Cohort at Risk: Long-Term Consequences of Conflict for Child School Achievement

We investigate the long-term effects of households' exposure to violent conflict on children's educational attainment in primary school, studying cognitive and non-cognitive skills as possible causal channels. Our identification strategy exploits the locality-level variation in the intensity of the Israeli-Palestinian conflict in the West Bank during the Second Intifada (2000–2005). We show that an increase in family experience of conflict has large negative long-term effects on the educational attainment of children measured by grade point averages. Impaired non-cognitive rather than cognitive skills are identified as channels through which exposure affects children's educational achievement.


Introduction
Violent conflicts can have enormous social and economic consequences for the affected civilian population. These consequences can be especially dramatic for children. A growing body of research from different countries provides robust evidence of the short-run adverse effects of conflicts on a large array of children's outcomes, including, for instance, health (Minoiu and Shemyakina, 2014;Akresh et al., 2012), psychological well-being (Blattman and Annan, 2010), child labor (Di Maio and Nandi, 2013), and education (Shemyakina, 2011;Brück et al., 2014;Di Maio and Nandi, 2013). However, exposure to conflict may also affect outcomes of individuals later in life, particularly if this happens during early childhood which may be a sensitive or even critical period for the development of important skills in life (Conti and Heckman, 2012).
This paper contributes to the literature in two ways: first, we investigate the long-term effects of households' exposure to violent conflict on children's educational attainment. So far, only few authors have investigated such long-term effects, and the results remain inconclusive and contextdependent. For instance, whereas León (2012) and Akresh and De Walque (2008) find that exposure to conflict in early childhood reduces completed years of schooling by 0.3 to 0.5 years, Arcand and Wouabe (2009) find no such effect. Second, and perhaps more importantly, we provide detailed analyses of the causal channels through which exposure to conflict may affect educational outcomes -which is critical for designing and implementing effective policies that protect children from the negative conflict effects (Justino, 2011). We are not aware of any papers in the related literature that have looked at skill development as a potential mechanism. We fill this gap by explicitly analyzing cognitive skills (IQ) and non-cognitive skills (personality and behavioral problems) as possible causal mechanisms. In some way, we provide a mirror image of the work by Heckman and coauthors on the long-term beneficial effects of preschool programs, see e.g. Heckman et al. (2013). More generally, by looking at how exposure to conflict in the early stages of a child's development affects her skills and outcomes later on in life, this paper contributes to our knowledge on how the early childhood environment affects long-term outcomes (see, for instance, Currie (2001); Heckman (2000); Krueger and Whitmore (2001); Gould et al. (2011)).
Specifically, our study is set in the context of the Israeli-Palestinian conflict. We study intelligence, non-cognitive skills and school grades in West Bank students seven years after the end of the Second Intifada, and we compare children whose families were directly affected by this conflict with those who were not. The Second Intifada took place between September 2000 and February 2005. During that period, a re-surge in the intensity of violence between Palestinian organizations and Israeli Defense Forces (IDF) resulted in more than 38,000 casualties (dead and wounded). It claimed the life of about 4,000 Palestinian and 1,000 Israeli civilians (18% of which were children), with the number of fatalities varying markedly across localities and over time. 1 For example, 2002 saw a peak in the number of deaths in the context of a large-scale Israeli military offensive, Operation Defensive Shield, in response to Palestinian suicide attacks. This military operation was concentrated primarily among some very violent localities in the West Bank, such as Jenin, Nablus, and Hebron (Mansour and Rees, 2012). During the conflict, the IDF have engaged in operations involving violence also against civilians in Palestine, such as occupations of Palestinian houses, beatings and abuse. Tight restrictions were imposed on the movement of people and goods within the Palestinian territories and between the territories and Israel (see, for instance, B 'Tselem, 2002;Calì and Miaari, 2013). As noted by Mataria et al. (2009), these mobility restrictions further depressed the Palestinian economy, leading to a rise in unemployment, a decline in real wages, and a reduction in the quantity and quality of food. The initial building of the separation barrier across the West Bank contributed to a further division of Palestinian communities by isolating them from their surroundings (UNSCO, 2014). The social and political consequences of this conflict did not stop with the end of the Second Intifada but continue to this day (Jaeger et al., 2012). An important aspect of the Second Intifada relevant to researchers is the temporal and spatial variation in the level of violence -which can be fruitfully exploited for the identification of causal effects. This is also made possible by the detail of information on the intensity of the conflict as it is made publicly available by non-governmental organizations such as B'Tselem. 2 Another feature of our analysis is the use of a direct, self-reported measure of household exposure to a violent event during the conflict. Such direct measures of exposure are often missing from conflict data. While retrospective self-reports have limitations, for instance due to recall bias, they improve our understanding of the consequences of conflict (Brück et al., 2015). Moreover, an individual family's exposure to the conflict might depend on a range of characteristics that are also linked with children's educational attainment, and is thus likely not exogenous. Therefore, we exploit the locality-level variation in conflict intensity to identify exogenous variation in households' experience of conflict-related events. An earlier example of an analysis that employs the number of fatalities in a given locality to measure conflict intensity during the Second Intifada is Brück et al. (2014). The authors use the year-to-year variation in the number of (Palestinian) victims at the school locality to investigate the short-term effects of violence on students' probability to pass the high-school final exam. We also use the locality-level number of fatalities as a measure of conflict intensity, but since we look at long-run effects of the conflict, we employ the cumulative number of victims over the entire Second Intifada period. Our identification relies on the assumption that the number of fatalities in the household locality only influences children's grades several years later through family exposure to conflict. This assumption may not hold if there exist time-invariant unobservable factors at the locality level correlated with both conflict intensity and children's school performance. To circumvent this problem, we control for a range of local labor market and institutional characteristics, such as unemployment rate, poverty rate, the proportion of the locality under the Israeli administration as well as the presence of the separation wall.
A further threat to the validity of our exclusion restriction arises from the endogenous sorting of households across localities in response to increased levels of violence. In particular, if households systematically move away, and children's primary education outcomes are correlated with the household decision to move, this can lead to biased estimates of family exposure to conflict. While we cannot address this concern directly, we note that in the West Bank during the Second Intifada, internal and external mobility has been extremely difficult and is still at a very low level. For example, Mansour and Rees (2012) point out that most Palestinian families living in the West Bank could not migrate because of the mobility restrictions imposed by Israel. Furthermore, data from the most recent (2007) census suggest that less than 10% of the current West Bank residents have moved away from the their mother's place of residence at birth (PCBS, 2010).
For our empirical analysis, we draw on data collected in 2013 in the West Bank, which contain information on a random sample of primary-school students in grades 5 to 9 and their families. These data are well suited for our purposes. First, they contain administrative information on children's primary school grades for three subsequent school years, i.e., 2010/2011, 2011/2012 and 2012/2013. Second, they contain retrospective self-reported information on parents' experience of violence during the Second Intifada. Finally, the data contain detailed information on children's skills, thereby enabling us to shed light on the potential mechanisms through which family's exposure to conflict may affect children's school achievement. Throughout the analysis, we combine the individual level data with locality-level information on the number of fatalities from the beginning of the Second Intifada in September 2000 to February 2005, allowing us to exploit the previously discussed geographical variation in fatalities.
Overall, our results demonstrate that family exposure to conflict during the Second Intifada has long-term effects on their children's primary education achievement in the West Bank. An additional event of household exposure to conflict reduces their children's grade point average (GPA) by 5.74 points (on a 0 to 100 scale), which corresponds to a decline of approximately 0.3 standard deviations. As regards the potential mechanisms through which the parents' exposure to conflict may affect children's performance in primary school, we show that more exposure to conflict leads to lower non-cognitive skills, such as conduct problems or a reduced level of conscientiousness. In contrast, there is little evidence that cognitive skills, although being important determinants of school achievement, are affected by family exposure to conflict.
The remainder of this paper is organized as follows. Section 2 presents a description of the data. Section 3 describes the empirical specification and identification strategy. Section 4 shows our main results and related robustness checks. Section 5 provides evidence on cognitive and non-cognitive skills as potential causal channels. Section 6 summarizes our findings and concludes.

Data
The data used in this paper are drawn from a new dataset collected in 2013 in the West Bank and East Jerusalem in the context of a joint research project between Wuppertal University, Hebrew University and Al-Quds University funded by the German Research Foundation (DFG). The dataset contains information on a sample of approximately 6,000 primary school students (around 4,000 girls and 2,000 boys -girls were oversampled) in grades 5 to 9 and their families. Students were randomly selected, stratified by grade level, from a random sample of 100 single-sex primary schools across the West Bank, stratified by region (North, Center and South) and school authority (public or UNRWA schools). 3 The dataset combines administrative and survey data. First, it contains individual grades in each subject for three subsequent school years, i.e., 2010/11, 2011/12, and 2012/13. This information was obtained from the Ministry of Education's administrative records. We use this information to create our main educational outcome in primary school: student GPA, which is computed as the average of the most recent grades (i.e., school year 2012/2013) obtained in the six main subjects: Arabic, English, Science, Mathematics, Social Science and National Education, and Religious Education. This set of subjects covers more than 80% of the weekly lessons (UNESCO, 2011). Students are graded on a scale ranging from 0 to 100.
Second, our dataset provides information at the individual and household level collected through paper-and-pencil student and family questionnaires. Of particular importance for our study is the parent-reported information on whether households themselves were exposed to violent events during the Second Intifada. This information is used to create the main explanatory variable in our analysis. The following question was asked of each household respondent: "During the Second Intifada, has your household experienced any of the following events?". The list of possible answers comprises seven items and reflects two aspects of the exposure to conflict: violence and economic consequences. For the purpose of this paper, we focus on the following violent events: "House was searched"; "House was occupied while you were in it"; "House was occupied and you were forced out"; "House was sealed or destroyed". Whereas one quarter of households reported that the house was searched, between 5% and 8% reported any of the other events. Furthermore, households are also asked about the year and month of the event. Among the households who reported exposure to conflict, 40% did not provide an answer regarding the date. The answers of those who gave a date are summarized in Figure B.1 in the Appendix. A small number of households reported events that happened during the First Intifada, some more reported events that happened after the end of the Second Intifada. These events were excluded from the analysis. Inspection of the reported dates further indicates that the events often took place in the same month and year, suggesting that answers to our exposure questions refer to the same episode. Apparently, there is no single "correct" way to use the self-reported information on exposure to conflict in our setting. For ease of interpretation, we use the raw count of events reported during the Second Intifada (a score ranging from zero to four). In a robustness analysis, we consider alternative definitions of exposure to conflict, and demonstrate that our results are not sensitive to these changes.
Third, our data contain information on children's cognitive and non-cognitive skills, which we use to study the mechanisms by which household exposure to conflict affects educational attainment. Cognitive ability is measured by three major cognitive tests (verbal, numerical, and figural), each consisting of subtests covering a wide range of item content (e.g., analogies, series, sentence completion, vocabulary). The subtests were selected and adapted from established tests of general ability: the Cognitive Ability Test (Thorndike and Hagen, 1971), Milta -a Hebrew version of the Lorge Thorndike Test (Ortar and Shachor, 1980), Standard Progressive Matrices (Raven, 1983), and the Culture Fair Intelligence Test (Cattell and Cattell, 1965).
Our indicators of non-cognitive skills are often used measures of mental health and personality traits. We assess student mental health using the Strengths and Difficulties Questionnaire (SDQ), a 25-item behavioral screening questionnaire designed to measure mental health problems in children and adolescents between 3 and 16 years old (Goodman, 1997;Goodman et al., 2000). 4 SDQ items are used to construct three indicators of children's mental health: the total difficulties scale, and the corresponding sub-scales for "internalizing" symptoms (emotional symptoms and peer problems) and "externalizing" behavior problems (conduct problems and hyperactivity/inattention). "Internalizing" symptoms are linked to anxiety and depression, whereas "externalizing" behavior problems are related to deviant and aggressive behaviors. Items for these two subcales are listed in Table A.1 in the Appendix . To facilitate interpretation, we standardize scores for internalizing and externalizing problems to have mean zero and standard deviation one.
Student personality is measured in terms of the Big Five model, a widely accepted approach for conceptualizing personality (Barenbaum and Winter, 2008;Goldberg, 1993), which is increasingly used also in labor market research to assess "soft" skills (Heckman and Kautz, 2012). According to this model, personality traits can be organized in five basic dimensions: openness, conscientiousness, extraversion, agreeableness, and neuroticism. We use a parent-reported 15-item version of the Big Five questionnaire (see Table A.2 in the Appendix). Scores on each dimension were z-standardized.
With regard to the local context, we use locality-level information on labor market and institutional characteristics, such as type of locality (rural, urban or refugee camp), unemployment and poverty rates, whether the locality is affected by the separation wall, and the proportion of a locality under area C. 5 Localities are the smallest administrative units defined by the Palestinian Central Bureau of Statistics (PCBS) based on their economic inter-linkages. There are 528 localities overall in the West Bank. Data on such a low geographic level are available only occasionally. For 4 The SDQ exists in three versions, to be answered by children, teachers, and parents, respectively, and is validated in many languages, including Arabic (see www.sdqinfo.org). In this paper, we use the parent-reported version included in the parental backgound questionnaire. 5 Since 1995, the West Bank is divided into three areas: A, B and C. Figure 1 shows a map of the West Bank and the geographical division of Palestinian territories into those areas. The Palestinian National Authority (PNA) has control over both administration (e.g., health, education, sanitation, water, and taxation) and security in area A, whereas control by the PNA is limited to administration in area B, and is absent in area C (Vishwanath et al., 2014). Communities in area C are at higher risk of adverse life circumstances due to the lack of primary services (UNSCO, 2014). instance, locality-level unemployment rates are available only for 2007 and are drawn from the PCBS. Information on unemployment for localities in East Jerusalem comes from the Applied Research Institute Jerusalem (ARIJ, 2014) and the Jerusalem Institute for Israel Studies. Poverty rates (available for 2009 only), the proportion of the locality under area C, and information on whether a locality is affected by the separation wall are obtained from the World Bank (Vishwanath et al., 2014). These contextual variables are then merged with the individual-level data using information on the residence of the household living in a locality.
Our most important contextual variable is the number of Palestinian fatalities at each child's school locality from September 2000 to February 2005, and is drawn from B'Tselem. These data measure local area conflict intensity during the Second Intifada. While the measures of conflict intensity used in the literature differ considerably, 6 the number of conflict-related Palestinian victims can be regarded as the best proxy for Palestinian exposure to conflict in the West Bank during the Second Intifada (Amodio and Di Maio, 2016).
Our analytical sample is constructed as follows: only observations with missing data for students' achievement in primary school and households' experience of political violence are excluded from the analysis. To minimize the loss of observations, missing values in any other variable are substituted by mean values. Additionally, an indicator for such missing values is created. 7 After these restrictions, we obtain a working sample that contains 4,235 children in grades 5 to 9 enrolled in 98 primary schools located across 74 distinct localities throughout the West Bank. 8 Table 1 reports descriptive statistics on the main variables used in the analysis. It consists of three panels, Panel A for the sample at the individual level, and Panel B and C for the corresponding samples at the school level and locality level, respectively. The children in our sample are on average 12.8 years old, and 70% are girls. Their GPA is on average 67.8 points and has a standard deviation of 17.6 points. On average, parents report to have been exposed to 0.45 conflict-related incidents during the Second Intifada. 9 41% of children in our sample have refugee status, meaning they are descendants of families who have been displaced after the 1948 Palestine war. 25% of fathers and 19% of mothers have a college degree, and 37% of fathers and 39% of mothers have a high school diploma. As regards children's mental health and personality, we show summary statistics before standardization. Children score, on average, 12.6 in terms of total difficulties, 6.1 in terms of "internalizing" symptoms, and 6.5 in terms of "externalizing" problems. These averages are fairly high compared to the corresponding figures from Western countries. For example, in the US children of the same age score on average 7.1 in terms of total difficulties, 4.0 in terms of "externalizing" problems and 3.0 in terms of "internalizing" symptoms (see www.sdqinfo.org).

Empirical Specification and Identification Strategy
The first part of our analysis is to estimate the effects of parents' exposure to conflict on their children's educational attainment. To this end, we estimate the following linear model for student school achievement: where the index isl denotes a child i enrolled in primary school s and residing in locality l at the year of interview. The outcome variable Y isl represents GPA of child i for the school year 2012/2013. Our variable of interest is Exposure isl , defined as the number of violent events during the Second Intifada reported by the household respondent (child's mother or father) and detailed in the previous section. Thus, β captures the impact on student GPA at the interview year of an additional event of parents' exposure to political violence at least seven years earlier. X isl is a vector of individual covariates, including a child sex, refugee status, dummies for children's and parental age, indicators for the maximum of father's and mother's educational level as well as indicators for missing values of these covariates. U sl is a school type indicator (UNRWA vs. public school). Z l is the set of locality-level labor market and institutional characteristics, such as unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, and type of locality (rural, urban or refugee camp). Finally, ε isl represents an idiosyncratic error term.
The need for an identification strategy arises from the potential correlation of household experience of violence during the Second Intifada with various unobservable determinants of student achievement. Such correlation, in particular related to family background characteristics but also education infrastructure, may confound our relationship of interest. One obvious concern might be, for example, that richer and better educated families -who have children of better academic ability -may reside in certain areas less prone to Palestinian violence that may have provoked counter-violence by the IDF. While we are able to control for observed background characteristics to address this concern, there may still be unobserved confounders.
We therefore apply an IV strategy to identify the long-term effect of household exposure to conflict on children's primary school achievement, with the number of Palestinian fatalities at the locality level being used as instrument for family exposure to conflict. However, since the present analysis focuses on the long-term academic outcomes of children, seven years after the conclusion of the conflict, we can exploit only the cross-locality variation in the total number of Palestinian victims (and not the variation over time within the locality during the Second Intifada) to identify the effect of family exposure to conflict. Thus, one potential threat to the validity of our instrument is the omitted variable bias resulting from the presence of time-invariant unobservable factors at the locality level correlated with both conflict intensity and child primary education performance. We argue that this problem is substantially mitigated by controlling for locality-level contextual variables.
Model (1) is estimated using two stage least squares (2SLS), with the following first stage regression: Exposure isl = η + δ Fatalities l + θX isl + ζU sl + σZ l + ν isl (2) where Exposure isl is regressed on Fatalities l , defined as the locality-level number of all conflictrelated Palestinian victims during the Second Intifada. X isl , Z l , and U sl are defined in the same way as in Equation (1). Standard errors are clustered by locality, the level of variation of our instrument.
To illustrate the variation of our instrument, Figure 1 shows a map of the West Bank. Each circle represents a locality contained in our sample. The size of the circle indicates the locality-level number of Palestinian fatalities during the Second Intifada. Overall, Figure 1 shows a substantial degree of variation in the number of Palestinian fatalites, with the highest levels of conflict intensity being concentrated in some very violent localities, such as Nablus, Jenin, Tulkarm, Hebron and Ramallah. Figure 2 shows the relationship between the total number of fatalities during the Second Intifada at the governorate level (there are 11 governorates in the West Bank) and the average level of exposure to conflict as reported by parents. This graph reflects the (first stage) variation we exploit in our analysis. Table 2 reports the estimates of the effects of an increase in household exposure to conflict during the Second Intifada on children's primary school achievement obtained from an OLS regression (Column 1) and from a 2SLS regression (Column 4). In addition, we show the corresponding results from the reduced form (Column 2) and first stage (Column 3) regressions. As described in the previous section, in each regression we include individual-level, school-level, and locality-level control variables.

Main Results
The OLS coefficient suggests a negative and significant association between households' exposure to conflict and their children's school performance: an additional event of parents' experience of conflict is associated with a 1.32 point decrease in student GPA in primary school, which corresponds to approximately 8% of a standard deviation. The reduced form regression suggests that GPA drops by approximately two points per one hundred fatalities in the locality.
The first stage regression shows that one hundred more fatalities in the locality increase parents' exposure to conflict by 0.35 events on average. With a first stage F-statistic of 46, our instrument easily passes conventional thresholds for strong instruments. The 2SLS coefficient on parents' exposure to conflict remains negative and statistically significant, and is larger than our OLS estimate. We find that one additional event of parents' exposure to conflict induced by the Second Intifada reduces the primary school GPA of their children by 5.74 points. This effect represents a decline of about 30% of a standard deviation. As previously mentioned, we also estimate a range of regressions with alternative operational definitions of the exposure variable, obtaining very similar results. Details can be found in Table  B.1 in the Appendix. Importantly, as shown in Columns E and F, our instrument does not predict exposure to conflict before or after the Second Intifada. We interpret this as evidence that the instrument does not pick up any trends in location-specific levels of violence.

Robustness Checks
We assess the robustness of our main results to a number of further specification changes. Results are reported in Table 3. First, in Column (1) we check the sensitivity of our results to a change in the functional form of the first stage and reduced form regressions. The major concern here is that exposure to conflict and GPA may not follow linear functions of the number of fatalities. To allow for this possibility, we have recoded our instrument into 4 categories (as shown in Figure 1): 0 fatalities, 1-10 fatalities, 11-50 fatalities, and >51 fatalities. Our point estimate changes somewhat, we lose precision and the first stage F-statistic decreases, but the qualitative result remains.
Second, to investigate heterogeneous effects across the distribution of GPA, in Columns (2) to (4) we report the treatment effects at the 25th, 50th and 75th percentile. Our results suggest that the effect of family exposure to conflict induced by the Second Intifada is concentrated in the lower quantiles of the GPA distribution. In other words, the long-term effects of family exposure to conflict induced by the Second Intifada seem to be driven by poor academically performing students.
Third, another concern regards the sensitivity of our findings with respect to the year in which school achievement in primary school is measured. As previously mentioned, our dataset provides information on student GPA not only for the school year 2012/2013 but also for the two preceding Notes: Standard errors are reported in parentheses and are clustered at the locality level. Individual-level controls include sex of child, sex of household respondent, dummies for children's and parental age, dummies for parental educational levels, as well as indicators for missing values in these covariates. The school-level control variable is an indicator for UNRWA versus public school. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, and type of locality (rural, urban or camp). * Significant at 10%; ** significant at 5%; *** significant at 1%. school years, i.e., 2010/2011 and 2011/2012. Hence, we run two additional 2SLS regressions of model (1) using student GPA for the school years 2010/2011 (Column 5) and 2011/2012 (Column 6) as alternative outcomes. Results remains qualitatively unchanged relative to the main specification: the coefficient on family exposure to conflict has negative and significant effects on children's GPA in primary school. Note, however, that point estimates are actually smaller in the two earlier years than in 2012/13. This suggests that effects on student GPA do not tend to subside over time. We interpret this as corroborating evidence for the plausibility and usefulness of our long-term analysis. Fourth, in Column (7) we verify the robustness of our results when using a broader measure of conflict intensity that includes the locality-level number of all Palestinian and Israeli victims during the Second Intifada. The reason for this check is that both sides of the conflict may react in a regular and predictable way to violence against them. This would imply that an increase in the number of conflict-related Israelis fatalities may lead to more violent actions against the civilian population in the West Bank. Again, the estimated parameter resembles closely the one obtained in the benchmark specification.
Next, we check what happens when we include governorate fixed effects (Column 8) in model (1). Identification then rests on within-governorate variation in the intensity of conflict. Differences in exposure to violence or children's educational attainment that are linked to the larger region are hence controlled for -addressing potential concerns about the validity of our instrument. Again, we find that the point estimate hardly changes. Standard errors increase a bit, but the first stage F-statistic remains safely above 20. In a related robustness check, we exclude single governorates one at a time to examine whether the results are driven by a specific governorate. This exercise, of which we do not report detailed results, confirms that our main results are robust.
As a final check, in Columns (9) and (10) we split the sample in two parts based on the date of birth of the children in our sample. Specifically, in Column (9) we restrict the sample to all children born before February 2000. These children have attended school for at least one year during the Second Intifada. One might argue that some part of the effect we measure is not due to parental exposure to conflict but rather to disruptions to teaching, such as closing of schools, detainment of teachers, road closures keeping children from reaching schools and so on (Brück et al., 2014). In Column (10) we focus on children born in February 1999 or later, who have entered school in September 2005 (after the end of the Second Intifada), and whose schooling has not been directly affected by the conflict. Taken together, the results in Columns (9) and (10) do not provide any evidence for differential effects of household exposure to conflict by date of birth.

Potential Mechanisms
We now turn to the second part of our analysis: identification of potential mechanisms underlying the relationship between family exposure to conflict and primary school achievement of children. We hypothesize that exposure to conflict and violence in the early stages of a child's development may have long-term detrimental consequences on skills, both cognitive and non-cognitive, which in turn translate into worse school outcomes. Our hypothesis is related to a recent strand of the literature suggesting that non-cognitive skills have independent predictive power for a wide range of socio-economic outcomes. For instance, Heckman et al. (2006) show that non-cognitive skillswhich include personality traits -are important determinants of academic and economic success, and that their long-run effects seem comparable to the ones of cognitive skills. Currie and Stabile (2009) argue that non-cognitive skills are conceptually linked with mental health. They show that mental health problems in early childhood, in particular those associated with "externalizing" and "internalizing" behavior problems, have negative and persistent effects on future educational outcomes.
With our data, we are not able to completely identify the causal path from exposure to conflict to educational attainment via impaired cognitive and non-cognitive skills. However, we are able to study if -using the same identification strategy as before -exposure to conflict causally impacts these mediating variables. Finding no effect would rule out certain candidate channels.
Before we report the results of 2SLS regressions with the mediators as dependent variables, we show that they have an "effect" on educational attainment at least in a descriptive sense. As regards cognitive ability, we consider the sub-scores for numerical, verbal and figural tests, as well as the total score obtained in the three tests; as concerns non-cognitive skills, we use the score for total strengths and difficulties (SDQ), the corresponding sub-scores for "internalizing" Notes: Standard errors are reported in parentheses and are clustered at the locality level. Individual-level controls include a female dummy for children and parents, dummies for children's and parental age, dummies for parental educational levels, as well as indicators for missing values in these covariates. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, as well as indicators for UNRWA school and type of locality (rural, urban or camp). * Significant at 10%; ** significant at 5%; *** significant at 1%. and "externalizing" behavioral problems, as well as the Big Five personality traits (i.e., openness, conscientiousness, extraversion, agreeableness and neuroticism). Table 4 shows the results of OLS regressions of GPA on z-standardized measures of cognitive ability, scores for internalizing and externalizing problems, as well as the Big 5 personality dimensions. Column (1) shows the results of twelve separate regressions, where each mediating variable is included one at a time. As expected, cognitive ability is strongly related to GPA. Moreover, all measures of non-cognitive skills are significant predictors of GPA as well. Since the coefficients show the estimated association in terms of standard deviation changes, a comparison of the size of the coefficients illustrates the quantitative relevance of non-cognitive skills. For instance, externalizing problems and conscientiousness are about half as strong predictors as cognitive skills.
Column (2) shows results when the cognitive ability sub-scores and the SDQ sub-scores are included jointly. We find that the influence of internalizing problems disappears almost completely. The coefficient on externalizing problems becomes smaller but remains strong. In Column (3) we report the results when the Big 5 and cognitive ability are included jointly, and we find that conscientiousness remains the only personality trait with a sizeable association with GPA. Overall, cognitive and non-cognitive skills have independent "effects" on children's educational achievement. The most important non-cognitive skills are conscientiousness and (the absence) of externalizing, i.e. behavioral, problems.
To establish whether these variables lie on the causal path from exposure to conflict to GPA, we now estimate the same specification as in model (1) with children's cognitive and non-cognitive skills as outcomes. We report the results of this analysis in Table 5. When examining cognitive ability (Panel A), we do not find any significant effects, with the magnitude being close to zero for numeracy and positive coefficients for the figural and verbal tests. In contrast, the results for noncognitive skills (Panel B) reveal that family experience of conflict increases children's SDQ total score by 0.2 standard deviations. This coefficient masks heterogeneous effects between "internalizing" problems (Column 2), where our coefficient of interest remains positive but the magnitude is substantially reduced, and "externalizing" behavior problems (Column 3), for which we find a significant 0.25 standard deviations effect of family exposure to conflict. This evidence is in line with the results of Currie andStabile (2006, 2009), who find that "externalizing" behavior problems are more likely to lead to negative outcomes than "internalizing" problems. We therefore believe that increased "externalizing" behavior problems may be one important channel through which exposure to conflict at least seven years in the past affects children's GPA today.
We now move to the examination of the Big Five personality traits (see Columns 4 to 8 of Panel B). Our main result is that parents' exposure to conflict during the Second Intifada reduces children's level of conscientiousness by 0.23 standard deviations. In light of the evidence reported in Table 4, this finding suggests another important channel, possibly related to the "externalizing" problems channel. Extraversion, agreeableness and neuroticism also seem to be affected by family exposure to conflict. However, as demonstrated in Table 4, these personality traits show no significant correlation with GPA, once cognitive skills and the other personality dimensions are controlled for. The remaining personality trait, openness, is instead not affected by the exposure to conflict. Standard errors are reported in parentheses and are clustered at the locality level. All explanatory variables were z-standardized Individual-level controls include children's age and month of birth, parental age, gender of parents and children, and parental education. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, as well as indicators for UNRWA school and type of locality (rural, urban or camp). * Significant at 10%; ** significant at 5%; *** significant at 1%.  Notes: Standard errors are reported in parentheses and are clustered at the locality level. Individual-level controls include a female dummy for children and parents, dummies for children's and parental age, dummies for parental educational levels, as well as indicators for missing values in these covariates. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, as well as indicators for UNRWA school and type of locality (rural, urban or camp). * Significant at 10%; ** significant at 5%; *** significant at 1%.

Discussion and Conclusion
In this paper we answer two related questions: does a family's exposure to violent political conflict affect the educational attainment of its children not only in the short-but also in the long-run? And if so, can we identify important channels that could be addressed to mitigate the adverse affects of political conflicts on educational outcomes?
While several studies from a variety of different countries have shown short-term consequences of exposure to conflict on children's educational achievement, less is known about a causal relationship in the long-run. We provide new evidence that household exposure to conflict during the Second Intifada has long-term impacts on the human capital of students in the West Bank. Specifically, we examine the effect of house searches or occupations by the Israeli army between 2000 and 2005 on students' grade point averages in primary school (grades 5 to 9) in 2013.
Our identification strategy exploits the variation in the locality-level number of Palestinian fatalities during the Second Intifada as an arguably exogenous determinant of a family's exposure to conflict-related events. According to our estimates, one additional event reduces their children's GPA by approximately 30% of a standard deviation of the grade point average distribution. This result is quantitatively robust to a number of specification changes, such as different definitions of the exposure variable, using a categorical instead of a continuous instrument, including Israeli fatalities as well in our instrument, or using GPAs in different school years. We also show that the effect of exposure to conflict is concentrated in the bottom and middle of the GPA distribution. In contrast, only small insignificant effects can be shown for the likelihood of being in the top quartile of the distribution.
With regard to causal channels, we estimate the effect of exposure to violence on detailed measures of cognitive and non-cognitive skills. We find no effect of exposure on performance in a long and thorough cognitive skills (IQ) test. In contrast, important non-cognitive correlates of educational achievement, such as externalizing behavior problems and conscientiousness, appear to be affected. This finding provides an interesting mirror image of findings on the effect of early education programs. For instance, with regard to the Perry Preschool Program, Heckman et al. (2013) conclude: "Although Perry did not produce long run gains in IQ, it did create persistent improvements in personality skills. The Perry program substantially improved externalizing behaviors (aggressive, antisocial, and rule-breaking behaviors)" (p. 2053). It almost appears as if exposure to conflict did the same -with a negative sign: although exposure is not detrimental to IQ, it leads to worse personality skills, which in turn affect educational attainment and other future outcomes.
Several limitations of our study are worth being mentioned. Although we have no theoretical prior as to the direction in which OLS would be biased compared to the true causal effect of exposure, the size of our 2SLS coefficients may raise concerns about the validity of our results. Large 2SLS-compared to OLS-estimates are quite common in applied research and three explanations can be put forward. The first is measurement error in the explanatory variable, which (if classical) attenuates the OLS coefficients. Clearly, if some households who have experienced violence do not report this in our survey, whereas others over-report events, and if misreporting is uncorrelated with the error term, OLS estimates would be biased towards zero. However, it seems at least equally plausible to assume that misreporting is in some way related to unobserved determinants of student achievement. The second explanation is that the IV approach identifies (a weighted average of) complier-specific causal effects, which can potentially be larger than OLS estimates. It is plausible that at least among always-takers (households prone to conflict events no matter what the level of conflict in the locality) the effect is smaller than among compliers (households only affected if the level of conflict in the locality is high). If this also holds for never-takers is unclear. A third explanation -potentially damaging to the IV approach -is that the exclusion restriction does not hold. Of course, this cannot be entirely ruled out in our setting. To partially address this concern, we report the reduced form coefficients for all specifications in Tables B.2 and B.3 in the Appendix. These results clearly show that locality-level conflict intensity during the Second Intifada is associated with worse educational attainment more than seven years later, and that this association is robust to a number of specification changes. Thus, even if the exclusion restriction does not hold, there is robust evidence that the violent political conflict has long-term consequences for educational attainment mediated through non-cognitive skills. This reduced form analysis is similar to what most of the literature has done when data on individual experience of conflict are lacking (see, for instance, Brück et al. (2014)).
Finally, in terms of policy recommendations, our paper suggests that children whose families have been exposed to conflict should be identified and invited to participate in interventions that are designed to improve social skills and self-control. Such interventions will likely lead to better educational attainment in school and possibly better labor market outcomes. When this is not possible during a conflict, children should be the main focus of interventions in the post-conflict recovery period.
A Description of non-cognitive skills measures Parents were asked to state how much they agreed with each statement about their child on a 3-point scale ("Not true", "Somewhat true", "Certainly true"). Both internalizing and externalizing problems are defined on a 0 to 20 scale. The total score for SDQ is given by the sum of subscores for internalizing and externalizing problems (on a 0 to 40 scale), and does not include the prosocial behavior score (Goodman, 1997). * This conduct problem was accidentally omitted of the instrument, i.e., the scale representing conduct problems consists only of four items and was scaled up to a maximum of 10.     Notes: Standard errors are reported in parentheses and are clustered at the locality level. Individual-level controls include a female dummy for children and parents, dummies for children's and parental age, dummies for parental educational levels, as well as indicators for missing values in these covariates. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, as well as indicators for UNRWA school and type of locality (rural, urban or camp). * significant at 10%; ** significant at 5%; *** significant at 1%. Notes: Standard errors are reported in parentheses and are clustered at the locality level. Individual-level controls include a female dummy for children and parents, dummies for children's and parental age, dummies for parental educational levels, as well as indicators for missing values in these covariates. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, as well as indicators for UNRWA school and type of locality (rural, urban or camp). * Significant at 10%; ** significant at 5%; *** significant at 1%.
(3) Notes: Standard errors are reported in parentheses and are clustered at the locality level. Individual-level controls include a female dummy for children and parents, dummies for children's and parental age, dummies for parental educational levels, as well as indicators for missing values in these covariates. Locality-level variables contain unemployment rate, poverty rate, the proportion of locality under area C, the presence of the separation wall, as well as indicators for UNRWA school and type of locality (rural, urban or camp). * Significant at 10%; ** significant at 5%; *** significant at 1%.