populations (UNICEF 2014). This campaign was specifically designed to vaccinate children against polio and did not cover other diseases. In 2017, another vaccine campaign was launched, targeting refugee children under the age of five with the goal of completing the vaccines that they had missed (Turkish Medical Association 2019).

3. Data and Empirical Methodology

Data

We combine two types of data in our empirical analysis: (a) province-level data on refugee inflows to Turkey, number of health-care professionals, number of hospital beds, and trade volume with Syria and (b) individual-level data on childhood vaccination, disease prevalence, and socioeconomic outcomes.

Data on Refugee Inflows, Health-Care Professionals, Hospital Beds, and Trade

The information on the number of refugees in Turkish provinces was obtained from two sources. First, UNHCR (2022) provides data on the number of total Syrian refugees in Turkey on a monthly basis from 2011 to the present.⁸ Second, following Altindag, Bakis, and Rozo (2020), we combine the estimates on province-level refugee populations in 2014 released by the Ministry of Interior with the aggregate counts from the United Nations High Commissioner of Refugees (UNHCR) to estimate refugee numbers in each province.⁹ For all 81 provinces in Turkey, we obtain the share of refugees in a particular province by dividing the number of registered refugees by the total province population. Since refugees may move into other provinces or out of the country after their registration, the figures released by the UNHCR reflecting the number of registered refugees in each province are likely to include some degree of measurement error.

Figure 1 illustrates the geographical distribution of Syrian refugees using the share of refugee inflows in the province population across Turkey in 2014. The provinces with the highest refugee-to-native population ratios are Kilis (61 percent), Hatay (15 percent), and Sanliurfa (13 percent), which are all located on the Turkish–Syrian border. Provinces farther from the border have generally received fewer refugees relative to their population. Istanbul, Izmir, and Ankara are the three largest cities in Turkey and are centers for economic activity. Hence, they also have relatively high refugee ratios. The average share of refugee inflows in the native population across Turkey was approximately 2 percent in 2014.

We use two additional sources of data to construct our instrument. First, we use data on the population of each Syrian governorate in 2011, prior to the civil war, from the Syrian Central Bureau of Statistics. Second, we utilize Google Maps to calculate the travel distance between each governorate in Syria and each province in Turkey. Note that there are six border crossings between Turkey and Syria: two in Hatay and one each in Gaziantep, Kilis, Mardin, and Sanliurfa. Syrian refugees used different border crossings depending on their home governorates and their destination provinces in Turkey. We take the shortest travel path between two locations to calculate our distance measure. Due to the open-door policy in Turkey toward Syrian refugees, Syrian refugees had no reason to use illegal pathways to enter the country.

Moreover, we use data on the trade volume between each Syrian governorate and 81 Turkish provinces provided by the Turkish Statistical Agency to control for economic linkages between these regions.

Finally, we use data on the number of health professionals and hospital beds provided by the 2008–2016 Health Statistics Yearbooks. These data are published annually by the Turkish Ministry of Health. The data include aggregate figures from private and public sectors.

Data on Vaccination, Disease Prevalence, and Related Outcomes

We use two sources of individual-level data. First, we use four rounds of the Turkish Demographic and Health Surveys (TDHS) conducted in 2003, 2008, 2013, and 2018.¹⁰ Second, we use five rounds of the Turkish Health Surveys (THS) conducted every two years from 2008 to 2016. We will describe these data sources next and present summary statistics.

The TDHS data are nationally representative household surveys that provide information on demographics and health outcomes for women and children, including the vaccination of young children. More specifically, it asks women whether their children aged between 0 and 3 have been vaccinated against certain diseases. This data set also captures in which of the 81 provinces the respondent lives.

Panel A of table 1 presents descriptive statistics for the demographic characteristics and vaccination outcomes of children using the 2003–2018 TDHS data. Section I of Panel A shows that the average age of children was 1.2, and approximately half of them were girls. It also provides information on mothers’ characteristics, including their education, whether they speak Turkish, the type of area they live in (rural vs. urban), and age. The mothers average age was 28.3, approximately 28 percent of them live in a rural area, 5 percent do not speak Turkish (i.e., the interview was conducted in Kurdish, Arabic, or another language), and 26 percent have completed high school or above.

Section II of Panel A in table 1 provides summary statistics for vaccination outcomes. We observe that the total number of vaccines received by children increased from 5.72 in the pre-war period to 6.47 in the post-war period.¹¹ We examine whether a child completed the three doses of the hepatitis B and the diphtheria, pertussis, tetanus vaccines as well as if they received a single dose of the tuberculosis and measles vaccines. We focus on the completion of three doses for the hepatitis B and the diphtheria, pertussis, tetanus vaccines since having completed all doses of vaccines is considered a good measure of the strength of immunization and is therefore frequently used in the literature and by organizations such as the Global Alliance for Vaccines and Immunizations (GAVI) (Arevshatian et al. 2007; GAVI 2015). In Section II of Panel A in table 1, we observe slight increases in the national percentages of children who were fully vaccinated against individual diseases, with the exception of measles. As shown in table S2.1 in the supplementary online appendix, tuberculosis and measles require one dose of vaccine each, while hepatitis B and diphtheria, pertussis, and tetanus vaccines require three doses to be completed. In all surveys combined, 88 percent of children had completed the tuberculosis vaccine, nearly 64 percent of children had completed the hepatitis B vaccine, 69 percent had completed the diphtheria, pertussis, and tetanus vaccines, and 66 percent of children had completed the measles vaccine.¹²

The THS data are also nationally representative household surveys, providing rich information on health outcomes, including disease prevalence among children. In particular, the THS data provide information on whether children aged 0 to 6 have experienced different infectious and noninfectious diseases in the past six months. For our purposes, we confidentially obtained information on 26 regions from the THS data.¹³

Panel B of table 1 presents summary statistics for the disease prevalence among 0- to 6-year-old children using the 2008–2016 THS data. Section I indicates that the average age of children was 3 years, and approximately half of them were girls. It also provides information on the educational attainment of the household head. Section II provides summary statistics on the prevalence of diseases among children. In the pre-war period from 2008 to 2010, approximately 9 percent of children had experienced an infectious disease within the past six months, and this value increased to 10 percent in the post-war period from 2012 to 2016. Similarly, the prevalence of other infectious diseases, such as upper and lower respiratory diseases, exhibited an increase of approximately 1 to 2 percentage points. A similar marked increase at the national level was not observed for cancer, diabetes, or anemia, with diarrhea being the only exception.

Identification

We compare children’s outcomes in locations that are exposed to larger refugee inflows with those in locations that are less exposed to these inflows before and after the Syrian civil war began. However, the resettlement of refugees is a potentially endogenous decision. In particular, refugees may decide to settle in provinces with growing labor markets and better health infrastructure, which would result in a spurious negative correlation between refugee inflows and infectious disease prevalence. Hence, ordinary least squares (OLS) would underestimate the effects of refugee inflows on the health outcomes of children. However, it is also plausible that refugees settle in smaller cities with lower cost of living and potentially worse health infrastructure. In that case, the OLS estimates may be upward biased. Moreover, the settlement decisions of refugees may be affected by a range of other factors, including ethnic or religious networks, differences in local policies, and potential overcrowding in areas close to the border. In addition, the measurement error in province-level refugee inflows is likely to create attenuation bias in the OLS estimates.

We estimate a similar specification at the 81-province level using the four rounds of TDHS data over 2003–2018, comparing children’s vaccination outcomes in provinces more affected by the Syrian refugee inflows to less affected provinces before and after the civil war began. At the individual level, we control for the child’s age, gender, and month of birth, whether the mother lives in a rural area, whether her mother tongue is Turkish, and indicator variables for her educational attainment. At the province level, we control for the trade volume of each province with Syria and the baseline trade volume interacted with a time indicator (both in logs). Following Aygün, Kirdar, and Tuncay (2021), we include 5 region-specific linear trends and 5 region–year fixed effects in separate regressions to account for unobserved regional trends.¹⁵

Our empirical framework includes region/province fixed effects to account for any time-invariant heterogeneity across regions/provinces and year fixed effects to control for any macroeconomic shocks at the national level. Our instrument thus relies on variation within regions/provinces observed before and after the Syrian civil war began. In addition, the inclusion of trade volumes at the regional level accounts for the potential disruption of trade linkages between Turkey and Syria due to the Syrian civil war. Moreover, we use specifications that include the interaction of the baseline trade volume with time to control for the differential effects of baseline economic linkages between regions over time. The latter might be important if regions with initially stronger economic linkages with Syria face a greater or weaker change in their health outcomes for reasons that are unrelated to refugee inflows from Syria. Finally, any time-invariant characteristics of regions, such as distance to the border or initial economic development, are already controlled for using region fixed effects.

4. Effects of the Syrian Refugee Inflows

Infectious and Noninfectious Disease Prevalence

We begin by testing whether the prevalence of infectious diseases changed in response to Syrian refugee inflows, and we also examine whether the noninfectious disease prevalence changed as a placebo check. Table 2 provides the first-stage regression results. Column (1) regresses the share of Syrian refugee inflows in the region population on the distance instrument while controlling for region and year fixed effects. Column (2) controls for region-level trade volume with Syria, and Column (3) adds baseline trade volume with Syria interacted with the year fixed effects. Finally, Columns (4) and (5) separately account for 12 region-specific linear trends and 12 region–year fixed effects. The findings indicate a strong positive correlation, implying that the Turkish regions closer to more populated Syrian governorates received more refugee inflows. The F-statistics are far greater than 10, indicating that the instrument strongly predicts the share of refugees in the region population.

Table 3 presents the estimates of the impact of refugee inflows on disease prevalence in Turkey. The IV estimates reported in the first row of table 3 indicate that the children located in regions that received large refugee inflows experienced an increase in their probability of catching an infectious disease compared to those in less affected regions. The magnitude of the estimated coefficient implies that a 1-standard-deviation increase in the refugee share results in a 3.2-percentage-point (0.021 × 1.534) increase in the probability of having an infectious disease, corresponding to a 32 percent increase relative to the mean value.²⁰

Moreover, we observe similar increases in the incidences of highly infectious upper and lower respiratory diseases.²¹ The IV estimates in the second and third rows of table 3 indicate that a 1-standard-deviation increase in the refugee share leads to 1.9 and 1 percentage point increases in the incidences of upper respiratory and lower respiratory diseases, respectively. These values correspond to 5 and 10 percent increases relative to the mean values.

In contrast, we find no evidence that refugee inflows had a significant impact on the incidences of other diseases. The remaining IV estimates in table 3 reveal no evidence of consistently significant changes in the prevalence of cancer, diabetes, diarrhea, or anemia in regions that received large refugee inflows compared to less affected regions.²² This result is not surprising because cancer, diabetes, and anemia are noninfectious diseases, and while diarrhea is an infectious disease, it is not vaccine preventable, and its transmission mostly occurs through drinking water and food, with person-to-person transmission being much less common.²³

Table S2.3 provides a placebo check using pre-treatment data from 2008 to 2010 and assigning the 2016 values for the refugee share and the instrumental variables for each province to the 2010 data. The IV estimates indicate no evidence of a significant effect on any outcomes with the exception of anemia, which has a negative and significant coefficient estimate for the first specifications. However, this effect disappears once we control for 12 region–year fixed effects. Overall, these results indicate that pre-trends in infectious disease outcomes are unlikely to drive our estimates. Moreover, we include the time-varying GDP and unemployment rate measured at the 12-region level to control for changes in local economic conditions during our sample period. Table S2.4 shows that our estimates are robust to adding these additional control variables. Finally, our results are robust to excluding regions with the highest shares of refugee inflows and to excluding highly populated provinces from the sample, as shown in tables S2.5 and S2.6.

Health-Care Professionals and Hospital Beds

In this section we examine whether Syrian refugee inflows had a significant impact on the number of health-care professionals and hospital beds per capita at the province level. The per capita values are calculated by dividing the number of health-care professionals and hospital beds by the total population, including natives and refugees, in each province. Table 4 presents the estimates of the impact of refugee inflows on health-care resources in Turkey. Columns (1)–(5) provide the OLS estimates, while Columns (6)–(10) provide the IV estimates. The IV estimates in Columns (6)–(10) for the first four outcomes are all negative and significant. Thus, provinces that received a larger share of refugee inflows experienced a decrease in the numbers of doctors per capita, midwives per capita, and nurses per capita compared to less exposed provinces. However, we find no evidence of a significant impact on the number of hospital beds per capita. These results are similar to Aygün, Kirdar, and Tuncay (2021), who also report significant declines in the number of doctors per capita and midwives per capita, an imprecisely estimated but large decline in the number of nurses per capita, and no significant effects on the number of hospital beds per capita. The magnitude of the IV estimate in the first row of Column (10) with a full set of controls indicates that a 1-standard-deviation increase in the refugee share results in 0.06 fewer doctors per person (0.108 × 0.520), corresponding to a 4 percent decrease relative to the mean value.^24,25 Similarly, a 1-standard-deviation increase in the refugee share leads to 0.03 fewer midwives per person and 0.06 fewer nurses per person. These effect sizes correspond to a 3 percent decrease in the number of midwives per person and a 3 percent decrease in the number of nurses per person.²⁶

We also examine whether these effects are driven by the number of health professionals or the total population, and table S2.8 presents estimates of the effects of Syrian refugee inflows on the number of health professionals and hospital beds in each province. For ease of interpretation, we take the natural logarithm of the dependent variables. The IV estimates indicate that the numbers of doctors, midwives, nurses, and hospital beds increased in absolute terms in provinces that received a larger share of refugee inflows, suggesting that the government increased investments in these health resources. However, these increased investments were not sufficient to maintain constant ratios of these health resources to the province population since the province population is growing due to large refugee inflows. As a result, we observe a decrease in health resources in per capita terms in more affected regions as reported in table 4.

As a placebo check, in table S2.9, we restrict the sample to the pre-treatment period from 2008 to 2011 and assign the 2016 values of the refugee share and instrument variables for each province to 2011 data. The IV estimates reported in Columns (4)–(6) indicate no evidence of a significant pre-trend in these outcome variables observed prior to the arrival of Syrian refugees.²⁷

We conduct three robustness checks to test the sensitivity of our health-care resource estimates to alternative sample specifications. First, we include the 26 region GDP and unemployment rate, and separately add 26 region-specific linear trends and 26 region–year fixed effects in alternate specifications to account for time-varying changes in local economic conditions and unobserved regional trends over time. Table S2.10 reports these results, which are consistent with our main estimates. The only difference is that we observe a significant decrease in the number of hospital beds per capita in specifications that separately include 26 region-specific trends and 26 region–year fixed effects. Second, we exclude Gaziantep, Adiyaman, and Kilis, which constitute the Gaziantep NUTS2 region that received the largest number of Syrian refugees as a share of its population, to test whether our results are sensitive to their exclusion. The results reported in table S2.11 are consistent with our primary estimates. Finally, we exclude Istanbul, Ankara, and Izmir, which are the most populous cities with large labor markets and a good health infrastructure, to test whether potential sorting of refugees into these provinces affects our results. The results presented in table S2.12 are consistent with our main estimates.

Taken together, these results indicate that although the number of health professionals increased in regions that received larger shares of refugees, this increase was not sufficiently large to offset the decrease in the per capita availability of health professionals in more affected provinces. As a result, the host provinces that received a disproportionate inflow of refugees experienced a potential shortage of doctors, nurses, and midwives, as well as of hospital beds per person, relative to less affected regions.

Vaccination Outcomes

One of the potential consequences of declining access to health-care providers in regions that receive large refugee inflows is a reduction in preventative health-care behaviors, including the vaccination of children. In this section we proceed by testing the effects of Syrian refugee inflows on vaccination outcomes. Table 5 presents the estimates of the impact of refugee inflows on vaccination outcomes in Turkey. Columns (1)–(5) provide the OLS estimates, while Columns (6)–(10) provide the IV estimates. The IV estimates in Columns (6)–(10) are all negative and significant for all vaccination outcomes. Based on these results, children in provinces that received a larger share of Syrian refugees were less likely to be fully immunized than children in less affected provinces. The magnitude of the IV estimate in the first row of Column (6) implies that a 1-standard-deviation increase in the refugee share results in 0.43 fewer vaccines (0.067 × 6.345 = 0.43) received by children, corresponding to a 7 percent decline compared to the outcome mean. Similarly, a 1-standard-deviation increase in the refugee share results in a 7-percentage-point decline in the probability of receiving three doses of the hepatitis B vaccine, an 8-percentage-point decline in the probability of receiving the three doses of the diphtheria, pertussis, and tetanus vaccine, a 4-percentage-point decline in the probability of receiving the tuberculosis vaccine, and a 10-percentage-point decline in the probability of receiving the measles vaccine. These effect sizes correspond to 11 percent, 13 percent, 4 percent, and 16 percent declines compared to the mean values.²⁸