Abstract
This study analyzes the impact of terror attacks on students’ academic achievement in Israel between 2001 and 2005, during the Second Intifada. Using within-student variation in exposure to terror attacks before exams, I find that a fatal terror attack before an exam adversely affects performance. The adverse effect, however, disappears for exams held five days or more after the attack. I explore potential explanations for these results, suggesting psychological impacts best explain the short-lived effect. Moreover, the temporary decrease in test scores does not affect the quality of diploma earned, suggesting no long-term effect on human capital accumulation.
I. Introduction
Community-wide traumatic events, such as terror attacks and mass shootings, are a common phenomenon around the world.1 Beyond their direct effects in deaths, injuries, and damaged property, these events have large adverse psychological and economic effects on individuals in the broader community. Children and adolescents may be particularly vulnerable.2 Indeed, recent studies show that exposure to such events is associated with lower student performance, an important predictor for social and economic well-being (Hanushek and Woessmann 2008).3
Traumatic events can affect student performance through different channels. The literature generally focuses on two of these: (i) the psychological ramifications of such events, including symptoms of anxiety and stress,4 and (ii) deleterious effects on key school inputs that shape the learning environment, such as reduced instructional time, impaired teaching quality, and damage to infrastructure.5 Understanding the role of each channel is essential for designing effective policies to reduce the negative impact of such shocks. However, since performance is usually measured either once overall or once in each academic year, it is often challenging to distinguish between psychological effects and effects on the school supply side.
Using a unique data set, this study overcomes the existing challenges in the literature to study (i) the short-term effect of exposure to terror attacks on students’ performance while school inputs are fixed and (ii) the potential extent of that exposure on long-term human capital accumulation. To this end, I use a series of fatal terror attacks during the Second Intifada in Israel (2001–2005) to identify their causal short-term effects on students’ performance in multiple end-of-high-school matriculation exams and the implications on end-of-high-school matriculation outcomes. The matriculation exams are a series of national examinations in different subjects that Israeli students take on different days during a particular period of the year in order to obtain a matriculation diploma when they finish high school. The matriculation diploma is a prerequisite for higher education in Israel and has a large impact on students’ future prospects. Access to college majors is determined largely by matriculation performance, with many lucrative professional programs requiring minimum overall average scores for admission. As a consequence, matriculation scores can affect an individual’s entire academic career and subsequent labor market outcomes.
The Israeli context offers a useful opportunity to study the effects of traumatic events on students’ short-term performance and the potential effect on long-term human capital accumulation for several reasons. First, the Second Intifada in Israel was characterized by multiple terror events during a relatively short time but with moderate and intermittent intensity, and schooling and educational routines as well as other public services were never interrupted during this period.6 Hence, I am able to exploit random variation in the time and place of terror attacks relative to exam timing without any significant disruptions to infrastructure or school supply inputs. Second, the Israeli matriculation system provides variation both between and within students, with students taking different exams on different dates within a few weeks. This allows me to track the performance of the same student over a short time and compare their relative performance in subsequent exams, considering the elapsed time from the terror attack to the exam. Finally, since the exams are a key component of the matriculation diploma awarded to Israeli students who complete certain requirements by the end of high school, I am able to examine how exposure to terror events before these exams affects the likelihood of a student obtaining a matriculation diploma and its quality, measured by the student’s composite score.
In my empirical analysis, I combine two highly detailed data sets. The first data set contains incident-level information on all fatal terror attacks in Israel between 2001 and 2005. The second data set contains test scores for about a quarter of a million Israeli 12th-grade students taking national matriculation exams over the course of several weeks each year. These student-level exam records are linked to data on high school locality (where exams take place), residence locality (usually the same as the school), and matriculation diploma status. Merging the data on 132 deadly terror attacks (41 of which took place during matriculation exam periods) with nearly two million exam observations by date and location makes it possible to determine the intensity of terror in any location during a given period preceding any matriculation exam.7
In the first part of my analysis, I examine the impact of exposure to terror attacks shortly before an exam on exam performance. My identification strategy exploits the fact that students take the matriculation exams over several weeks, which enables me to examine the relationship between exposure to a terror attack and scholastic performance across the same student’s exams. Concretely, I apply a student fixed effects strategy to identify the causal short-term effects of a fatal terror attack in the student’s area up to two weeks before an exam on the student’s exam score. The identification strategy relies on the premise that within-student variation in exposure to terror attacks shortly before exams is not correlated with unobserved determinants of educational performance. This is a plausible assumption because matriculation exams are compulsory, and the dates of these exams are determined by the Ministry of Education long in advance, with no opportunity for rescheduling within an examination period. Indeed, I do not find any indication that terror attacks affect decisions about test-taking (that is, they do not lead students to skip or defer exams). This provides a context for analysis in which variation in exposure to terror attacks shortly before an exam for a given individual across multiple exams is essentially random.
In my preferred specification, which includes student-by-exam-period fixed effects and controls for exam subjects and proficiency level, I find that the occurrence of a fatal terror attack in a student’s area reduces the student’s performance on that exam only if the attack took place within the four days before the exam. More precisely, each additional fatality from a terror attack in a student’s vicinity within a four-day window before an exam significantly reduces their test score by 0.005 of a standard deviation.8 The effect is driven predominantly by attacks that cause a relatively high number of fatalities (an attack causing more than ten Israeli fatalities reduces test scores by 0.1 standard deviation) and falls with the distance between the attack and the student’s location. Also, the effects seem to occur mainly at the top of the score distribution and do not decrease the likelihood of a student passing the exam. I find no significant heterogeneity in the short-term effects of terror by student characteristics or exam subjects. Importantly, I find that the effect is transitory and short-lived. In particular, I do not find that terror attacks that took place five days or more before the exam have a significant effect on exam performance, which suggests resilience among students. I discuss potential explanations for my findings and suggest that psychological stress is an important mechanism behind the short-lived effect of terror on students’ performance. Also, the inclusion of student fixed effects in the analysis enables me to rule out any school supply-side effects on the student learning environment that are fixed within students over a short time (for example, factors such as school quality, school accessibility, absenteeism, or educational investment decisions).
In the second part of my analysis, I explore whether the short-term adverse effect of terror exposure on exam performance also has implications for the likelihood of a student obtaining a matriculation diploma, and for the quality of that diploma (based on the student’s composite score)—two end-of-high-school outcomes that are meaningful for postsecondary schooling. To this end, I use student-level data and exploit the variation among students in their exposure to terror attacks during examination periods at the same school. This within-school variation is considerable, since students in any given school take examinations in several elective subjects, and the dates of exams vary by subject and cohort. Estimating a model with school fixed effects, I observe no effect of terror exposure before exams on a student’s likelihood of obtaining a matriculation diploma and only a small and economically insignificant effect on the quality of the diploma earned. I discuss possible explanations for this result and suggest that since the quality of the final matriculation diploma relies on multiple evaluations, poor performance in one test caused by short-lived effect of terror exposure does not necessarily translate into a persistent effect on human capital outcomes.
This study contributes to the existing literature on how various types of violent events affect students’ academic achievement. For example, Beland and Kim (2016), Gershenson and Tekin (2018), Rossin-Slater et al. (2020), Ang (2021), Levine and McKnight (2021), Cabral et al. (2021), and Bharadwaj et al. (2021) examine the short- and long-term consequences of exposure to gun violence on educational, economic, and health outcomes, and Jarillo et al. (2016), Monteiro and Rocha (2017), and Brück, Di Maio, and Miaari (2019) study the effect of cumulative exposure to conflict violence on students’ performance at the end of the school year. However, those studies analyze the effect of either cumulative or one-time exposure to violence on outcomes typically measured once for each school or student overall or in each academic year or semester. Hence, the mechanisms behind their negative results may include both psychological impacts and school supply-side disruptions, such as school absenteeism, reduced teaching quality, and other forms of degradation in the learning environment. Sharkey et al. (2014), Poutvaara and Ropponen (2018), and Michaelsen and Salardi (2020) focus on short-term effects by comparing the exam performance of students exposed to violence a few days before the exam to that of other students who were not exposed, under the assumption that school supply-side effects are unlikely to operate over a short period. However, they do not consider the long-term consequences of such short-term exposure.
This work is the first to study both the short-term effects of exposure to terror attacks and the potential implications of that exposure for long-term human capital outcomes. This study complements the existing literature by providing novel evidence of human resilience, showing that exposure to terror has a significant impact on performance, but only over a short time window. Moreover, the exam-level data and the within-student variation in exposure to terror attacks helps provide compelling evidence for the effects of terrorism on academic achievement when school inputs are not affected and overcomes potential selection bias caused by the possibility that violent events may also affect the decision to take the exam.
In the following, the next section provides background on the Second Intifada and the Israeli matriculation exam system. Section III describes the data and presents summary statistics. Section IV presents the analysis for the short-term effect on exam performance. Section V presents the analysis for the effect on educational outcomes with longer-term implications, and Section VI concludes.
II. Background
A. The Israeli–Palestinian Conflict and the Second Intifada
The Israeli–Palestinian conflict began almost 70 years ago and is characterized by alternating periods of low- and high-intensity violence. One of the most intense periods of violence was the Palestinian uprising known as the Second Intifada, which was marked, on the Israeli side, by repeated terror attacks and assaults against Israeli civilians. The Second Intifada began on September 29, 2000, peaked in 2002, and subsided in 2005. In the years since 2005, the number of attempted terror attacks within Israel has fluctuated but has not returned to 2004 levels, and suicide bombings are no longer widespread.9
During the Second Intifada, Israelis experienced more than 100 fatal terror attacks within the country’s pre-1967 borders and more than 200 fatal terror attacks in Israeli settlements in the West Bank and Gaza Strip, which together claimed the lives of more than 1,000 Israelis (civilians and security forces). More than 80 percent of the fatal terror incidents within Israel’s pre-1967 borders were suicide bombings aimed almost exclusively at Israeli civilian targets.10 Most were in the country’s largest and most densely populated cities (particularly Jerusalem and Tel Aviv). Concurrently, the Israel Defense Forces (IDF) operated inside Palestinian cities, resulting in more than 3,300 deaths among Palestinian civilians and militants (B’Tselem 2008).
The violence that characterized the Second Intifada persisted throughout the whole period, though its intensity varied over time and across localities. Importantly, however, in comparison to large terror attacks and ongoing or recurrent violence experienced by developing countries, the attacks that Israel sustained during the Second Intifada were on a moderate scale. For example, the most deadly attack (at the Park Hotel in Netanya on March 27, 2002) caused 30 fatalities, and the homicide rate (per 100,000 population), including acts of terrorists, was 3.1, which is about half the homicide rate in the United States at the time.11 In that sense, it more closely resembles local community violence, such as that frequently found in the United States, or the repeated small-scale terror attacks seen in several countries in Europe in recent years.
B. The Israeli High School Matriculation Exam System
High school in Israel runs from the 10th to the 12th grade. Students on an academic track are required to obtain a matriculation certificate when they finish high school in order to proceed to tertiary education. Almost all high school graduates take at least one matriculation exam during their high school years, but only about two-thirds earn a certificate by the time they graduate from high school.12
To earn a certificate by the time of their high school graduation, students must pass a series of national exams in core and elective subjects. Within subjects, students are tested at different proficiency levels, and credits are awarded accordingly: one credit for the least demanding exam through five credits for the most difficult exam. A minimum of 21 credits are needed to qualify for a matriculation certificate.13 Students choose their matriculation study programs (proficiency levels and elective subjects) during the 10th or 11th grade and study the material for each subject over the course of at least one academic year before taking the relevant matriculation exams.14
In addition to the number of credit units earned in each subject, each student’s matriculation certificate specifies a final score for each subject and an overall composite score produced by averaging the student’s final subject scores. For their part, final scores in each subject are the mean of the student’s score in the national exams just described, which are “external” to the school (see below), and an “internal” school-level score (also known as the protective score) awarded by the student’s own teacher.15 Scores are on a 1–100 scale, with a passing grade of 55.
Credit units in a given subject may be awarded based on either a single exam or a set of exams, with each exam paper conferring a certain number of credits. In the latter case, the sum of the credit units on the individual exams is the total for the subject. Where students take more than one exam in a given subject, they may do so on the same day or during different exam periods.16 Throughout this paper, the term “exam” refers to any individual exam paper. Online Appendix Table A1 provides sample contents of a matriculation certificate and a sample exam schedule with scores.
The national matriculation examinations are administered biannually during two periods: a winter season in January and February, for the mandatory subjects only, and a summer season in May and June, for all subjects.17 The Ministry of Education sets the dates and times of the exams well in advance, and exams in a given subject are administered simultaneously across the country.18 Exams are taken at the student’s home school and last 2.5 hours on average. Within subjects, all exams (that is, exams at all difficulty levels) are administered simultaneously. High school students are assigned to the matriculation exams in advance by the school based on their coursework during the academic year. Hence, it is not possible for students to take a different exam than the exam to which they were assigned. Students who are dissatisfied with their scores on a specific exam may choose to be retested in the same subject in a subsequent examination period at either the same or a different proficiency level as long as they remain in high school. However, since most exams are taken during the summer season at the end of 12th grade, retaking exams before graduation is rare.19
The national exams are prepared, administered, supervised, and graded by an independent agency. Therefore, scores are comparable across schools and students within examination periods. In particular, examiners do not know the names, schools, or teachers of students whose exams they grade. For this reason, and because each examiner grades exams from different schools in different regions, the probability of normalizing scores in response to observing a low score distribution in a particular school or region is very low. Only if scores are relatively low nationwide does the Ministry of Education normalize scores across the board. Moreover, the grading process takes several weeks, and scores are only available to students a few months after the end of the examination period.
Obtaining a high school matriculation certificate is one of Israel’s most economically important educational milestones. The matriculation certificate is a prerequisite for study at universities and most academic and teachers’ colleges. Students are admitted to postsecondary programs on the basis of their composite scores (that is, the average of their final scores in all subjects), their credit units and final scores in subjects relevant to the program, and their scores in a psychometric examination (akin to the American SAT).20 Each university ranks applicants according to the same formula, thus producing an index based on a weighted average of the student’s average score on all their matriculation exams and the psychometric examination. This ranking determines students’ eligibility for university admission, and even which major they can choose within the university. Therefore, a negative shock affecting exam performance, such as a terror attack, could have implications for students’ postsecondary schooling, whether by limiting their ability to pursue higher education at all or by restricting the type of higher education pursued and the quality of the institution ultimately attended.
III. Data and Summary Statistics
A. The Terror Data
The data used in this study comprise a daily record of terror incidents resulting in at least one Israeli fatality within Israel proper (excluding Israeli settlements in the West Bank and Gaza Strip) between November 2000, shortly after the beginning of the Second Intifada, and the end of 2005, when terror attacks within Israel had mostly subsided.21 The main source of this data is B’Tselem, the Israeli Information Center for Human Rights in the Occupied Territories (B’Tselem 2008), cross-checked against information publicly available from the Israel Ministry of Foreign Affairs.22 The database contains information on all Israeli fatalities from terror attacks (including civilians and security forces), itemized by date and location of the fatal incident. In this data set, terror attacks are defined as premeditated, politically motivated violence perpetrated against noncombatant targets, where the term “noncombatant” is construed as including, in addition to civilians, military personnel who were unarmed and/or not on duty at the time of the incident.23
Over the full study period, Israel experienced 132 separate attacks that together claimed 632 Israeli lives, of whom 85 percent were civilians. Of these, 41 attacks causing 222 fatalities took place during matriculation exam periods. Figure 1 plots daily terror fatalities within Israel from the beginning to the end of the sample period. The shaded areas identify periods when matriculation exams were administered. The figure clearly shows large variation in the frequency and severity of attacks, with a peak in 2002, and both high- and low-intensity intervals coinciding with examination periods.
Daily Terror Fatalities Within Israel During the Sample Period
Notes: The figure plots daily Israeli fatalities (civilians and security forces) from terror attacks within Israel excluding the West Bank and Gaza Strip from November 2000 through December 2005. Shaded areas represent matriculation exam periods.
Figure 2 depicts the distribution of fatalities across Israel’s “natural regions”—the third level of Israel’s formal administrative units—during the sample period.24 Panel A shows the distribution of all fatalities from November 2000 to December 2005, while Panel B shows the distribution of only fatalities that took place during matriculation examination periods. As can be seen, the distribution of fatalities was characterized by geographic as well as temporal variation. The number of fatalities and incidents was especially high in the most populated areas of the country, in particular in the Jerusalem District.
Distribution of Terror Fatalities Across Israel’s Natural Regions, 2001–2005
Notes: Panel A presents the distribution of all Israeli fatalities (civilians and security forces) from terror attacks across Israel’s natural regions excluding the West Bank and Gaza Strip from November 2000 through December 2005. Panel B presents the same data for fatalities from terror attacks occurring during examination periods from November 2000 through December 2005.
B. The Education Data
The educational database contains administrative records of all high school students collected by the Israeli Ministry of Education for the school years 2001–2005 (coinciding with the terror data). This data set includes both matriculation test data and rich demographic information on students. The test data cover any exam taken by any student in any subject during the study period, offering observations for students’ performance on different dates and across the full range of subjects. Demographic information includes the student’s gender, parents’ education level, number of siblings, country of origin, ethnicity, and identification of the high school and locality at the area (natural region) level (due to privacy limitations of the data).
The data set used in the analysis is at the exam level and includes the student’s score on the exam, the student’s internal school score, the number of credit units earned, and the date and time of the exam.25 Note that I observe all exams taken in 12th grade, meaning that if a student retook an exam during 12th grade, I observe both outcomes. The exam-level data were augmented by additional information, including whether the student successfully obtained a matriculation certificate after completing high school, the total credit units awarded in the certificate (including the total in each subject), and the student’s composite matriculation score. The primary measure of exam performance in this study is standardized exam performance by subject, proficiency level, and examination date, though the results are robust to using scale scores while controlling for subject, proficiency level, and examination period.
The sample is limited to 12th-grade students who attended schools in the regular Jewish state system (excluding regional schools) and followed the same general matriculation curriculum. There are several reasons for focusing only on 12th-graders. First, most matriculation exams are given in the 12th grade (more than 60 percent in the 12th grade and about 25 and 10 percent in the 11th and 10th grades, respectively). Second, it is more difficult for 12th-graders to simply not show up for an exam, and to make up for this by “retaking” it later, as described in Section II.B. Students in the 10th or 11th grade are more likely to not take an exam, or to not do their best, when they are affected by a terror attack, simply because they have the option of retaking the exam in the next examination period.26 Third, the probability of students switching schools during the 12th grade, and particularly during the matriculation examination period, which is only several weeks long, is negligible. Finally, both academic performance and students’ reactions to terror may vary with age and over time, complicating efforts to compare within or between students.
C. Summary Statistics
The final data set was generated by combining the exam-level database with the data on terror incidents by exam date and examinee’s area.27 Panel A in Table 1 presents summary statistics for the exam-level data in the sample. The sample consists of 1,950,618 matriculation examinations taken by 248,033 students at 420 schools in 37 areas throughout Israel between 2001 and 2005.28 The average score on the matriculation exams is 72.3, and the average internal school score is 78.5. The average number of credit units conferred per exam is 1.76. About 15 percent of the exams received failing scores (under 55). A small number of exams (6.5 percent) were retake tests (where the student had taken the same exam in a previous exam period), and 4.8 percent of the exams were a first exam preceding a retake (where the student retook the same exam in a future exam period).29 Finally, in terms of the spacing between exams, in any given matriculation examination period, students typically have at least three days between one exam and their next one, with an average of about eight days between successive exams.
Descriptive Statistics: Exam and Student Variables
Panel B in Table 1 presents summary statistics at the student level. Seventy-five percent of the students earned their matriculation certificate, with an average matriculation composite score of 77.2 and an average of 24.7 credit units. The students in the sample took 13 matriculation exams on average, of which eight were taken in the 12th grade, and about half of the total (6.7 exams on average) were taken during the summer examination period of 12th grade. In terms of broader demographics, 17 percent of the students in the sample studied in religious schools, 81 percent were born in Israel, and parents’ average schooling was 12 years.
It is possible to establish the intensity of local terror-related violence in the days before students take a given exam using information on the exam dates and students’ locations. Table 2 presents summary statistics for the number of fatalities from terror attacks in a given area for time windows running from the day of the exam, but before the exam started, to 15 days before the exam. Columns 1 and 2 report the percentages of exams and students, respectively, with a positive number of terror fatalities in the relevant time window before an exam. Columns 3, 4, 5, and 6 report the means, standard deviations, and extreme values for the number of fatalities in the relevant time window conditional on a positive number of fatalities. Although terror attacks during the sample period are not rare, a very small proportion of the matriculation examinations were affected by terror in the local area during the preceding days. For example, only 1.3 percent of the examinations were exposed to a terror attack in the four-day window before the exam.
Descriptive Statistics: Terror Fatalities for Different Time Windows Preceding an Exam
IV. The Effect of Terror on Exam Performance
A. Empirical Specification
My empirical strategy to identify the causal effect of fatal terror attacks shortly before an exam on students’ performance takes advantage of two important features of the data. The first is the panel structure of the test-score data, given that Israeli students take multiple exams over a period of several weeks at the end of their final school year. The second is the temporal and geographical variation in the frequency and intensity of terror attacks during the Second Intifada. Based on these two features, I am able to compare performance of the same student in different exams, taking into account the variation in exposure to terror attacks occurring in the student’s area before the exam.
To consider different levels of severity in exposure to terror attacks, for each exam taken by a given student, I construct an intensity measure of exposure. This intensity measure is based on the cumulative number of fatalities from terror attacks in the student’s area in a specific time window before the exam (the variable in Table 2, described in the previous section). I then exploit the quasi-random variation in that measure within the student and exam period to estimate the causal impact of terror attacks on student performance. Furthermore, I also use different time windows before the exam to explore the magnitude and persistence of the effects of exposure to terror. Thus, I am able to show how the effect of exposure to terror attacks on performance evolves over time.
In the base specification, I use a given student’s standardized score in a particular matriculation exam as the dependent variable and regress it on the measure of the terror intensity in the student’s area during an n-day window preceding the exam (including the day of the exam but before the exam started). In particular, I estimate the following specification:
1where yiqrt denotes standardized exam performance for student i taking an exam in subject q in area r on date t, γit are student-by-exam-period fixed effects, and αq are subject fixed effects. Fatalitiesiqr,t–j is the number of Israeli fatalities from terror attacks in area r of student i on day t – j. When no terror attack (or a terror attack without fatalities) occurred in area r on day t – j, Fatalitiesiqr,t–j is equal to zero. The structure of the data allows to explore the persistence of the direct effect by changing n to different thresholds. If n = 1, then β measures the immediate effect on scholastic performance of terror intensity (number of fatalities) on the day of the exam or the day preceding. If n = 5, then β measures the effect of terror intensity on the exam day or during the five days preceding. Ziqt is a vector of exogenous explanatory variables related to the exam, including fixed effects for proficiency level (both for the credit units earned by the student in the exam paper and the total credit units earned by that student in the subject), and an indicator for a retake exam.30
A critical assumption for inferring a causal effect of β is that unobserved determinants of students’ test scores within an exam period are uncorrelated with variation in terror attack intensity. While exposure to a terror attack is at the area level, student-by-exam-period fixed effects ensure that I am comparing the performance of the same student across exams within the same testing window, where some exams may be taken shortly after a fatal terror attack and others not, but the student’s ability, schooling quality, previous exposure to terror, and other personal factors, such as conditions at home, are fixed. Also, since neither students nor schools have control over the timing of tests, the occurrence of a terror attack before different exam dates is random. Hence, it is reasonable to assume that all potential confounders are orthogonal to terror attack timing. To account for persistent differences by subjects and proficiency levels, subject fixed effects and variables relating to specific exams are included.31
B. The Time Window for the Effect of Terror on Exam Performance
The main parameter of interest, β in Equation 1, measures the marginal effect of an additional fatality from a terror attack in the student’s area within an n-day window preceding an exam on the student’s performance in that exam. To identify the specific timing of the impact (n), Figure 3 plots the estimated coefficients of β with 90 percent confidence intervals for different values of n from zero to 15, where n = 0 refers to the day of the exam but before the exam began. The point estimates in Figure 3 show that terror fatalities have the greatest effect on student performance when the terror incident occurs on the day of the exam. Each additional fatality from a terror attack just before the exam, on the same day, reduces performance by 0.016 standard deviations (SE = 0.007). The effect remains significantly negative, though smaller, when the window for the effect is expanded up to a week before the exam. For windows of more than a week the effect is very small and statistically not different from zero.
Effect of Cumulative Terror Fatalities on Matriculation Exam Scores by Time Windows
Notes: The figure plots the coefficients and their 90 percent confidence intervals from different specifications of Equation 1. Standardized matriculation exam scores were regressed on the number of Israeli fatalities from a terror attack in the student’s area within an n-day window preceding the exam, with each specification allowing for a different value of n from zero to 15, where n = 0 refers to the day of the exam but before the exam started. The regressions also include student-by-exam-period fixed effects, fixed effects for exam subject, fixed effects for proficiency level (for both the subject and the specific exam), and an indicator for a retake exam. Standard errors are clustered at the area and date level.
Following the results in Figure 3, I estimate the model by allowing for lag and lead effects of terror attacks on test performance. Specifically, I estimate separately the effect of fatalities from terror attacks in the student’s area on each day from eight days before the exam up to eight days after the exam, as in the following specification:
2The estimates of the lag dummies help us to identify the specific timing of the impact, and as a result to choose the appropriate n for Equation 1. The estimates of the lead dummies can serve as a placebo test for the effect of future fatalities each day for up to eight days following the exam. Figure 4 reports the estimated coefficient of each β in Equation 2 with 90 percent confidence intervals. The estimated coefficients for the effect of future fatalities on test scores are not statistically different from zero, and the estimated coefficients for the effect of past fatalities on test scores show a negative and significant effect for up to four days before the exam. Online Appendix Figure A1 provides an additional placebo test by replicating Figure 3 but regressing the outcome variable on future rather than past fatalities from terror attacks. The point estimates in Online Appendix Figure A1 are very small and not statistically different from zero, showing that fatalities in the 15 days following the exam are not significantly associated with test scores. This analysis further supports the current findings by which the association between (prior) fatalities and test scores represents a real relationship and not a statistical or methodological artifact.
Effect of Daily Terror Fatalities on Matriculation Score in the Eight Days Before and After the Exam
Notes: The figure plots βj coefficients and their 90 percent confidence intervals from a regression of Equation 2, where j runs from −8 to 8. The dependent variable is the standardized matriculation exam score, and the independent variable is the number of Israeli fatalities from a terror attack in the student’s area within the eight days prior to the exam, on the day of the exam, and within eight days after the exam, allowing for a separate effect on each day. The regression also includes student-by-exam-period fixed effects, fixed effects for exam subject, fixed effects for proficiency level (for both the subject and the specific exam), and an indicator for a retake exam. Standard errors are clustered at the area and date level.
As in Figure 3, the effect of terror is most negative and significant when an attack occurs in the student’s area on the exam day but before the exam begins. Moreover, the marginal effect of each additional fatality from a terror attack in the student’s area one day or four days before the exam is also negative and statistically significant, although smaller (a marginal effect of −0.006 standard deviations in both). The marginal effect for two or three days before the exam is also negative, but smaller and less precise.32 The marginal effect becomes very small and nonsignificant from the fifth day onward.33 The estimated coefficients for the effect of future fatalities on test scores are not statistically different from zero.
Figures 3 and 4 show clearly that a fatal terror attack within a four-day window preceding an exam has a significant effect on students’ performance, but a fatal terror attack five days or more before the exam has no effect. Therefore, the analysis on the short-term effect of terror on student performance focuses on the four-day window preceding an exam. However, the result that the effect is transitory and not persistent is also important. In Sections IV.E. and IV.F, I discuss further the transitory nature of this finding.
C. Identification Issues
My analysis of the time window for the effect suggests that the affected exams are those taken in the four days following a fatal terror attack in the examinee’s area. Hence, these can be referred to as treated (exposed) exams and students who took these exams as treated (exposed) students. In order to identify and quantify the causal short-term effect of terror on test performance by estimation of Equation 1 with n = 4, a key assumption is that the occurrence of a terror attack within four days before an exam is orthogonal to other factors that affect student performance. Using within-student variation in exposure to terrorism helps to address most concerns that could confound the causal relationship between terror attacks and exam performance, as discussed in Section IV.A. But if there is selection in exam-taking due to terror attacks, the results could still be biased. In this section I address this concern by presenting a balancing test and testing for selection in exam-taking.
1. Balancing test
Online Appendix Tables A3 and A4 display means and standard deviations by exposure status for student characteristics, matriculation variables, and exam variables. The tables also report the differences between the means, providing a balance test that allows us to compare the treated and untreated students. Online Appendix Table A3 reports means, standard deviations, and differences for matriculation variables and student characteristics at the student level, while Online Appendix Table A4 reports means, standard deviations, and differences for exam variables at the exam level. In both tables, Column 1 reports means and standard deviations for treated (exposed) students, and Column 2 reports the same for untreated (unexposed) students. Both tables also report the differences (Column 3) and the differences conditional on year and area fixed effects (Column 4) between the means with standard errors clustered by year (exam date in Online Appendix Table A4) and area.
Students who were exposed to terror attacks shortly before a matriculation exam have better exam and matriculation outcomes, even when controlling for time and area fixed effects. However, conditional differences (while controlling for time and area fixed effects) for students’ background characteristics and exam characteristics are not statistically significant (controlling for school fixed effects instead of area fixed effects does not change the results). The fact that the sample of exposed students includes a higher proportion of higher-achieving students compared to unexposed students can be explained either by mechanical correlation—higher-achieving students take more demanding matriculation programs that include more credit units, and hence take more exams; therefore, they are more likely to experience a terror attack before an exam—or by selection in test-taking. I discuss this in more detail in Section IV.C.2.
In Columns 5 and 6 in Online Appendix Table A4, I stratify the sample of exposed students by exposure status of their exams. The difference and the conditional difference between these two columns and their standard errors clustered by exam date and area are reported in Columns 7 and 8, respectively. Terror-affected exams are not statistically different from other exams in their characteristics, except in their subject credit units and days between subsequent exams. Subjects that confer more credit units (or proficiency levels which confer more credit units in a given subject) require more exams within the same day. Hence, an exam is more likely to be affected by terror if it is in a subject (or at a level) conferring more credit units. Interestingly, while exams taken by exposed students are usually spaced closer together (that is, with fewer days between successive exams), their terror-affected exams follow the preceding exam by, on average, an additional day and a half compared to unaffected exams, suggesting that if more space between exams improves performance, the present estimates are downwardly biased.34 It is also worth noting that while exposed students appear to retake exams in the next examination period at a marginally higher rate, a closer look at the exams of exposed students reveals that this is not driven by terror-affected exams.
2. Selection in test-taking
The main concern over causal identification that cannot be addressed by the student fixed effects strategy is selection in test-taking. That is, if terror attacks before scheduled exams cause some students to not show up for the exam, the exposure variable is no longer exogenous. Specifically, students who did not show up for an exam due to a terror attack would not be observed in the data as exposed students, even though they were affected by the “treatment” of terror exposure. In that case the estimated effect will only apply to the sample of students who took all their exams, and who may be very different from students who did not take all their exams. Although Section II.B described why students are unlikely to not show up for exams, to address this concern I estimate the effect of terror exposure on the rate of test-takers at the school–exam type level and test whether terror exposure during the exam period, during the spring semester, or during the full academic year affects the number of exams taken at the student level.
Online Appendix Figure A2 plots the estimated coefficients of β and their 90 percent confidence intervals from a regression similar to Equation 1, with the rate of test-takers within schools by subject, exam proficiency level, and date as the dependent variables. School-by-exam period, subject, and exam proficiency level fixed effects serve as additional controls. The results in Online Appendix Figure A2 show very small and imprecise correlations between the number of Israeli fatalities in the school’s area within different time windows before each exam and the rate of test-taking within the school. In fact, some coefficients are positive, although not statistically significant. This evidence suggests that fatal terror attacks do not increase attrition in exam-taking, and if there is any effect it may be in the opposite direction.
As an additional test, I use student-level data and exploit the within-school variation in terror exposure during the summer exam period (May and June), during the spring semester (February–June) or during the full academic year (September–June) to estimate whether it affects the number of exams taken by students. In particular, I estimate the effect of the number of fatalities from terror attacks during each of the relevant periods in the student’s area of residence on both the total number of exams taken by a given student and the number of exams taken during the summer exam period (when most exams are scheduled), while controlling for school and year fixed effects, as well as for student characteristics (parents’ education, number of siblings, a gender dummy, an indicator for being born in Israel, and a set of indicators for ethnicity). Online Appendix Table A5 reports the results, where the dependent variable in Columns 1–3 is the number of exams taken in the summer examination period, and the dependent variable in Columns 4–6 is the number of exams taken overall in the 12th grade. The estimated coefficients in Online Appendix Table A5 are very small and insignificant, suggesting that terror attacks do not affect the number of exams taken. These results, together with the findings of low exam retake rates in Section IV.C.1, provide additional support for the absence of selection in test-taking as a result of terror attacks before exams.
D. Results
1. Main results
Table 3 presents the estimated results for β for variations of Equation 1 where n = 4 (following the results in Section IV.B). Column 1 reports the student-by-exam-period estimate of β from Equation 1. It indicates that an increase of one fatality from a terror attack in a student’s area within four days before an exam is associated with a decline of 0.005 standard deviations (SE = 0.0015) in the student’s score on the exam.35 Columns 2–6 in Table 3 present estimates of β when the student-by-exam-period fixed effects are replaced by other unit fixed effects. Allowing for student and exam period fixed effects, or school-by-exam-period fixed effects, or area-by-exam-period fixed effects all produce similar negative point estimates.36 The robustness of the result to excluding student fixed effects in Columns 3 and 4 provides support for the assumption that any confounding factors at the student level are orthogonal to terror attack exposure. Moreover, the standard errors do not appear to be sensitive to alternative clustering. The standard errors in Table 3 are clustered at the area and exam-date level. However, the results are robust to alternative clustering, such as by school, by school and date, or by area and date.37
Short-Term Effect of Terror Fatalities on Matriculation Scores
The estimated effects are roughly equivalent in magnitude to the impacts of other score-shock instruments characterized by transitory exposure studied in previous research, such as excessively hot temperatures or above-normal pollution levels. For example, a terror attack causing at least 20 fatalities (the 90th percentile of exposure in my sample conditional on exposure) in a given area is associated with a reduction of at least 0.1 standard deviations in test scores on affected exams, compared to exams that did not follow a fatal terror attack in the area.38 Michaelsen and Salardi (2020) and Sharkey et al. (2014) present similar-sized effects for exposure to crime among elementary school students who live or attend school very close to the crime location.
To further explore whether the negative effect in the four days before the exam differs along the test score distribution, Figure 5 reports estimated results and 90 percent confidence intervals of β from specifications similar to Equation 1, with n = 4, and an indicator for receiving a higher score on the exam at different thresholds as the dependent variable. The x-axis of Figure 5 reports the exam score thresholds from 55, the minimum score needed to pass the exam, to 95, the top decile. As shown in Figure 5, a fatal terror attack before an exam mainly affects the upper distribution of test scores. An additional terror fatality in a student’s area in the four days preceding an exam reduces the probability of receiving a score higher than 85, the upper quarter of the exam score distribution in the sample, by 0.17 percentage points (SE = 0.0009). On the other hand, a fatal terror attack before an exam does not decrease the likelihood of passing the exam. This is in contrast to the finding by Brück, Di Maio, and Miaari (2019) that the number of Palestinian fatalities in the West Bank during a given academic year reduces the probability of Palestinian high school students passing an exam, but has no effect on students in the upper tail of the distribution. However, the mechanisms for the effect observed by Brück, Di Maio, and Miaari (2019) are different from this study.
Effect of Terror Fatalities on the Probability of Receiving a Given Score on a Matriculation Exam
Notes: The figure plots the coefficients and 90 percent confidence intervals of β from specifications similar to Equation 1, with n = 4, where the dependent variable is not the standardized test score, but an indicator for receiving a test score higher than a certain threshold for different values of thresholds. Score thresholds run from 55 (the minimum passing score) to 95. The regressions also include student-by-exam-period fixed effects, fixed effects for exam subject, fixed effects for proficiency level (for both the subject and the exam), and an indicator for a retake exam. Standard errors are clustered at the area and date level.
2. Nonlinear effect
The baseline specification assumes a linear effect of the number of fatalities from terror attacks on test scores. However, the impact of terror may vary with its intensity. Table 4 reports estimates from different specifications similar to Equation 1 with alternative nonlinear effects. Columns 1 and 2 report the results from specifications that allow for a nonlinear effect in Equation 1 by also including a quadratic term for the level of terror or an indicator for any attack, respectively. Columns 3, 4, and 5 report the results from specifications that allow for a nonlinear effect in Equation 1 by replacing the fatality-count variable with a dummy indicating whether terror attacks causing any fatalities, above five fatalities, or above ten fatalities, respectively, occurred in the examinee’s area in the four-day window before an exam. Column 6 reports the results from specifications that allow for a nonlinear effect in Equation 1 by replacing the fatality-count variable with three dummy variables which divide the number of terror fatalities into different bins: (i) one to five fatalities, (ii) six to ten fatalities, and (iii) more than ten fatalities from a terror attack in the student’s area within four days of an exam.
Nonlinear Short-Term Effect of Terror Fatalities on Matriculation Scores
The estimated effects of terror intensity from the nonlinear specifications in Columns 1 and 2 are similar to the estimated effects from the linear specification reported in Table 3. The results in Columns 3–5 show that only a terror attack causing more than five or ten fatalities has a negative and statistically significant effect on test scores, reducing test scores by 0.083 standard deviations (SE = 0.020) and 0.103 standard deviations (SE = 0.031), respectively. The results in Column 6 support this pattern. While the three coefficients are significantly different from each other (the p-value for the differences between the coefficients of the three dummy variables is 0.005), it appears that the effect of six to ten fatalities does not significantly differ from the effect of more than ten fatalities (the p-value for the differences between the coefficients of these two dummy variables is 0.349).
Figure 6 provides extended results for the specifications in Columns 3–5 in Table 4. In particular, each point in the figure is the point estimate from a specification of Equation 1 that replaces the fatality-count variable with a dummy variable indicating whether terror attacks causing a number of fatalities equal to or above the relevant cutoff occurred in the examinee’s area in the four-day window before an exam. According to Figure 6, and similar to the results in Table 4, as the number of fatalities increases the effect becomes more negative, up to a decline of 0.1 standard deviations in test scores, and statistically significant. These results suggest that high-intensity terror, rather than terror in and of itself, is responsible for the previously estimated effects.39
Nonlinear Effect of Terror Fatalities on Matriculation Score by Intensity
Notes: The figure plots the coefficients and 90 percent confidence intervals from regressions of standardized matriculation exam scores on a dummy variable that indicates whether a terror attack causing more than a given number of Israeli fatalities occurred in the examinee’s area within four days of the exam. The regressions also include student-by-exam-period fixed effects, fixed effects for exam subject, fixed effects for proficiency level (for both the subject and the exam), and an indicator for a retake exam. Standard errors are clustered at the area and date level.
3. The role of distance
A main assumption in the empirical strategy is that students are sensitive to terror fatalities only in their area of residence. To explore the role of geographical distance, I expand Equation 1 by adding three additional variables relating to terror fatalities in the four days preceding the exam outside the student’s area, as follows: (i) the number of terror fatalities in the student’s subdistrict but outside their area, (ii) the number of terror fatalities in the student’s district but outside their subdistrict, and (iii) the number of terror fatalities in Israel but outside the student’s district. By comparing the impact of attacks in near and farther away areas, I can shed light on the mechanisms driving their impact. This will be discussed further in Section IV.F.
Table 5 reports the results. In Columns 1 and 2, the specification reports fatalities in the rest of the subdistrict and the rest of the district, not including those in the student’s own area or subdistrict, respectively. The other columns add fatalities in the rest of the country, with Column 4 also including Israeli civilian fatalities in the West Bank and Gaza Strip. p-values for the equality of the various coefficients are reported as well. The results in Table 5 show that the effect of terror fatalities on test scores decreases with distance. None of the coefficients for fatalities outside the student’s area, whether in the subdistrict, district, or country as a whole, are statistically significant. Therefore, the results suggest that physical distance mitigates the effect of terror on academic achievement. Moreover, accounting for potential effects from other areas is immaterial to the results of the main analysis (the estimates in Row 1 resemble those in Table 3).
Short-Term Effect of Terror Fatalities on Matriculation Scores by Distance
The results in Table 5 also suggest that the estimates reported in the main analysis in Table 3 can be considered as a lower bound, since the area units are the smallest geographical units that I am able to identify due to privacy limitations, and the denser areas, where most of the attacks took place, are about 30–50 km2. Hence, for a smaller radius of exposure we could potentially see a larger effect. However, it makes sense to examine the impact of terror attacks on a wide geographical range since terror attacks receive nationwide media attention.40
4. Heterogeneity effect
To examine heterogeneity in the short-term effect of terror by student characteristics and exam subjects, Tables 6 and 7 report the results from estimations of Equation 1 with n = 4, for samples stratified by various student characteristics and exam subject types. Columns 1 and 2 in Table 6 report the estimated results of Equation 1 separately for males and females, respectively. In Columns 3 and 4 of Table 6 the samples are stratified by parents’ education, operationalized by the sum of both parents’ years of schooling, with high education levels defined as a sum exceeding 24 (the median value in the sample).41 Columns 5 and 6 in Table 6 report the results separately for secular and religious schools, respectively, and in Columns 7 and 8 the samples are stratified by students’ origins (native-born or immigrant). The p-values for coefficient equality within each group are reported as well. The estimated negative effects are slightly larger among females, students who have more educated parents, secular students, and Israeli-born students, though none of these differences are statistically significant. Previous studies that have tested for differential effects of community-wide violent events by gender have also found no such effects (Sharkey et al. 2014; Brück, Di Maio, and Miaari 2019; Cabral et al. 2021).
Heterogeneity in the Effect of Terror Fatalities on Matriculation Scores Across Subpopulations
Table 7 reports the results stratified by exam subject type in the same format as Table 6. Columns 1 and 2 report the estimates for exams in STEM and non-STEM subjects, respectively. Columns 3 and 4 stratify the sample by exams in mandatory subjects and exams in elective subjects. Both sets of results in Table 7 show a negative and significant effect for all subject types, with no statistical difference between STEM and non-STEM subjects or between mandatory and elective subjects. Findings in previous studies regarding heterogeneous effects of community-wide violent events by exam subject are not conclusive. Monteiro and Rocha (2017); Gershenson and Tekin (2018); Brück, Di Maio, and Miaari (2019); and Michaelsen and Salardi (2020) all find that math test scores are more sensitive to conditions in the learning environment affected by community-wide traumatic events. However, Sharkey et al. (2014), who focus on the immediate effects of extreme violence, find significant negative effects on test scores in language, but not in math, and only among elementary school students and not middle school students.
Heterogeneity in the Effect of Terror Fatalities on Matriculation Scores by Exam Subject Type
E. Robustness Checks for the Time Window
The results in Section IV.B show that the effect of terror attacks on students’ performance is concentrated in the four days before the exam, suggesting that this effect is short-lived. This raises concerns that, as terrorism was an ongoing threat in Israel during the sample period, students simply had become habituated to it. To reduce such concerns, Online Appendix Figure A3 replicates Figure 3 while restricting the sample to include only students tested in 2001, the first year of the Second Intifada, when students were less used to experiencing terror attacks. The patterns observed in this figure are similar to the patterns in Figure 3, suggesting that even when students are less used to experiencing violence, the negative effect on test scores dissipates after several days.
Another limitation is that as the analysis design exploits variation within students, it may be able to shed light only on very short-term effects. To estimate the effect of exposure to terror attacks over longer time periods, I exploit the variation in terror intensity during the spring semester (February–June) over time within schools to compare students’ performance on the same exam (that is, on exams covering the same material at the same proficiency level) who faced different levels of terror intensity in the months before the exam. Specifically, I estimate a model similar to Equation 1 where the student-by-exam-period fixed effects are replaced by exam fixed effects, school fixed effects, and year fixed effects for different values of n that allow for wider windows of days for exposure (that is, 30, 60, or 90 days). Since I focus on exposure during the spring semester, the sample is restricted to include only exams taken in the summer examination period.
Online Appendix Table A6 presents the results, where Panel A reports the results for the linear effect of terror intensity exposure (that is, the independent variable is the number of fatalities from terror attacks within the relevant time window), and Panel B reports the results for the nonlinear effect of terror intensity exposure (that is, the independent variable is an indicator for whether a terror attack causing more than five fatalities ever occurred within the relevant time window). For comparison with the analysis in Table 3, Columns 1 and 2 show the results for the effect of terror exposure in the four-day window before the exam. Columns 3–8 show the results for the effect of terror exposure in a 30-day window, 60-day window, and 90-day window before the exam. Specifications in odd-numbered columns include only exam, school, and year fixed effects, while specifications in even-numbered columns also include controls for student characteristics, subject fixed effects, and fixed effects for the number of exams taken by the student in the exam period but before the current exam. The results reported in Online Appendix Table A6 suggest that the short-lived effect is robust. While the estimated effect for exposure in the four-day window before the exam is similar to the student-by-exam-period estimator reported in Table 3, exposure to terror within a one-, two-, or three-month period before the exam has no significant effects on performance.
The identification assumption for a causal interpretation of these findings is that, conditional on school and year fixed effects, variation in terror intensity in the months before exams is orthogonal to any other determinant that affects student performance. To test the validity of this assumption, Online Appendix Table A7 reports the coefficients from regressing the intensity of terror (measured either by the number of fatalities or by whether any attack causing more than five fatalities occurred) in the student’s area during the summer exam period (May and June) or the spring semester (February–June) on students’ characteristics, along with year fixed effects and school fixed effects. Online Appendix Table A7 presents the small and insignificant coefficients, which together with the results reported in Online Appendix Table A5, indicate that students who were exposed to terror during the months before exams do not differ from those who were not exposed in either their observable characteristics or the number of exams taken. We cannot exclude the possibility that other time-varying unobservables may be correlated both with exam performance and the intensity of terror attacks. However, because the inclusion of the student and subject characteristics has little effect on the coefficients estimated in Online Appendix Table A6, it is unlikely that the results are driven by other unobserved variables.
F. Discussion of Possible Mechanisms
Acute stress is likely to be an important mechanism in the short-term effect of terror on students’ performance. Psychological studies show that rates of “substantial stress” among the general population are extremely high in the first few days after a terrorist incident, but are already in decline within the first two weeks, and the proportion experiencing substantial stress is negatively associated with distance from the attacks (Schuster et al. 2001; Vázquez, Pérez-Sales, and Matt 2006; Whalley and Brewin 2007; Tsai and Venkataramani 2015). Similar patterns were also found in economic studies on individuals’ self-reported well-being after a terror attack (Clark, Doyle, and Stancanelli 2020; Bryson and MacKerron 2018) and in studies on how community violence affects children’s outcomes that are more directly related to psychological stress, such as sleep, cortisol levels, attention, and impulse control (Sharkey 2010; Sharkey et al. 2012; Heissel et al. 2018). The finding in this study that terror attacks have a significant negative impact on student performance, but that the effect is transitory (concentrated in exams occurring within four days after a terror attack) and decreases with physical distance, are consistent with these patterns.
Of course, since I lack access to important data, such as direct measures of student stress levels before exams, other mechanisms could also be at play. Nevertheless, my short-term analysis, which exploits variation in exposure within students across days within the same testing window, allows me to rule out as possible mechanisms determinants of learning that might correlate with terror attacks over time but that are fixed within students and exam period (such as school quality, absenteeism, economic conditions at home, individual educational investment decisions, and outmigration), as well as any concern over selection in test-taking.42
Another mechanism that could be relevant is the effect of terror on the student’s environment, manifested in changed behavior among parents, teachers, and examiners, which could represent indirect psychological effects of terror on students. However, while I cannot separate out the direct effects of terror on students from effects transmitted through the student’s environment, several factors reduce the concern that such indirect effects play an important role in my findings. First, since my analysis focuses on final-year high school students, it is less likely that the degree of exposure and its effect depend on the parents. Second, during exam periods, students only come to school for exams, so they have few encounters with teachers in the days before an exam and thus are less likely to be affected by their response. Third, while exam supervisors may act differently after a terror attack, to the extent that this is the case it seems more likely that supervisors would help students who seem to be under special stress. This would work against the current finding of a negative effect on scores.43
Finally, since terrorism receives massive media coverage in Israel, a possible channel for the effect on students may be more commotion in the media space, which could distract students from studying before exams. However, while this is indeed so, it is not clear to what extent the effects of media coverage can be separated from the effects of stress, because exposure to media coverage of a tragedy has itself been shown to generate symptoms of anxiety and distress (Slone 2000; Schlenger et al. 2002; Holman, Garfin, and Silver 2014). Moreover, the media in Israel is very centralized, and students throughout the country are exposed to the same media coverage after any terror attack. Hence, the decline in the observed effect with physical distance from the site of an attack suggests that it is not media coverage per se that leads to fear and distress, but exposure to news about attacks and fatalities in a familiar place, where one knows people who were or could have been hurt.
V. Long-Term Consequences of the Short-Term Effect of Terror Exposure
The findings so far suggest that only attacks that occur shortly before an exam impair exam performance among tested students. However, poor performance on matriculation exams may have long-term implications, as described in Section II.B.44 In this section, I use repeated cross sections of student-level data to examine whether the negative short-run effects of terror exposure on performance observed in Section IV could potentially result in persistent consequences. I do this by focusing on two end-of-high-school outcomes with meaningful long-term implications, as described in Section II.B: the probability of obtaining a matriculation certificate and the quality of the certificate earned (measured by the composite score).
A. Empirical Specification
To evaluate whether the short-run negative effects reported above result in persistent consequences, I exploit the variation in terror exposure shortly before exams among students at the same school. Since students are required to take examinations in several elective subjects, and the dates of exams vary by subject and cohort, students at the same school may experience different levels of terror exposure before exams due to being in different cohorts or choosing different sets of elective subjects. Also, as previously discussed, students who choose more demanding matriculation programs, and who as a result must take more exams, are more likely to experience a terror attack before an exam. Hence, I assume that conditional on the number of exams taken, within-school variation in exposure to terror attacks shortly before exams is unlikely to be correlated with potential student outcomes. That is, the number of exams taken is exogenous to terror exposure in the short run. This is a plausible assumption based on the discussion in Section IV.C.2. Therefore, I estimate the following specification:
3where Yirsy is the education outcome of interest of student i in school s in area r who completed 12th grade in year y. For the probability of obtaining a matriculation certificate, Yirsy is a dummy variable that takes the value one if student i obtained a matriculation certificate. For the quality of the certificate earned, Yirsy is the matriculation composite score of student i on a 1–100 scale. δy and θs are year and school fixed effects, respectively, where the latter control for unobservable time-invariant differences across schools, and the former account for time-varying differences at the national level that could influence both students’ performance and the severity and frequency of terror attacks. The variable of interest is Exposeirsy, which measures the short-term exposure to terror attacks of student i in school s in area r in year y. Following the result that high-intensity terror attacks drive the short-term effect, I estimate Equation 3 using two different measures for Exposeirsy: (i) the average intensity of terror attacks to which student i was exposed across all his or her matriculation exams in year y, that is, in the 12th grade,45 and (ii) the number of exams exposed to high-intensity terror attacks (attacks causing more than five fatalities in the four-day window before the exam) of student i in year y. Examsiy is a vector of fixed effects for the total number of exams taken in 12th grade by student i who completed 12th grade in year y. It accounts for the mechanical correlation between exposure to terror attacks shortly before an exam and the number of exams taken. Xi is a vector of student-level controls, including parents’ years of education, number of siblings, a gender indicator, a dummy for being born in Israel, and a set of indicators for ethnicity. Standard errors are clustered at the school level.
Another point that needs to be considered is the effect of sustained exposure to terror throughout the academic year. In the short-term analysis, the student-by-exam-period estimator eliminates the effects of sustained terrorism and allows us to capture idiosyncratic shocks from terror attacks shortly before an exam. However, when evaluating end-of-high-school outcomes, this sustained effect must also be taken into account. Moreover, sustained exposure may also affect the ramifications of any short-term exposure. Hence, in some specifications I also include Fatalitiesry, which measures terror exposure (by counting the number of fatalities from terror attacks) during either the spring semester or the full school year to account for cumulative previous exposure.
The differences between exposed and unexposed students in the number of exams taken documented in Section IV.C.1 suggest that, overall, exposed students are more likely to take more demanding matriculation programs. This creates an imbalance between exposed and unexposed students that may bias the results. Therefore, as a robustness check, I also estimate Equation 3 using a balanced matched sample. I employ a coarsened exact matching (CEM) procedure (Blackwell et al. 2009) to balance between exposed and unexposed students by matching each student who was exposed to a terror attack in the four-day window before an exam to a control student from the same school who took the same number of exams during the 12th grade. I found matches for 95 percent of the exposed students (= 16,241 out of 17,023), and the final sample contained 32,482 students. Online Appendix Table A8 reports the balancing test for the matched sample, where Column 1 reports means and standard deviations for exposed students, and Column 2 reports the same for matched unexposed students. Differences are reported in Column 3, and differences conditional on year, school, and number of exams fixed effects are reported in Column 4. The table shows that exposed students and the matched unexposed students are similar in their characteristics.
B. Results
Table 8 reports the estimated results for the effect of terror attacks before matriculation exams on end-of-high-school outcomes with long-term implications based on Equation 3. The dependent variable in Columns 1–3 is the indicator for obtaining a matriculation diploma, and the dependent variable in Columns 4–6 is the matriculation composite score. Panel A reports the results where Exposeirsy captures the intensity of the student’s terror exposure as an average of fatalities from all attacks in the four days before any exam, and Panel B reports the results where Exposeirsy is defined as the number of exams exposed to attacks causing more than five fatalities. Columns 2 and 4 also report the coefficients for the effects of number of fatalities during the spring semester, and Columns 3 and 6 also report the coefficients for the effects of number of fatalities during the whole school year.
Effect of Short-Term Exposure to Terror on Matriculation Outcomes with Long-Term Implications
The results show that exposure to fatal terror attacks before matriculation exams has no significant effect on the likelihood of obtaining a matriculation certificate. The effect on the matriculation composite score is marginally significantly negative, but its economic magnitude is small. For example, according to Columns 4–6, Panel B, each additional exam that was exposed to a terror attack causing more than five fatalities is associated with about a 0.45-point reduction in a student’s composite score. But since most students (80 percent of the exposed students) have only one such exposed exam, this is a small effect considering that the composite score standard deviation is almost 11 points. Also, the estimated coefficients are not sensitive to the inclusion of previous cumulative exposure, suggesting that the effect of additional exposure to a terror attack shortly before an exam does not vary with previous exposure to violence. Online Appendix Table A9, which reports the results of Equation 3 when using the matched sample (in the same format as Table 8), displays similar estimated effects as in Table 8.
These findings may be explained by the structure of the Israeli matriculation system, which prevents poor performance in one test from translating into a persistent effect on human capital outcomes. As described in Section II.B, the quality of the final matriculation diploma relies on multiple evaluations in different subjects, including internal and external exams. Also, the lack of an effect on the likelihood of a student obtaining a matriculation diploma is consistent with the results reported in Section IV.D.2, showing that exposure to a fatal terror attack before an exam does not increase a given student’s risk of failing, but rather tends to reduce test scores within the upper exam score distribution.46 In evaluation systems where one exam has significant weight, one episode of poor performance could have lasting effects with long-run implications. Furthermore, we would also expect to find more persistent effects in a multiple-evaluations system if the negative shocks on performance were more frequent, as in Ebenstein, Lavy, and Roth (2016) and Park (2022).47
Note also that the estimated effects of sustained exposure are small and insignificant, suggesting that cumulative exposure to fatal terror attacks has no significant effect on end-of-high-school outcomes. This is in contrast to previous studies that find negative effects of sustained exposure to violence on students’ end-of-year exam performance (for example Jarillo et al. 2016: Brück, Di Maio, and Miaari 2019). However, in those studies, the sustained exposure is associated with significant interruptions to the school routine (such as temporary school closures and absenteeism), which are not likely to occur in the Israeli context due to the stability of the education system and lower level of violence.
VI. Conclusion
This study explores the effects of terrorism on student performance in end-of-high-school exams. Using a large sample of Israeli high school matriculation exams during the Second Intifada (2001–2005), I present evidence that terror attacks shortly before an exam have a significant negative impact on student performance. The effect is driven mainly by attacks causing a relatively high number of fatalities and decreases as the physical distance between the examinee’s locality and the attack location increases. Importantly, I find that the effect on performance is transitory and concentrated on exams occurring within four days after a terror attack, which suggests resilience among students. This pattern is similar to findings in the literature regarding the transitory and local effect of community-wide traumatic events on the prevalence of stress in the general population.
My identification strategy uses within-student variation in exposure to fatal terror incidents during the Second Intifada, taking advantage of the temporal and geographical variation in terror attacks during that time. This enables me to address various selection concerns that could confound the causal relationship between terror attacks and exam performance. Moreover, the empirical setting allows me to rule out alternative explanations that are fixed by student and exam period, such as school quality, school accessibility, and student absenteeism, leaving psychological impacts on the learning process and cognitive acuity during the exam as the potential primary mechanisms. Hence, policies that aim to help students deal with exposure to these events over the short term should consider adjusting the timing of exams, when this is possible. In cases where more enduring effects are a concern, policies should focus on educational inputs that also could be affected and that are more likely to have persistent long-term negative effects.
Furthermore, the present findings also show that the negative short-lived effect on exam performance has very little impact on end-of-high-school outcomes, which suggests no permanent repercussions for students’ human capital formation. In particular, I find no association between a student’s exposure to terror attacks before exams and the probability of obtaining a matriculation certificate, and only a small and economically insignificant effect on the quality of the certificate—two outcomes that have meaningful long-term implications for access to higher education in Israel. This can be explained by the structure of the Israeli matriculation system, where a test score in one exam has limited ramifications. Policies for designing assessment systems based on multiple evaluations at different times can help to mitigate the impact of temporary stressors and distractions such as terrorism during test-taking on subsequent outcomes.
My findings can also be generalized to the effects of terror or other types of community- wide traumatic events on other activities that require cognitive acuity, such as productivity at work. However, a caveat that should be considered is that my analysis, which exploits variation in exposure across areas, only picks up the effects of excess exposure relative to a national exposure baseline. Since terrorism was an ongoing threat in Israel during the sample period, in countries that are less prone to community violence the effects of occasional events may be larger and more persistent. Also, any country-wide effects would not be captured by the current analysis. Finally, this study focuses on the impact of terror attacks on the general population. The effects are likely larger and more persistent for students who are directly impacted by a violent event, whether by witnessing it firsthand or having relatives or friends injured or killed, as shown in recent studies on mass shootings (Cabral et al. 2021; Bharadwaj et al. 2021; Levine and McKnight 2021).
Acknowledgments
The author is grateful to Josh Angrist, Itai Ater, Rema Hanna, Victor Lavy, Hani Mansour, Sandra McNally, Maria Padilla-Romo, Cecilia Peluffo, Yaniv Reingewertz, Analia Schlosser, Yannay Spitzer, Petra Todd, Margaret Triyana, Sarit Weisburd, and Assaf Zussman, along with seminar participants at UC San Diego, the Hebrew University of Jerusalem, Tel-Aviv University, Ben-Gurion University, the University of Haifa, the IDC, the NBER Economics of National Security Summer Institute Conference, the EALE Annual Conference, and the Terrorism, Economics and the Behaviour of Agents Workshop, who provided helpful comments and suggestions. The author also thanks Israel’s Ministry of Education for allowing access to restricted data through the Ministry’s online protected research lab and Eliad Trefler of the Ministry’s online protected research lab for his invaluable assistance. Margalit Samuel provided excellent research assistance. Financial support from the Henry Crown Institute of Business Research in Israel is gratefully acknowledged. The empirical analysis of this study was carried out at online protected research lab of the Israel’s Ministry of Education. Anyone interested in acquiring the data should contact the director of the Ministry online protected research lab. The data will be provided to those who go through the data request procedure. The author is willing to assist. More details on the Ministry online protected research lab and the request procedure can be found at https://cms.education.gov.il/EducationCMS/Applications/spss/dafault.htm.
Footnotes
↵1. For example, the United States has experienced an average of 334 mass shootings annually since 2014 (https://www.gunviolencearchive.org/). Europe has experienced an increasing frequency of terror attacks either by Islamist or far-right groups in the last decade (Institute for Economics & Peace 2017, 2020).
↵2. See reviews by Whalley and Brewin (2007); Brom, Pat-Horenczyk, and Baum (2011); Dimitry (2012); Lowe and Galea (2017); Travers, McDonagh, and Elklit (2018); Rowhani-Rahbar, Zatzick, and Rivara (2019); and Lee, Kim, and Kim (2020) on children’s and adolescents’ psychological reactions to community-wide traumatic events.
↵3. For example, see Burdick-Will (2013); Sharkey et al. (2014); Beland and Kim (2016); Jarillo et al. (2016); Monteiro and Rocha (2017); Gershenson and Tekin (2018); Poutvaara and Ropponen (2018); Brück, Di Maio, and Miaari (2019); and Michaelsen and Salardi (2020). Recent studies also provide evidence of lifelong consequences, including negative educational and health impacts (Rossin-Slater et al. 2020; Levine and McKnight 2021; Cabral et al. 2021; Bharadwaj et al. 2021).
↵4. Community-wide traumatic events increase the risk of symptoms of anxiety and stress, occasionally accompanied by depression, aggressive behavior, social and emotional problems, and impaired cognitive development, among children and adolescents in the broader community (Pfefferbaum et al. 2000; Hoven, Duarte, and Mandell 2003; Pat-Horenczyk et al. 2007; Schiff et al. 2007; Whalley and Brewin 2007; Al-Krenawi, Graham, and Kanat-Maymon 2009; Dimitry 2012; Duffy et al. 2015). These emotional responses may lead to poor academic performance by impairing students’ cognitive functioning and behavior, manifesting in an inability to concentrate, reductions in working memory, and difficulty sleeping (Osofsky and Osofsky 2004; Sharkey 2010; Sharkey et al. 2012; McCoy, Raver, and Sharkey 2015; Heissel et al. 2018).
↵5. See, for example, Beland and Kim (2016); Jarillo et al. (2016); Monteiro and Rocha (2017); Gershenson and Tekin (2018); Brück, Di Maio, and Miaari (2019); and Cabral et al. (2021). Several studies, which focused on developing countries characterized by ongoing or recurring violence, suggest that exposure to traumatic events can also affect educational investment decisions (Justino 2012; Koppensteiner and Menezes 2021).
↵6. The deadliest attack (at the Park Hotel in Netanya on March 27, 2002, the eve of Passover) caused 30 fatalities. During the years 2001–2005, Israel’s homicide rate (per 100,000 population), including acts of terrorists, was 3.1. By comparison, in the United States at the same time it was 5.8, and in Western Europe the average was 1.5. At the other extreme, during Mexico’s period of drug-related conflict in 2006–2011, the homicide rate was 9.2 per 100,000 (source: https://www.unodc.org/gsh/en/data.html). Israel is a relatively stable country, and even during sustained terror campaigns the government, health system, and educational services have largely remained fully functional (Dimitry 2012). End-of-high-school matriculation exams, which are scheduled long in advance, were never rescheduled.
↵7. The geographical location data I use is for Israel’s “natural regions,” which is the third level of Israel’s formal administrative units. These units range in size from 34 km2 (the Tel Aviv area, the densest natural region) to 4,820 km2 (in the south of the country, the least dense). For more details on these units, see Section III.A.
↵8. Specifications with separate student and exam period fixed effects, or school-by-exam-period fixed effects, or area-by-exam-period fixed effects produce similar results.
↵9. For a detailed chronology of the Second Intifada, see Jaeger and Paserman (2008).
↵10. Public buses were the most popular targets, accounting for one-third of all casualties, but coffee shops, restaurants, discos, and pubs were targeted as well (Becker and Rubinstein 2011).
↵11. Source: https://www.unodc.org/gsh/en/data.html.
↵12. According to Israel’s Central Bureau of Statistics (CBS), in 2004, for example, 91 percent of all high school graduates in the mainstream Jewish education system (that is, excluding Arab and Ultra-Orthodox students) took at least one matriculation exam, and 63 percent earned matriculation certificates (source: Statistical Abstract of Israel 2006, Table 8.22; https://www.cbs.gov.il/he/publications/doclib/2006/8.shnaton%20education/st08_22.pdf).
↵13. At least 16 credits must be earned through tests in seven mandatory subjects, and at least one elective subject must be studied at an advanced level that confers four or five credits. The seven core subjects are mathematics, English, Hebrew (or Arabic for Arab students), history, literature, religious studies, and civics.
↵14. Students in a given school or class may not face the same sequence of exams due to heterogeneity in the proficiency level of the exams and the selection of elective subjects.
↵15. Usually, the internal school score is the weighted average of a final year exam and the average of the student’s scores in all tests in this subject during the academic year (the weights recommended by the Ministry of Education are 30 and 70 percent, respectively). The internal school score is determined before the matriculation exam and cannot be changed afterwards.
↵16. It is usually the school’s decision as to how individual students spread out their exams over the high school years.
↵17. The summer season is the main examination period, and all matriculation exams are offered at this time. During the winter, exams are held in the compulsory subjects. There is also a special assessment period in July in which students may improve their scores only in math and English. Students take most of their exams in the summer exam period (40 percent take all exams in the summer exam period, and 50 percent take at least two-thirds of their exams in the summer exam period).
↵18. Exams are generally scheduled for one of three time slots: morning, midday, or afternoon.
↵19. Most Israelis begin a period of mandatory military service (normally three years for males and two for females) after their high school graduation. Students who wish to be retested after their military service usually take preparatory courses that require investment of money and time.
↵20. Students may choose to upgrade their matriculation certificate after high school graduation. Indeed, a quarter of 2002 matriculation examinees who were unsuccessful at graduation completed their certificate requirements by 2010 (CBS press release 2011; https://www.cbs.gov.il/he/mediarelease/DocLib/2011/217/06_11_217b.pdf). However, obtaining or upgrading certification after graduation is usually a costly process that postpones one’s access to postsecondary schooling and delays entry to the labor market. In any case, the matriculation outcomes of these examinees are not included in the data used in this study, since these examinees were not high school students at the time they took the exams.
↵21. The analysis presented in this study excludes Israeli settlements in the West Bank and Gaza Strip because compared to Israelis living within the country’s pre-1967 borders, settlers experienced a much higher level of politically motivated violence and a different type of terrorism during this period. Furthermore, some of the security forces killed in terror attacks in these areas were on duty at the time of the attack. In Online Appendix Table A2, I include the West Bank and Gaza Strip in the analysis as a robustness check.
↵22. The B’Tselem data are thought to be accurate and reliable and have been used by other researchers (for example, Jaeger and Paserman 2008; Gould and Klor 2010; Zussman 2014). The Ministry of Affairs data can be found at https://www.gov.il/en/departments/topics/palestinian-terror-and-incitement/govil-landing-page.
↵23. The definition used here is the definition of terrorism contained in Title 22 of the United States Code, Section 2656f(d): “The term terrorism means premeditated, politically motivated violence perpetrated against noncombatant targets by subnational groups or clandestine agents, usually intended to influence an audience.”
↵24. The Israel CBS divides Israel (excluding the West Bank and Gaza Strip) into six districts and 15 subdistricts. Each subdistrict is further divided into “natural regions,” smaller geographic areas defined by continuity and uniformity both in their physical structure, climate, and soil, and in the demographic, economic, and cultural traits of their populations. Hereafter in this paper the term “area” refers to these natural regions.
↵25. Recall that the credit units awarded to a student for a given exam in a given subject are not necessarily equal to the total credit units awarded the student in that subject because some subjects are tested through a series of separate exams. These exams may be taken on the same day or in different exam periods.
↵26. Online Appendix Figure A2 shows that terror attacks indeed do not affect the rate of test-takers among 12th-graders. This will be discussed further in Section IV.C.2.
↵27. Information on students’ locality of residence is limited because small settlements are unidentified. Thus, I assign examinees to geographical areas through the proxy of school location. For students who have an identified locality, only 5 percent live outside the area of their school (this percentage is relatively small because most students who live outside their school’s area attend regional schools, which are not included in the sample). However, as a robustness check, in Online Appendix Table A2 I exclude students known to live in a geographical location other than that of their school.
↵28. There are 52 natural regions in Israel, but the sample contains just 37 areas because it is limited to students who attend regular state high schools in the Jewish system. The 15 natural regions omitted from the sample are either Arab-populated regions or do not have their own schools due to an especially sparse population.
↵29. The retake indicator refers to all cases where the same student took the same exam in a previous examination period, either earlier in the 12th grade or in a previous grade (10th or 11th). The exam-preceding-retake indicator means that the exam was retaken by the same student in a later examination period. Since the sample includes only 12th-graders, and I do not observe exams taken after graduation, this measure comprises exams taken in the winter examination period and then retaken in the summer examination period, or exams taken in the summer examination period and then retaken during the special assessment period in July (only math and English).
↵30. As previously discussed, students choose their matriculation study program years before the dates of exams are determined, and shifting between exams within an examination period is not possible (shifting between examination periods is possible, but rare for 12th-graders). Moreover, in Section IV.C.2. I show that fatal terror attacks do not affect exam-taking. Hence, the credit units earned by the student are exogenous to the timing of terror attacks. The results also do not change when I control for (i) the number of days since the previous exam, (ii) whether the exam is the student’s first matriculation exam taken within the current exam period, or (iii) the number of exams taken in the current exam period before the current exam.
↵31. Recall that the average student takes eight matriculation exams during the 12th grade, most of them (6.7 on average) in the summer period. Note also that not all students in a given school or class face the same sequence of exams in a given exam period, due to heterogeneity in the proficiency level of the exams and the selection of elective subjects.
↵32. The p-value for the null hypothesis that the coefficients in Equation 2 for the day of the exam, a day before, two days before, three days before, and four days before are statistically equal (β0 = β1 = β2 = β3 = β4) is 0.133. The p-value for the null hypothesis that these coefficients are not statistically different from zero (β0 = β1 = β2 = β3 = β4 = 0) is 0.000 with an F statistic of 5.58.
↵33. The p-value for the null hypothesis that the coefficients in Equation 2 for the fifth day before the exam, the sixth day before, the seventh day before, and eight days before are not statistically different from zero (β5 = β6 = β7 = β8 = 0) is 0.87, with an F-statistic of 0.31.
↵34. A regression analysis with and without controlling for examination period and exam characteristics shows that each additional day between exams reduces test scores by 0.009 standard deviations, a statistically significant effect. However, when adding student fixed effects the coefficient becomes 0.0002, which is not statistically significant. Also, none of the main results of my analysis change when controlling for days between exams.
↵35. Online Appendix Table A2 presents a series of robustness checks for whether the main results are sensitive to the sample included in the analysis. Specifically, I estimated β while including students residing in Jewish settlements in the West Bank and Gaza Strip and counting civilian fatalities within those areas (Column 1), including regional schools attended by students who live outside the school’s area (Column 2), including only students who live in the same area as the school they attend (Column 3), and including only treated students who were exposed to fatal terror attacks in the four days preceding at least one of their exams (Column 4). All yield similar estimated coefficients as shown in Column 1 in Table 3.
↵36. The specifications in Columns 3 and 4 include controls for student characteristics (parents’ education, number of siblings, a gender dummy, an indicator for being born in Israel, and a set of indicators for ethnicity).
↵37. With alternative clustering the standard errors range between 0.0013 and 0.0019.
↵38. Ebenstein, Lavy, and Roth (2016) found a decline of roughly 0.13 standard deviations in scores on exams taken during days characterized by elevated pollution. Park (2022) found that on days characterized by very high temperatures, exam performance drops by approximately 13 percent of a standard deviation relative to days when the temperature is normal.
↵39. Reliable data on attacks causing varying numbers of injuries and attacks with no fatalities are not available. However, the finding that only “large” attacks matter suggests that attacks causing no fatalities will not have an effect.
↵40. The distance of considered exposure in my study is larger compared to recent studies that focus on exposure to homicides and that use finely geocoded crime data. Koppensteiner and Menezes (2021) consider exposure to homicides within the radius of 25 meters from the school, and Michaelsen and Salardi (2020) consider exposure within the radius of 2 km, 5 km, or 10 km from school. The distance of considered exposure in studies that focus on terror attacks or violent conflict is on a varied range. Brück, Di Maio, and Miaari (2019) uses geographic area with an average size of 8.5 km2, whereas Clark, Doyle, and Stancanelli (2020) considered individuals who live in states close to Boston as exposed to the 2013 Boston marathon bombing.
↵41. I use parents’ education as a proxy for student ability and socioeconomic status. I do not use the internal school score as a measure of student ability and invoke it to stratify the sample since this score is on a 0–100 scale and takes no account of the proficiency level at which the student is taking the subject. Thus, students taking low-proficiency matriculation exams who earn high internal school scores may in fact be less able than students taking high-proficiency exams who earn lower scores.
↵42. Terror attacks may also be correlated with environmental conditions such as traffic and the weather, which could also be endogenous to other unobserved determinants of learning and performance. For example, terror attacks may disrupt traffic, meaning students are more likely to run late, or they may correlate with weather conditions, which can also affect students’ performance (Park 2022). However, traffic could explain the effect only if the attack occurs on the day of the exam (during exam periods students only attend school when they have an exam scheduled), but I find significant effects for attacks up to four days before the exam. Regarding weather conditions, I find no correlation between the intensity of terror attacks and either temperature or humidity (results available upon request).
↵43. Supervisors of national matriculation exams are not teachers, and do not know the students they are supervising. With respect to graders, exams are graded over a long process lasting several months by independent examiners in a blind process (that is, graders do not know the names, schools, or teachers of students whose exams they are assigned). For this reason, any negative shock such as a terror attack before a particular exam is unlikely to affect the grading process.
↵44. Since it is difficult (and therefore rare) to simply retake an exam following a negative shock to performance, poor performance on a matriculation exam could have significant effects on a student’s postsecondary academic options, including reducing the chances of university admission or access to more desirable college majors. Previous studies on environmental shocks show that in the context of high-stakes exams, where test scores are used as signals of ability or potential, even transitory shocks can have persistent consequences (Ebenstein, Lavy, and Roth 2016; Park 2022).
↵45. To calculate this, I add up the number of terror fatalities occurring in the student’s area within four days prior to any exam taken by the student during the 12th grade and divide that sum by the number of all exams taken by the student during that year.
↵46. Receiving lower scores could still influence later outcomes by affecting the quality of the certificate earned, and, as a result, the type of postsecondary schooling obtained.
↵47. Ebenstein, Lavy, and Roth (2016), using similar data as the present study (Israeli students taking matriculation exams between 2000 and 2002), show that transitory disturbances to cognitive performance caused by variations in air pollution on exam days have a significant effect on long-term educational attainment and adult wages. A possible reason for these differing findings is that high levels of pollution are a frequent phenomenon in Israel (according to the Ministry of the Environment, air pollution levels measured in 2001–2002 were above the world standard target value), implying that students could be exposed to negative pollution shocks on multiple exam days. By contrast, terror exposure in my sample was infrequent (90 percent of the exposed students experienced an attack before only one exam, and the rest were exposed before just two exams).
- Received July 2021.
- Accepted September 2022.
This open access article is distributed under the terms of the CC-BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at: https://jhr.uwpress.org.
References
Jump to section
Related Articles
Cited By...
- No citing articles found.