Education Interrupted: Enrollment, Attainment, and Dropout of Syrian Refugees in Jordan

Abstract The children affected by the Syrian conflict, including the large population of Syrian refugee children hosted in neighboring countries, are at risk of becoming a ‘lost generation’ due to interruptions in their schooling. This paper examines how educational outcomes of Syrian refugees in Jordan have evolved from pre-conflict to during conflict and displacement. We rely on nationally representative survey data from Jordan in 2016 and in-depth interviews with Syrian refugee youth. We use discrete-time hazard models and compare dropout pre-conflict, during the conflict, and during displacement for different stages of schooling. Syrian refugees in Jordan faced disrupted schooling in Syria due to the conflict, followed by several multidimensional supply- and demand-side barriers to education in Jordan. Yet ultimately enrollment rates, at least through 2016, have recovered to pre-conflict levels for basic education among the group of Syrians in Jordan, with important lessons for other countries struggling to protect refugee children’s education. Host countries’ policy response to refugee education plays a critical role in whether and for how long refugee children resume schooling after displacement.


Introduction
As of 2020, the global number of displaced persons reached a new high of 82.4 million (UNHCR, 2021). Displacement has become increasingly protracted, with particularly concerning implications for the well-being of children, who make up 42 per cent of the displaced (UNHCR, 2021). Among refugees, only 50 per cent have access to primary education (vs. 90% of non-refugee children), only 22 per cent attend lower secondary (vs. 84% of non-refugee adolescents) and only 1 per cent attend university (compared to 34% of youth globally) (UNHCR, 2016). However, microeconomic research on conflict and education has focused on populations that stayed in the conflict country. Studies generally show a negative impact of conflict on educational outcomes (Bertoni, Di Maio, Molini, & Nistic o, 2019;Diwakar, 2015;

Conceptual framework and hypotheses
To conceptualize the potential evolution of educational outcomes from pre-conflict to during conflict and displacement, we draw on conceptual frameworks for household decisionmaking among forced migrants (Abbasi-Shavazi, Mahmoudian, & Sadeghi, 2018) and adapt them to the issue of education. Displacement can be conceptualized in stages: home country/pre-flight (including conflict exposure), flight/transit, asylum/temporary settlement, and longer-term settlement, return or resettlement (Abbasi-Shavazi et al., 2018). For the case of Syrian refugees in Jordan, we distinguish four distinct stages: pre-conflict; conflict exposure in Syria; initial displacement; and protracted displacement. Drawing on the literature on the policy environment for refugees and challenges to refugee integration (Dryden-Peterson, 2016;Justino, 2014), in the latter phase, we emphasize the role of host-country education policies and practices as well as households' evolving socio-economic status. Per the conceptual framework (Figure 1), refugee education outcomes are influenced by preconflict policies and practices in the country of origin as well as households' socioeconomic status pre-conflict. Pre-conflict education trends are pertinent for considering the counterfactual of what education might have been achieved in the absence of conflict and displacement. These factors may continue to affect refugees' educational outcomes during conflict and displacement, e.g. through norms around education.
Conflict itself impacts the education system in the country of origin (affecting children who are school-age during conflict), potentially including experiences of internal displacement. Conflict can also alter the socio-economic status of households. Subsequently, during displacement, the education policies and practices of the host country are important for educational outcomes and may evolve substantially over time. Displacement also can alter refugee households' socio-economic status, with aid and economic policies in the host country shaping how socioeconomic status changes, particularly as displacement becomes more protracted.
Thus, studying refugees differs from considering those affected by conflict but still in their countries of origin for three main reasons: (1) while conflict's effects on education are ongoing for those who remain in the country of origin, refugees are only subject to these effects until they flee the country; (2) refugees experience different education policies and practices in their host country and; (3) refugees experience different socio-economic trajectories than those who remain in their country of origin. Refugees in a particular destination may also be a selected population from their origin country along a variety of dimensions; when considering refugees in a particular host country (such as Syrian refugees in Jordan) this does not create a problem in terms of estimating the evolution of outcomes from pre-conflict to during conflict and displacement among this particular refugee population, but it does limit generalizability to other populations.

Phases: Pre-conflict --> Conflict --> Initial displacement --> Protracted displacement -->
Pre-conflict education policies and practices in country of origin Pre-conflict education trends in country of origin Pre-conflict socio-economic status Impact of conflict on education system in country of origin (potentially including internal displacement) (school-age during conflict) Impact of conflict on socio-economic status Education policies and practices of host country (potentially evolving) Socio-economic status (including aid and economic policies) in host country (potentially evolving) Figure 1. Conceptual framework for refugee educational outcomes. Source: Authors' construction.
Drawing on the literature and our conceptual framework, we test the following hypotheses for Syrian refugees in Jordan: H1: School exit (dropout) will increase during periods of conflict and displacement compared to pre-conflict.
H2: There will be different patterns of school exit during conflict and initial displacement than during protracted displacement.
H3: Changes in dropout during conflict and displacement vary across the educational progression (across grades).

Refugee education and education policy in Jordan
In Jordan, the policy response to Syrian refugee education was rapid and based on comprehensive integration. Since April 2012, basic and secondary education were provided to Syrian refugees through the public Ministry of Education (MOE) system, free of charge (Culbertson et al., 2016). Within host communities, Syrian refugees may enroll in Jordanian public schools with enrollment priority given to Jordanian students (Culbertson et al., 2016). Double shift schools were created when there were insufficient places to include Syrians in a single shift with Jordanian students (Brussels II Conference, 2018). For Syrian refugee children in formal refugee camps, UNICEF provides education infrastructure and MOE provides Jordanian teachers (Culbertson et al., 2016).
However, Syrians can enroll in Jordanian schools only under certain conditions related to documentation and learning. Before 2016, children were required to have a service card (be registered as a refugee with UNHCR and with the Ministry of Interior) to enroll in school (Al Abed, 2017;Jordan Times, 2016). This requirement was waived temporarily in 2016 (Jordan Times, 2016) and permanently in 2017 (Al Abed, 2017). In 2017, the MOE also exempted Syrians from tuition and textbook fees (Brussels II Conference, 2018). However, refugee children who are three or more years older than the age that corresponds to their expected grade level cannot enroll in formal education (Education Sector Working Group, 2015).
Physical integration in the school system does not necessarily mean social integration or academic success (Ager & Strang, 2008;Dryden-Peterson, 2016). Globally and among Syrians, refugee students often do not have skills that match their expected grade for the age level (Brussels II Conference, 2018;Culbertson et al., 2016;Dryden-Peterson, 2016). Refugee children, including Syrians, are vulnerable to discrimination within school settings, including bullying from peers and prejudicial behavior from teachers (Culbertson et al., 2016;Dryden-Peterson, 2016).
Factors related to the context of their displacement may also affect refugees' demand for schooling. Poverty may hinder refugee households' ability to pay for books and transportation even when schooling itself is free (Culbertson et al., 2016). Poverty may also raise the demand for child labor, leading to the withdrawal of children from school (Justino, 2014). In Jordan, an estimated 86 per cent of Syrian refugees live below the poverty line (UNHCR, 2015), and child labor has been cited as a reason for non-enrollment (Education Sector Working Group, 2015). Among adolescent Syrian refugee girls, marriage is a factor that may reduce demand for schooling, with 18 per cent of refugee girls aged 15-19 married (Sieverding, Berri, & Abdulrahim, 2019). Furthermore, the returns to education in displacement may be low or perceived to be low (Justino, 2014), reducing the incentive to invest in education while in Jordan.

Quantitative data and analysis
We use the nationally representative 2016 Jordan Labor Market Panel Survey (JLMPS 2016) to analyze the education outcomes of Syrians 1 in Jordan. The JLMPS 2016 over-sampled areas Education interrupted: Syrian refugees in Jordan 1877 with a high proportion of non-Jordanians to be able to examine outcomes for Syrians. This sampling strategy is incorporated into the sample weights, which are used in our descriptive statistics (but not our multivariate models). The JLMPS includes a full educational history that allows us to assess the education experiences of Syrian refugees who were in Jordan in 2016 even when they were back in Syria. We also use the 2009 Syria Pan Arab Project for Family Health (PAPFAM) survey to illustrate enrollment patterns nationally in Syria pre-conflict. 2 Our quantitative multivariate analysis exploits the JLMPS 2016 retrospective data to assess the evolution of progression and school exit from pre-conflict to during conflict and displacement. We further contextualize these outcomes by descriptively examining measures such as enrollment. We calculate basic and secondary net enrollment ratios (NERs) for years before and since arrival in Jordan based on retrospective educational and residential mobility data (see the retrospective questions section in the Supplementary Material). 3 The level or grades of school completed is the key outcome underlying our multivariate analyses. Since our outcome of grades completed is right censored for those currently enrolled, survival analysis methods are required. We, therefore, structured the dataset as individual-grade observations, including grade 'zero' for entry. If an individual has never attended school, their highest year in school and exit was set to grade zero. If an individual had previously (but not currently) attended school, their grade of exit was the highest grade they completed. Those currently in school are right-censored (have no exit), yet we know how many grades they have persisted through schooling thus far and this information is incorporated into our estimates.
We estimate a discrete time hazards model for the grade (years of school completed). The hazard function, h ig, describes the probability of an individual i exiting school (the event T g ) in a particular grade, g, if he or she has not already left (Jenkins, 1995): We use a complementary log-log model, which is a proportional hazards model where a covariate proportionately raises (or lowers) the hazard of exiting school (Jenkins, 1995): The term h j denotes the baseline hazard, the probability of exiting the school at each grade level g j (for the reference individual, when covariates X ik are included). The estimated coefficients, when exponentiated, characterize how the hazard ratio changes with a one-unit increase in the covariate. Hazard ratios >1 mean a higher hazard of school exit, <1 mean a lower hazard of exit.
Since we are interested in how educational outcomes changed before and during the conflict, as well as in displacement, we limit the sample in our multivariate analyses to individuals aged 6-24 as of the end of December of the survey year (a sample of entry cohorts, as is ideal for survival analysis). We use this cut-off since children enter school (on time) in Jordan the year when they are aged six by the end of December (Ministry of Education, 2018). We further limit our analyses to the years 2006-2016, since going further back in time gives us a diminishing sample size. With these samples, we can compare those who completed their course of schooling in Syria to those whose schooling was disrupted by the conflict and those whose schooling started in Jordan. In the JLMPS 2016, our sample for multivariate analysis is 1,111 Syrians.
We are interested in how educational progression has evolved since the Syrian conflict, which started in 2011. We, therefore, identify the calendar year an individual was in each grade. 4 We initially estimated our models with only the baseline hazards for each grade in school (e.g. grade 3). We then aggregated some of the grades with similar baseline hazards (grades 1-5, 6-8, and 9-12; 5 note that basic education in grades 1-10 in Jordan and secondary is grades 11-12).
Thereafter, grades were also interacted with sex, denoted s, to account for differential patterns of dropout by sex. We then estimated calendar year (e.g. 2014) differences, denoted by c j , as periods of exposure to conflict and displacement. We refer to this model, shown in equation (3) as specification 1.
We are particularly interested in the coefficients c j on different calendar years, relative to the reference year of 2010 (immediately pre-conflict). The 2011 and onward estimates test H1, whether school exit increased during conflict and displacement. Examining the pre-2010 coefficients establishes whether there were changes over time in exit pre-conflict. Subsequently, we aggregated the calendar years into three periods: 2006-2010 (pre-conflict, the reference period), 2011-2013 (conflict and initial displacement), and 2014-2016 (displacement, at which point the majority of refugees were in Jordan). We refer to this model, shown in equation (4) as specification 2.
We are particularly interested in c 1 , the coefficient for the period of initial conflict and displacement and c 2 , the coefficient for the period of protracted displacement in Jordan. These coefficients can be compared to test whether there were different patterns of exit during conflict and initial displacement than during protracted displacement (H2).
We also interact periods of time and stages of schooling to see whether different grades might have had varying changes in dropout over time. We refer to this model, shown in equation (5) as specification 3.
h j, 2011À13 and h j, 2014À16 test H3, whether the changes during conflict and displacement may vary by grade (j ¼ 1-5, 6-8, and 9-12 as compared to entry).
In the models, we also control for the mother's and father's education, and the number and composition of siblings interacted with sex. Table A1 in the Appendix presents descriptive statistics on these time-invariant covariates for Syrians aged 6-24. Table A2 in the Appendix illustrates our time-varying data structure.
Conflict and displacement exposure affected different ages and stages of schooling across cohorts. While we can directly estimate differences over time, these differences are not necessarily caused by conflict. There may, for example, have been rising enrollment and falling dropout over time pre-conflict (we test for this using c j for individual calendar year effects pre-conflict). The cohorts entering and progressing through school may have changing composition over time in unobservable ways, and these omitted variables can bias estimates of change over time. Furthermore, the students still enrolled in school could experience varying degrees of selection over time. For instance, if conflict causes students who would have only completed a basic education to drop out early, the group of students who remains or returns to school in displacement would then be selected.
Education interrupted: Syrian refugees in Jordan 1879

Qualitative data and analysis
The qualitative data consists of 71 in-depth interviews with Syrian refugee youth aged 15-29 residing in Mafraq and Amman governorates. Half of the respondents were men (36) and half women (35). The majority arrived in Jordan between 2012 and 2013. Only 15 were still in school at the time of the interview, but 23 had previously attended school in Jordan; all (previously) attended public schools. Respondents were primarily attending or had attended the upper basic level (grades 5-10), with few having reached secondary. Interviews were transcribed, coded, and analyzed both thematically and in terms of life trajectories. From the thematic analysis, we present results related to youths' educational experiences, including barriers to and facilitators of education in Jordan, and school experiences in Jordan (including academic experience, school environment, and interpersonal relationships). In the life trajectory analysis, we focused on the timing of leaving school relative to the conflict, and re-enrollment and dropout decisions after arrival in Jordan. See the Supplementary Material for details on the qualitative methods. The lower enrollment rates of Syrians in Jordan in 2016 compared to nationally in Syria in 2009 may be due to conflict and displacement but may also be a continuation of pre-existing trends among this selected and disadvantaged sub-population. 6 We, therefore, focus hereafter solely on the group of Syrian refugees who entered Jordan. When comparing educational outcomes pre-conflict to during conflict and displacement for this specific population of Syrians, selection into Jordan is not relevant since the whole population selected into Jordan (although endogeneity and generalizing to other groups of Syrians, not in Jordan remain issues).

Enrollment among Syrian youth in Jordan
With the JLMPS data, we can also look at net enrollment rates (NERs) relative to the year of arrival in Jordan. The NER for a particular school level is calculated by dividing the number of pupils enrolled in education in that level who are of age for that level by the population for the same age group. In Figure 3, year zero is the year of arrival in Jordan. Years one and two signify time post-arrival in Jordan, while years -1 to -5 signify time pre-arrival (in Syria). The NERs for basic education fell somewhat in the two years before arrival (during the conflict in Syria), then stabilized or increased slightly, indicating the successful transition of students at this level into the Jordanian school system. NERs for secondary are very low post-arrival, due to delays in progression and possibly non-re-entry among older students. 7

School exit timing among Syrian youth
To statistically test for differences during conflict and displacement as compared to pre-conflict, we estimate the hazard models for the sample of Syrians in Jordan in 2016 that include single years of calendar time (specification 1), the hazard ratios for which are shown in Table 1. 8 Key to note is that the hazard of exit was rising pre-conflict, as it was significantly lower from 2006 to 2009 compared to the reference year of 2010 (immediately before the conflict). The drought and worsening economic conditions in Syria likely drove this trend (United Nations, 2010). Although the counterfactual of what would have happened in Syria without the conflict is difficult to know, the trend pre-conflict was notably not less dropouts. The hazard of exit was not significantly different in 2011-2013 than in 2010 for the Syrians who ended up in Jordan. In specification 2, we group the years into the pre-conflict period (2006)(2007)(2008)(2009)(2010), conflict and initial displacement (2011)(2012)(2013), and protracted displacement (2014)(2015)(2016). Compared to the reference 2006-2010 period, the hazard of exit is higher in 2011-2013 (during conflict and initial displacement) but insignificantly so. The hazard of exit is lower, significantly so, in 2014-2016, when Syrians were in Jordan. 9 This finding is counter to H1; rather than an increase in school exit during conflict and displacement, outcomes during conflict and initial displacement were similar to pre-conflict. Exit actually decreased in protracted displacement. Our results confirm H2; there are different patterns of exit during conflict and initial displacement compared to protracted displacement since the coefficients for 2011-2013 and 2014-2016 are significantly different.
There may be differential changes in school entry or exit for students in different grades in school during conflict and displacement (H3). We therefore fully interact with the grouped calendar years (2011-2013 and 2014-2016 vs. 2006-2010) and grade segments (entry, grades 1-5, grades 6-8, and grades 9-12) in specification 3, summarized in There was a significantly higher hazard of dropping out in grades 1-5 in 2011-2013 compared to 2006-2010 or 2014-2016 (but 2006-2010 and 2014-2016 were not significantly different). Students who were in the early stages of basic education when the conflict in Syria began and during the period when most refugees arrived in Jordan experienced higher hazards of dropout than those who were in the same levels of education during the periods immediately before the conflict or after arrival in Jordan.

Education interrupted: Syrian refugees in Jordan 1883
For grades 6-8, there was a not a significantly higher hazard of dropping out in 2011-2013 compared to 2006-2010. However, the hazards in 2014-2016 for grades 6-8 were significantly lower than in both previous periods. This means that children in grades 6-8 in 2014-2016, when nearly all were already in Jordan, were more likely to persist in school than those in grades 6-8 during the conflict or in Syria pre-conflict. The Syrian youth who made it to grades 9-12 had the lowest hazards of dropout in 2014-2016, significantly so compared to the other two periods. However, this group would be particularly selected, as relatively fewer of them would have made it through the preceding years of basic. Overall, the results support H3, that there were distinct changes in exit during conflict and displacement across grades. Children in grades 1-5, which had high enrollment rates pre-conflict, during conflict (the 2011-2013 period) did drop out at higher rates, but progression recovered in protracted displacement. Children in later grades (6-8 and 9-12) during protracted displacement had better outcomes than pre-conflict.

Supply-side barriers to education in Jordan
Syrian refugees who re-enrolled or considered re-enrolling in school after their arrival to Jordan faced several supply-and demand-side barriers. On the supply side, the main themes that arose from respondents' discussions were the registration process, academic difficulties, and discrimination. Most respondents reported that the school enrollment process in Jordan was straightforward and smooth. Those for whom it was not faced two types of challenges: lack of needed documentation and resistance from the school administration. At the time most of the respondents arrived in Jordan, in 2012 and 2013, school enrollment policies were changing and the documentation requirements that have since been waived were still in place. Some respondents were missing documentation of refugee status in Jordan, family books, or school records from Syria. However, in all the cases that came up within the qualitative sample, respondents were eventually able to overcome the lack of documentation, sometimes with help from UNHCR.
'Once we obtained the UNHCR registration card, we were then able to enroll in school despite the missing family book.' Young man, 17 years old, East Amman, attending school in Jordan.
Respondents' experiences were thus consistent with the quantitative results in indicating that those youth who wanted to re-enter school after arrival in Jordan were in fact able to do so from the perspective of administrative requirements.
A number of respondents reported school principals not wanting to admit them due to lack of space, or, in a few cases, implied bias against Syrian students. In some cases, respondents or their families were able to overcome this resistance through persistence. A few others went to the MOE for assistance and reported that the MOE obliged the school to enroll them.  Table 1, specification 3, for the hazard ratios. Models include controls for sex, parents' education, and siblings. Results jittered for ease of visualization.
'When I came from Syria, I brought with me my certificate of birth, but the school principal did not agree to enroll me because there was no policy to enroll Syrians in schools, she said. We tried several times to convince her, but she would not agree to it. So we went to the Ministry of Education. I complained to the Ministry, who contacted the school principal to tell her to enroll me and every other Syrian student applying.' Young woman, 16 years old, Mafraq, attending school in Jordan.
The combination of school interruptions, differences in the Jordanian and Syrian school systems, timing of arrival relative to the start of the school year, and lack of places contributed to the fact that some respondents were placed into a grade level below that expected for their age. 12 For several, the age-for-grade gap was the main factor contributing to the decision not to enroll in school after arrival in Jordan.
'I wanted to go back to school, but they wanted to place me in 7 th grade while I should have been in 11 th grade. That is a 5-year difference. So I did not want to enroll.' Young woman, 18 years old, East Amman, never attended school in Jordan.
Other respondents mentioned the difficulty of adapting to a school in a new environment, with a new curriculum and classmates they did not know.
'At first school was hard and I wanted to stop, I didn't want to study. I didn't know anything and the girls were younger than I was.' Young woman, 22 years old, Mafraq, previously attended school in Jordan.
Curriculum difficulty was a particular concern for the few respondents who had reached secondary or had already been at the secondary school level when they arrived in Jordan.
Yet more so than instructional factors, discrimination and interpersonal challenges emerged as an important theme for many respondents who had attended school in Jordan. Respondents who had positive relationships with their teachers described them as cooperative, showing interest in their futures, and not tolerating differentiation between Jordanians and Syrians.
'Teachers are not discriminating. They do not mention nationality; they encourage us Syrians and Jordanians to be one and not separated.' Young man, 15 years old, East Amman, attending school in Jordan.
Those who reported a negative relationship with teachers, by contrast, said that their teachers were verbally and physically abusive, at times to the point that this was a factor motivating dropout.
'I dropped out … they used to hit me for no obvious reason … last time they summoned me to the administration where someone slapped me on the face, they accused me of stealing!' Young man, 15 years old, Mafraq, previously attended school in Jordan.
In terms of their school peers, the majority of respondents formed friendships with their Syrian peers, especially among girls. A few also managed to become good friends with Jordanians. Yet many respondents reported being bullied by their Jordanian peers and feeling unwelcomed, so they avoided mixing with Jordanian students.
'Many girls used to tell us [Syrians] that Jordan is for them Jordanian nationals. Others on the contrary they were very helpful and did not allow others at school to bully us.' Young woman, 18 years old, East Amman, previously attended school in Jordan.

Demand side barriers to education in Jordan
Syrian households' difficult economic situation was the main demand-side theme that arose as a challenge to school enrollment in Jordan. Although a few respondents said that their parents could not afford school fees and supplies, more so than direct costs of education, the need to work kept male respondents out of school, as many were supporting their families. Young men mentioned specifically that they worked so they could help to pay for the family's accommodation.
'I did not inquire about education in Jordan. We had newly rented a house and there was no one but me to work and pay.' Young man, 23 years old, never attended school in Jordan.
For young women, marriage, which was also commonly linked to difficult economic circumstances, was an impediment to schooling as some dropped out of school after getting engaged. Married young women could not be at school and manage a household, or their families did not consent to their continuing school. A few other female respondents were out of school to take care of family members, or because their parents would not let them attend due to perceptions of insecurity in Jordan.
Besides financial challenges, a few youths mentioned that there was no value in education in Jordan, either because they could not use their education in the Jordanian labor market, or due to a general loss of hope in the future.
'My future was lost with Syria's destruction and the events. I was putting high hopes on having an education and later a decent job. When I stopped my education, my aspirations vanished.' Young man, 23 years old, Mafraq, never attended school in Jordan.
Despite the difficult circumstances, some youth perceived education as an achievement in one's life and wanted to continue or return to education. In this respect, parental encouragement (and sometimes requirement) for youth to stay in school was an important facilitator of education, as well as individual motivation.
'All of my siblings are educated, my mom urges us to study and repeats how education is very important.' Young man, 17 years old, Mafraq, attending school in Jordan.

Discussion and conclusions
For the 6.7 million Syrians displaced by conflict, there have been concerns of a 'lost generation' of children missing out on school (Brussels II Conference, 2018;UNHCR, 2016UNHCR, , 2021. As refugees' displacement lengthens, the efforts of host communities and the international community to ensure their education are increasingly important. While previous research demonstrates that conflict usually (but not always) harms educational outcomes for those who stayed in their country of origin (e.g. Fiala, 2015;Justino, 2013), there has been little research on the impact of displacement on refugees' educational outcomes.
Our paper contributes a conceptual framework for understanding refugees' educational outcomes during conflict and displacement and some of the first evidence on refugee outcomes during the conflict, initial displacement, and protracted displacement in a developing country. We hypothesized that school exit would increase during conflict and initial displacement as well as protracted displacement (H1). The analysis demonstrates that, contrary to our hypothesis, refugees can experience comparable or even better educational outcomes during protracted displacement compared to pre-conflict.
The results of the quantitative models show that there was an accelerated dropout at the basic education level during the peak conflict years (2011)(2012)(2013) when most refugee children were still in Syria or in the process of moving to Jordan. However, persistence in school among the Syrian refugees in Jordan in protracted displacement (2014-2016) was comparable to or higher than among this population pre-conflict. The evidence supports H2, that there are differences in school exit during initial conflict and displacement vs. protracted displacement. This finding must nevertheless be placed in the context of the Syrian refugees in Jordan coming from an educationally disadvantaged population within Syria; net enrollment rates at the basic level remain around 80 per cent. Enrollment in secondary schooling is low. However, those select youth who transitioned to secondary school persist at higher levels than pre-conflict. These Education interrupted: Syrian refugees in Jordan 1887 results confirm H3 (that changes in exit during conflict and displacement vary by grade) and are consistent with other research that finds that the age of exposure to conflict matters for educational impacts, particularly relative to compulsory schooling (Bertoni et al., 2019;Fergusson et al., 2020).
The recovery in Syrians' enrollments and progression highlights how a clear policy mandate for including refugees in local education systems combined with a Ministry of Education that actively supports that mandate can substantially improve access to education. A comparison of Syrian refugees' outcomes in Jordan to those in Lebanon helps illustrate the importance of policy responses. In Lebanon, it was not until 2014 that the Ministry of Education and Higher Education launched a policy framework for refugee education (Buckner, Spencer, & Cha, 2017). In 2015, Lebanon had only 39 per cent of children aged 5-17 enrolled in formal schools, compared to 66 per cent in Jordan (Brussels Conference Education Report, 2017). As of 2017/ 18, only 4 per cent of Syrian school-age children had never entered school in Jordan, and enrollment rates were nearly 99 per cent for ages 6-11 and had risen from 49 per cent at age 14 in 2014 to 68 per cent by 2017/18 (Tiltnes, Zhang, & Pedersen, 2019). In Lebanon, conditions had improved by 2017 with 70 per cent of those ages 6-14 enrolled in school (up from 52% in 2014), but only 13 per cent of those aged 12-14 were enrolled (UNHCR, UNICEF, & WFP, 2017). Although there were some differences in the education profiles of the older generation of Syrians who fled to Lebanon and Jordan, these differences were small relative to the enrollment differences observed in displacement, 13 and thus signal an important role for host-country policy (Sieverding & Calder on-Mej ıa, 2020;Verme et al., 2016).
That countries can, with a concerted effort, ensure refugees have similar education outcomes to pre-conflict is an important policy lesson for the countries hosting displaced children and youth. Although historical cases from developed countries (Bauer et al., 2013;Becker et al., 2020) have demonstrated refugees may invest relatively more in education due to its portability, this is the first evidence that developing country contexts can also support and potentially even improve refugee educational outcomes.
These findings also point towards important areas for future research. Although we have, unusually, nationally representative data for a large number of refugees in displacement and detailed retrospective data, retrospective data may have substantial measurement errors (Assaad, Krafft, & Yassin, 2018). Contemporaneous and frequent panel data or administrative records on refugees would be helpful for future research. Our retrospective data also lack key details on conflict and displacement experiences in Syria and arriving in Jordan. While our estimates show changes over time in dropout, they are not necessarily causal impacts of conflict and displacement, as we lack additional variation to identify impacts (such as location-specific exposure to conflict). Designing surveys to better capture conflict and displacement experiences and facilitate stronger identification strategies is an important area for future work. While our results do generalize to those Syrian refugees in Jordan, since refugees are a select group, our findings would not necessarily generalize to other groups of Syrians.
The qualitative data are not representative, although helpful for understanding drivers. Our findings concur with other studies (Education Sector Working Group, 2015;Salem, 2018) that interpersonal aspects of schooling, and particularly bullying by peers, affect dropout decisions among refugee adolescents. Interventions that support teachers' and native students' diversity awareness and perspective taking have shown promise in evaluations in Turkey (Alan, Baysan, Gumren, & Kubilay, 2021;Tumen, Vlassopoulos, & Wahba, 2021) and should be tested in Jordan and at scale. Testing academic support programs for refugees is another important area for future research. For those who did drop out and cannot return to formal schooling, nonformal education may be an important intervention that merits research.
There are varying and context-specific results in the literature in terms of the impacts of conflict on educational outcomes by gender (Bertoni et al., 2019;Buvini c et al., 2014;Diwakar, 2015;Shemyakina, 2011). For adolescent boys, we found the need to work was a key reason for being out of school. Particularly in the face of financial pressures to work, a negative school environment can serve as an extra push to leave school. For girls, the demand-side factors limiting school enrollment were more mixed but included marriage and family responsibilities. Future research is needed on educational supports, such as cash transfers (e.g. De Hoop, Morey, & Seidenfeld, 2019), school feeding programs (which were effective during the conflict in Mali, see Aurino, Tranchant, Sekou Diallo, & Gelli, 2019), as well as gendered interventions to prevent early marriage and child labor. Notes 1. In this paper we treat "Syrians" as synonymous with "Syrian refugees." The percentage of Syrians in Jordan that are currently registered as a refugee and arrived in Jordan in 2011 or later or left a previous residence in 2011 or later due to violence, persecution, or lack of security is 93 per cent (Krafft, Sieverding, Salemi, & Keo, 2019). 2. For more information on the PAPFAM 2009, see PAPFAM (2011). For more information on the JLMPS 2016, see . 3. Although retrospective questions may be subject to bias, among the Jordanians of age for our analysis sample who ever attended school, 75 per cent report starting grade one of basic the year they turned six and 96 per cent report starting grade one of basic the year they turned six or seven, suggesting relatively accurate recall. 4. See the retrospective questions section in the supplementary online material for the exact questions and details. 5. Too few Syrians in Jordan progressed to higher education to analyze beyond grade 12. 6. In addition to refugees being from areas that were particularly affected by conflict, the adult Syrian refugee population in Jordan is less educated than the national population in Syria pre-conflict (Sieverding & Calder on-Mej ıa, 2020), which may affect demand for education among refugees due to intergenerational persistence in educational attainment. Other differences between the Syrian refugees who fled to Jordan and the national population of Syrians (which existed pre-conflict) may also be pertinent to educational outcomes, including higher poverty, higher fertility, and earlier marriage  Sieverding & Calder on-Mej ıa, 2020; Verme et al., 2016). Other research suggests that differences between adolescent Syrian refugees in Jordan and Jordanians in school enrollment are almost entirely explained by differences in socio-economic status between the two groups (Krafft, Assaad, & Pastoor, 2021). 7. When including delayed students in gross enrollment ratios, secondary enrollment rates have come closer to recovering . 8. Figure A1, in the supplementary online material, presents a descriptive analysis of school persistence and exit by cohort, the results of which are consistent with the multivariate findings. 9. Estimating a model with time-varying country of residence corroborates that the hazard of exit was (insignificantly) lower in years when individuals were in Jordan compared to Syria. 10. We tested whether there were differential effects of the time trends and time trends interacted with grade by sex; there were not significant differences. 11. Since the model includes both main effects and interactions between time and grade segment, the marginal effects (predicted hazards, shown in the figure) of different combinations were tested for equivalence. 12. See Sieverding et. al (2018) for an examination of interruptions and re-enrollment, as well as age-for-grade and grade repetition. Syrians in Jordan were slightly older than their expected grade at some points in their trajectory, especially during latter basic and secondary, but did not repeat grades. A substantial fraction of Syrians in Jordan experienced interruptions of six months or more in their schooling, but often re-enrolled successfully after arrival to Jordan. 13. For instance, estimates among Syrian refugees aged five and older in 2014 show 7 per cent of Syrian refugees in Lebanon had no education compared to 11 per cent of Syrian refugees in Jordan; 22 per cent in Lebanon and 20 per cent in Syria had first stage primary; 54 per cent in Lebanon and 52 per cent in Jordan had second stage primary, 8 per cent in Lebanon and 11 per cent in Jordan had secondary attainment, and 5 per cent in Lebanon and Jordan had university or above attainment (Verme et al., 2016).

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
Research supported by a grant from the Ford Foundation on 'Youth and Vulnerability in the Middle East' to the Economic Research Forum.