Shocking news and cognitive performance

European Journal of Political Economy xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

European Journal of Political Economy journal homepage: www.elsevier.com/locate/ejpoleco

Shocking news and cognitive performance ⁎

Panu Poutvaaraa,b,c,d, , Olli Ropponene a

Department of Economics, LMU Munich, Schackstraße 4, 80539 Munich, Germany Ifo Institute - Leibniz Institute for Economic Research at the University of Munich, Poschingerstraße 5, 81679 Munich, Germany c CESifo, Germany d IZA, Germany e VATT Institute for Economic Research, Arkadiankatu 7, P.O. Box 1279, FI-00101, Helsinki, Finland b

A R T I C L E I N F O

ABSTRACT

JEL classification: D8 H42 I19 I21 J16

We study how shocking news affects cognitive performance. Identifying these effects makes societies more resilient by helping to adjust policy responses to reduce indirect costs of future atrocities. Our analysis is based on a school shooting that coincided with national matriculation exams, allowing a difference-in-differences analysis. We find a substantial negative effect on males: their average performance dropped by seven percent. The average performance of females was unaffected. Our findings suggest that a shocking event may call for psychological support for young people even in communities that are not directly affected.

Keywords: Shocking news School shootings Cognitive performance Gender differences

1. Introduction In recent years, the United States and several European countries have suffered from a number of terrorist attacks and school shootings. In addition to the tragic direct consequences for the victims and their families, such shocking events may dominate the news for days, having a traumatic effect on millions of other people. Addington (2003) uses National Crime Victimization Survey data to explore the effects of the Columbine High School shooting on students' fear levels, finding that students were slightly more fearful afterwards.1 Galea et al. (2002) document that the 9/11 terrorist attack resulted in thousands of New York City residents developing posttraumatic stress disorder, and Schlenger et al. (2002) show that this was also the case in other parts of the United States. Blanchard et al. (2005) find that the college-age population in the United States was still suffering from the 9/11 attacks in the fall of 2002, with a larger effect in cities closer to New York City. We analyze the indirect effects of shocking events, with focus on how the news of a shocking event affects cognitive performance. Finland has witnessed two school shootings in recent years, both of which received wide media coverage.2 The first one took place in

⁎

Corresponding author at: Ifo Institute - Leibniz Institute for Economic Research at the University of Munich, Poschingerstraße 5, 81679 Munich, Germany. E-mail addresses: [email protected] (P. Poutvaara), olli.ropponen@vatt.fi (O. Ropponen). 1 The Columbine High School shooting was the biggest U.S. news story in 1999 as measured by Cable News Network (CNN) ratings. Muschert (2009) explored the subsequent media dynamics, finding that the news coverage first focused on what happened in Columbine, and then moved to repercussions across the country. 2 There was a large number of stories about both Jokela and Kauhajoki school shootings in the newspapers in the immediate aftermath of these tragedies: Jokela shooting was followed by 901 stories and Kauhajoki shooting by 465 stories in the print media within the next 48 hours. In addition, TV channels provided the breaking news coverage related to the shootings almost all day long immediately after the shootings. (Hakala (2009): Koulusurmat verkostoyhteiskunnassa: Analyysi Jokelan ja Kauhajoen kriisien viestinnästä (School shootings in a networked society: An analysis of news following the crises in Jokela and Kauhajoki). In Finnish; available at: http://www.helsinki.fi/crc/Julkaisut/) http://dx.doi.org/10.1016/j.ejpoleco.2017.03.006 Received 15 September 2016; Received in revised form 16 March 2017; Accepted 17 March 2017 0176-2680/ © 2017 Elsevier B.V. All rights reserved.

Please cite this article as: Poutvaara, P., European Journal of Political Economy (2017), http://dx.doi.org/10.1016/j.ejpoleco.2017.03.006

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

November 2007 in Jokela in southern Finland, and the second one in September 2008 in Kauhajoki in western Finland. We focus on the second shooting as it coincided with national high-school matriculation exams, taken typically when students are aged 18 or 19.3 There were exams both before and after the shooting took place, allowing us to compare performance in exams that took place after the shooting with performance in exams that took place before it. To account for the fact that the average performance may differ between exams and student cohorts, we form a treatment group of exams that took place after the shooting in 2008 and a control group of exams that took place before the shooting in that year. Then we perform a difference-in-differences (DID) analysis to see how the average performance in the treatment group changed between 2007 and 2008, relative to the change in the average performance in the control group. (Note that treatment and control group are based on exams; any difference in student cohorts between 2007 and 2008 should be captured in how the average performance in the control group exams changed.) We study separately the effects of the school shooting in the region in which it occurred, in the region in which the first school shooting had taken place, and in the rest of the country. The effects of the shocking news can be best observed in the rest of the country. In the Kauhajoki region, pinpointing the effects is more complicated, as many students knew some of the victims or the perpetrator. In the Jokela region, the results could be driven by reactivation of painful memories among those who had lost a relative or a friend in the previous shooting. Economists have started to analyze the psychological effects of shocking events, like terrorist attacks or school shootings, only recently. Based on comparisons between interviews before and after 9/11, Metcalfe et al. (2011) show that these attacks reduced subjective well-being also in the United Kingdom. Montalvo (2011) shows that the terrorist attack in Madrid on March 11, 2004 significantly reduced support for the incumbent government in elections three days later. Bozzoli and Müller (2011) show that the terrorist attacks in London on July 7, 2005 considerably increased public support for security measures at the cost of civil liberties, with the willingness to trade off security for liberties being driven by threat perceptions. Abouk and Adams (2013) find that school shootings in the United States are followed by a 10 to 12% increase in enrollment in private high schools, and a decline in public school enrollment. Our paper is the first to study the effects of shocking news on cognitive performance. Identifying the effects of shocking news helps societies to plan policy responses to reduce indirect costs of future atrocities, thereby making societies more resilient. We study the effects of shocking news on the cognitive performance by using register data on the examination results. We have the test results for all students in a random sample of schools and therefore avoid the problems related to sample selection, as well as under- or overreporting of symptoms that might occur in survey data. In addition to reporting error and commensurability issues, collecting survey data includes ethical issues as well, because collecting such data could trigger traumatic memories, especially among those who reacted to the events most strongly.4 We study the reactions of men and women separately, as previous research has documented various gender differences with respect to reaction to stressful events. Females are found to suffer more often from acute and posttraumatic stress disorders (see Schlenger et al., 2002; Silver et al., 2002; Marshall et al., 2007). This suggests that females would also respond to the school shooting more strongly. On the other hand, there is a vast literature (see Grant et al., 2006) showing that social support protects young people from the negative effects of stressors (known as the buffering effect). Kendler et al. (2005) find that females have, on average, wider social networks than males. This suggests an opposite gender pattern: wider social networks could give females more protection against the adverse effects of shocking news. There is extensive evidence on gender differences also in other contexts, both in psychological and economic literature (Eagly, 1995; Blau and Kahn, 2000; Croson and Gneezy, 2009). Using DID analyses, we find that young men's average test score dropped by 4.3 percentage points, which is about a fifth of a standard deviation and therefore a relatively large effect. The effect on young women is small and statistically insignificant. Our findings suggest that self-reported symptoms may underestimate how severely males are affected by traumatic events relative to females. This result can be interpreted applying the model of limited attention by Banerjee and Mullainathan (2008). They argued that attention is a scarce resource that can be divided between detecting problems at home and problems at work. Having to spend more time on problems at home reduces productivity at work. If teenage women have more attention capital than teenage men, then shocking news would have less harmful effects on women's test performance. Matriculation exams play an important role in admission to universities in many countries. Our findings suggest that a shocking event may have serious repercussions on the lifetime career paths of a large number of young men. One policy implication of our findings is that in countries in which matriculation exams play a major role in university admissions, policy makers should consider allowing young people to resit exams the following year if a shocking event, like a death of a family member, occurs just before exams. Another policy implication is that a shocking event may increase the need for psychological support more widely than is commonly understood.5 While it is natural for health care services to prioritize survivors and bereaved families, a national shocking event may call for extra resources, even in communities that were not directly affected. Furthermore, our results highlight that highstakes exams are vulnerable to random shocks that cause fluctuation in cognitive performance. This may result in an inefficient allocation of talent when such exams are used in university admissions; see Lavy et al. (2014) for a related discussion in the context of air pollution and national exams in Israel.

3

We do not analyze the effects of the 2007 school shooting, as it took place after that year's examinations. As our study uses register data, the risk of burdening the victims is minimal, and meets the guidelines set in the meeting entitled “Ethical Issues Pertaining to Research in the Aftermath of Disaster” that was sponsored by the New York Academy of Medicine and the National Institute of Mental Health (see Collogan et al., 2004). Silver (2004) provides an excellent overview of the challenges associated with conducting methodologically rigorous studies of responses to traumatic experiences. 5 These should be interpreted as tentative suggestions. We hope that these suggestions stimulate additional psychological research to evaluate their validity. 4

2

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Our paper proceeds as follows. The follow-up section reviews the related literature. Section 3 provides background information on Finnish matriculation examinations and the school shooting. The data and the empirical framework are described in Section 4. The results are presented in Section 5, which starts by describing the development of interruption rates and statistics on average performance, and proceeds to an econometric analysis. Section 6 concludes. 2. Related literature Our study is related to multiple branches of literature. First, our paper is related to an increasing strand of literature in which the direct consequences of terrorist attacks or school shootings are examined, or such events are used as an exogenous source of variation to examine some other questions. Second, we contribute to the economics of education literature by analyzing the effects of shocking news on student performance. This is important for planning psychological support in schools following local or national tragedies, and it also raises questions related to how to mitigate the adverse lifetime consequences that such shocks may impose in the presence of high-stake exams. Third related strand of literature analyzes the effects of news and media. Fourth, our paper contributes to recent advances in economics on health or psychological effects of various shocks or conditions that have not been traditionally studied by economists. Previous papers have examined the psychological aftermath of terrorist attacks (Blanchard et al., 2005; Marshall et al., 2007; Su et al., 2009), as well as how these affect relative valuation of security measures and civil liberties (Bozzoli and Müller, 2011), financial markets (Kollias et al., 2011), entrepreneurial activity (Brück et al., 2011) or electoral outcomes (Montalvo, 2011), and analyzed the effects of school shootings on school enrollment (Abouk and Adams, 2013).6 The news on the 9/11 terrorist attack reduced subjective well-being also in the United Kingdom (Metcalfe et al., 2011). The exogenous variation in police presence caused by terrorist attacks or the threat thereof has been also used to study the effects of policing on crime (Di Tella and Schargrodsky, 2004; Klick and Tabarrok, 2005; Poutvaara and Priks, 2009; Draca et al., 2011).7 We show that the psychological effects of shocking news on student performance are economically relevant. Especially young men performed worse in important tests after the shocking news, reducing their chances of entering their preferred education, which in turn can have severe effects on lifetime earnings and collective welfare due to misallocation of human capital. To put our findings into perspective, it is useful to compare the magnitude of the effect we find (a fifth of a standard deviation) with the effects of other shocks or policy interventions in a school context. Based on US military deployments, Lyle (2006) finds that the negative effect of parental absence during the school year on children's test scores is about a tenth of a standard deviation. Lyle does not study the effect separately for boys and girls, so it is not possible to verify whether the gender pattern in his data is the same. Li et al. (2014) study the effect on test scores achievement of pairing high and low achieving classmates as benchmates. Their semester-long intervention increased low-achiever test scores by about fourth of a standard deviation. Oswald and Backes-Gellner (2014) study the effect of financial incentives on student performance. Their intervention increased the student performance by roughly a third of a standard deviation. Rud et al. (2014) find that having criminally involved parents reduces the likelihood of attending higher education by 2 to 6 percentage points. Swee (2015) analyzes the effects of the 1992–1995 Bosnian War on schooling attainment, and shows that at the municipal level, a one standard deviation increase in the number of war casualties per capita decreases the likelihood of attending secondary school by 4 percentage points. Beland and Kim (2016) show that homicidal shootings reduce math and English test scores in schools in which the shootings took place in the United States. The proportion of students achieving a proficient or advanced level in math test decreases by 5 percentage points in these schools. For English test the corresponding decrease is 4 percentage points. Unlike us, they do not study what were the effects of news, or whether there are gender differences in reactions to school shootings. Similar to our study, Lavy et al. (2014) also analyze the effects of adverse shocks on student performance in high-stakes examinations.8 While they study the effects of air pollution with the most likely mechanism being physiological, we study the psychological effects of shocking news on student performance. The results of Lavy et al. (2014) are similar to ours. They also find that the performance of young men is reduced more than that of young women. Young men are therefore likely to pay bigger costs of the exogenous shocks than young women. Intriguingly, positively experienced events may also have a negative effect on academic performance. Lindo et al. (2012) and Hernandez-Julian and Rotthoff (2014) found a negative relationship between the success of the university's football team and its students' subsequent academic performance for the University of Oregon and for Clemson University, respectively. Lindo et al. (2012) found a bigger negative effect for men, and Hernandez-Julian and Rotthoff (2014) for women. The literature on the effects of news and media is interlinked with the literature on the effects of terrorist attacks. The effects of the terrorist attacks on electoral outcomes (Montalvo, 2011) and subjective well-being in another country (Metcalfe et al., 2011) arise through the news of such atrocities. The same goes for the relative valuation of security measures and civil liberties (Bozzoli and Müller, 2011). However, the literature on the effects of news and media is much wider. For example, Besley and Burgess (2002) show that Indian state governments increase aid in response to floods more where newspaper circulation is higher and Brunetti and 6 Previous research related to school shootings has been devoted to the personality and background of the perpetrators, factors predicting school shootings, the cultural context in which school shootings take place, and the fear of victimization following school shootings. For excellent summaries of research on school shootings, see Muschert (2009) and Wike and Fraser (2009). 7 For other issues on the effects of terrorist attacks, see Brück and Schneider (2011) and Brandt et al. (2016). 8 The two papers were written independently. The first version of our paper was circulated as IZA DP 5009 in 2010 and the first version of Lavy et al. as NBER WP 20,647 in 2014.

3

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table 1 Timing of matriculation examinations in 2007 and 2008.

Friday Monday

Wednesday Friday Monday Tuesday Wednesday

Friday

Monday

Fall 2007

Fall 2008

14th September mother tongue (Finnish, Swedish), 1st exam 17th September psychology, philosophy, history, physics, biology

12th September mother tongue (Finnish, Swedish), 1st exam 15th September religion, life philosophy, civics, chemistry, geography, health education 17th September mathematics, long and short syllabus 19th September foreign language, long syllabus 22nd September mother tongue (Finnish, Swedish), 2nd exam 23rd September SCHOOL SHOOTING IN KAUHAJOKI 24th September psychology, philosophy, history, physics, biology

19th September mathematics, long and short syllabus 21st September foreign language, long syllabus 24th September mother tongue (Finnish, Swedish), 2nd exam 25th September 26th September religion, life philosophy, civics, chemistry, geography, health education 28th September other domestic language, long and intermediate syllabus 1st October foreign language, short syllabus

26th September other domestic language, long and intermediate syllabus 29th September foreign language, short syllabus

Notes: in addition to above, exam on Lappish is also taken by few people in Finland. None of them are in our data. The examinations in the treatment group (20 out of 38) are given in italics.

Weder (2003) and Bhattacharyya and Hodler (2015) show that free press reduces corruption. As for the effect of media on young people, Kearney and Levine (2015) conclude that MTV reality show 16 and Pregnant led to a 4.3 percent reduction in teen births. In recent years, the scope of questions studied in economics has greatly expanded, to include questions previously left to medical research or psychology. Almond (2006) shows that people who were in utero during the 1918 influenza pandemic had less education and lower earnings in 1960 to 1980 than somewhat older cohorts. This provides support to the fetal origins hypothesis that links certain chronic health conditions with circumstances during fetal development. Deschênes et al. (2009) and Simeonova (2011) find that fetal exposure to extreme weather conditions leads to lower birth weights. Eccleston (2011) finds that exposure to maternal stress in utero in New York City during the 9/11 attacks had adverse effects on birth weight and early schooling outcomes. We add to these earlier contributions by examining the effects of shocking news on cognitive performance. There is also a link to recent literature on attention as a scarce resource, initiated by Banerjee and Mullainathan (2008). Our results suggest that at least among young adults, attention capital remaining after receiving shocking news is, on average, larger among women.

3. Finnish school system and timing of shooting In Finland, all students aiming to graduate from high school have to pass matriculation exams in at least four subjects. The exams are administered nationally at the same time in all schools, with the same questions and grading criteria in the whole country. The matriculation exams are standardized tests used to evaluate high school students in the Finnish school system and these results are also used in determining access to universities and universities of applied science. Every spring and fall, there is an exam period of two to three weeks. Each test can be taken only on a given day and students have to register for their chosen exam several months in advance. On most exam days there are separate exams in different subjects. In the fall of 2008, the exam period took place from September 12 until September 29 (see Table 1). On September 23, a lone gunman murdered nine students and one staff member in Kauhajoki, before committing suicide. By that time, 18 out of 38 exams had already taken place. This massacre, which took place in the middle of the exam period, dominated the news for several days. We examine the effects of this shocking news by comparing the average of the test scores after the shooting to those in the same subjects in the previous year. In order to control for possible cohort differences, we also employ test scores from exams that took place before the shooting. One possible threat for the results based on the comparison of the average test scores is a selection of students. If there occurred a change in the composition of students taking the exam over the years, then the results might just be a reflection of this change. In our study the shooting cannot have affected the registration, because the registration took place several months in advance. In addition, we have the interruption rates of those who had registered.9 Our analysis in Section 5 begins by a study on the interruption rates. As we observe no statistically significant changes in the interruption rates, apart from a small effect among females from the Kauhajoki region, the change in the student composition should not play any role in our results about the effects of shocking news. 9 We employ the official definition for an interrupted exam provided by the Matriculation Examination Board. A given observation is considered as interrupted if the student had not finished all the required parts of this exam. The term “interruption rate” stands for the ratio of the number of interrupted exams to the number of registrations to the corresponding exams.

4

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table 2 Sample sizes. Fall 2006

Kauhajoki region Before After Jokela region Before After Rest of the country Before After

Fall 2007

Fall 2008

Females

Males

Females

Males

Females

Males

634 367

371 156

416 291

233 145

380 280

249 128

553 281

446 183

492 271

429 256

555 342

493 229

1641 890

1230 514

1240 925

997 527

1403 862

1023 572

4. Data and empirical framework The Matriculation Examination Board, which is responsible for organizing matriculation exams, kindly provided us with individual-level data on test scores for the years 2006, 2007 and 2008 for selected schools. Our sample covers all six schools in the Kauhajoki region, where the 2008 shooting took place, all six schools of the Jokela region, where the 2007 shooting took place, and 28 randomly selected schools in the rest of the country. In total, there are 417 high schools in Finland, of which our sample includes about 10%. We use test results for the fall only, to account for systematic differences in participation between the fall and spring periods. Table 2 presents sample sizes for completed tests. We divide our subjects into two groups. Our treatment group consists of subjects for which the exam took place in 2008 after the shooting, such as those in psychology and philosophy (see Table 1). Those are the subjects, whose average results might have changed due to the shooting. For the control group subjects (mother tongue etc.), the exams took place in 2008 before the shooting. The shooting cannot, therefore, have affected these subjects. Because the exams in 2007 were in the exactly same subjects as in 2008 and we have knowledge both on each subject and each school we can control for these in our regressions. Therefore, we can base our results on the comparison of the test scores in the same school and in the same subjects. Comparability with 2006 results is more limited due to reforms that took place in 2007. Nonetheless, we use 2006 results below to test the parallel trend assumption, restricting the attention to exams that were not affected by the reforms. As we have the test results for all students in a random sample of schools, we avoid the problems related to sample selection, as well as under- or over-reporting of symptoms that might occur in the survey data. Johnston et al. (2009) compare the survey responses of a self-reported measure of hypertension with objective measures of it. They show that the results derived using their self-reported measure differ from those described by their objective measures, suggesting that self-reported health measures may suffer from a reporting error. We measure individual performance in a test as the percentage of the theoretical maximum points. We first calculate for females and males together, and then for each gender, how the average performance in exams forming part of the treatment (control) group changed between 2007 and 2008. We give the same weight to each participant in each test. The change in the average performance in the control group captures the possibility that students in 2008 differed from those in 2007. The change in the average performance in the treatment group includes both the cohort effect, which is seen also in the control group exams and the effect of shocking news. To obtain the effect of the shocking news, we subtract the change in the average performance in the control group from that in the treatment group. The main equation we estimate is

yijt = β0 + β1Aij + β2Tit + β3Aij Tit + γs + ϵijt

(1)

where yijt is the standardized test score of student i in subject j in year t, β0 is a constant, Aij is a dummy variable for a subject with a value of one if the exam took place in 2008 after the shooting (like psychology), Tit is a dummy variable for the year 2008, Aij Tit is the interaction term, γs is a school fixed effect and ϵijt is the error term. While most of our analysis studies females and males separately, we also study whether the gender difference is statistically significant. To do that, we estimate the equation

yijt = β0 + β1Aij + β2Tit + β3Aij Tit + γs + δ 0 * malei + δ1Aij * malei + δ2Tit * malei + δ3Aij Tit * malei + ϕs * malei + ϵijt

(2)

where malei is a dummy variable which receives value of one if student i is a male. The interaction term ϕs * malei shows that we allow the school fixed effects to be different for females and males. As we observe several test results for most participants, there is a strong argument to cluster standard errors at the individual level. However, as we have thousands of observations with just a few for each individual, one could also argue that the effect of clustering of the standard errors should not play any significant role in our study. To verify whether this is the case, we present both standard errors. We also study separately those exams which took place on the same day in both years. This way we exclude the possible effect of timing of the exams on our results.

5

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table 3 Percentages of interrupted tests. Fall 2007

Kauhajoki region Before After Jokela region Before After Rest of the country Before After

Fall 2008

Females

Males

Females

Males

3.9 4.0

3.3 2.0

2.6 4.5

5.7 3.0

3.5 2.5

6.6 4.5

4.0 4.0

4.1 5.4

2.4 2.3

3.2 4.0

4.9 3.8

5.5 4.2

5. Results In addition to possible change in the performance following the shocking news, there might also occur another response to the shooting, namely an increase in the number of interrupted tests. We begin by studying whether there was any change in the number of interrupted exams. In Table 3, we report the percentage of tests that were interrupted in 2007 and 2008 in the Kauhajoki region, the Jokela region, and the rest of the country. In addition to the descriptive statistics in the table we conducted an econometric analysis of interruption rates (see Table A1) and find no statistically significant changes in the interruption rates in the treatment group in 2008 in the Jokela region or in the rest of the country.10 Table 4 reports the average performance in 2007 and 2008. The changes in average performance outside the Jokela and Kauhajoki regions are striking. The average test score of males in exams that took place before the shooting increased by 0.5 percentage points from 2007 to 2008. In the exams after the shooting, the average test score of males dropped by 3.1 percentage points. The average test score of females dropped by 2.2 percentage points in the exams before the shooting and by 2.4 percentage points in the exams after the shooting. Therefore, males appear to react more strongly. In the Kauhajoki and Jokela regions, the performance of both males and females deteriorated after the shooting, and more so in the Jokela region. Table 5 presents the DID estimations for the Kauhajoki region, the Jokela region and the rest of the country for females and males together. The estimated effect of news is always negative, and statistically significant in the Jokela region and in the rest of the country. In order to test whether the pronounced gender differences in Table 4 are statistically significant, Table 6 presents the DID estimations for the Kauhajoki region, the Jokela region and the rest of the country for females and males together with gender dummy interactions, taking females as the reference group. The estimated effect of news is small and statistically insignificant for females. The interaction term for how the effect of news for males differs from the effect of news for females is always negative, but statistically significant only in the rest of the country. It should be noted that the point estimate for the interaction term for the Jokela region is almost as large as in the rest of the country, so that it not being statistically significant is driven by the much larger standard error, reflecting fewer observations. As the total effect of news for males is the sum of the baseline effect and the interaction term, and the baseline effect (which is estimated for females) is also negative, it would be too early to conclude that there is no effect of news for males in the Jokela region. Table 6 allows stating for the Jokela region only that the estimated negative effect of news for females is statistically insignificant, and that the estimated gender difference between females and males is also statistically insignificant. Table 7 presents the DID estimations for the Kauhajoki region, the Jokela region and the rest of the country, estimated separately for females and males. The effect is negative in all regions, but the drop in performance is statistically significant only for males in the Jokela region and in the rest of the country.11 The point estimate of the effect on females is almost zero in the rest of the country. Fig. 1 illustrates the effects of the news in the rest of the country for both males and females. As the average performance of males outside the Kauhajoki and Jokela regions in the treatment group was 59.3 percentage points of the theoretical maximum in 2007 (see Table 4), the estimated −4.3 percentage point effect of the shocking news implies a 7percent drop in their overall average performance after the shooting. The reduction is about a fifth of the standard deviation of the standardized test scores (22.5) employed in the third column for males of Table 7. This is a large effect. As a point of comparison, Lyle (2006) studies the effect of parental absences and household relocations on children's academic performance using US military deployments, and finds that the negative effect of parental absence during the school year on children's test scores is about a tenth of

10 There seems to occur an increase in the interruption rate among Kauhajoki females. Yet our focus is on the results for the rest of the country. If females and males are pooled as shown in Table 5, the effect of news is statistically insignificant in all specifications. (Detailed analysis available upon request.) 11 Our prior was that the negative effect would be strongest in the Kauhajoki and Jokela regions. One possible explanation for the puzzling result that males' performance in the Kauhajoki region dropped less than elsewhere is that sometimes people perform surprisingly well in acute threat situations, and the threat was most acute in the region in which the shooting had taken place. Mobbs et al. (2007) show that an acute threat can activate different areas of brain than a more distant threat. Nonetheless, this should be viewed just as a tentative explanation and it should be noted that the males' average performance deteriorated somewhat also in the Kauhajoki region, although the change was not statistically significant.

6

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table 4 Standardized average performances. Fall 2007

Kauhajoki region Before After Jokela region Before After Rest of the country Before After

Fall 2008

Females

Males

Females

Males

54.9 (1.0) 58.6 (1.1)

50.7 (1.5) 54.3 (1.6)

54.2 (1.1) 56.7 (1.1)

48.0 (1.5) 50.4 (1.7)

56.6 (0.9) 60.6 (1.3)

56.3 (1.2) 56.1 (1.4)

60.6 (1.3) 56.1 (1.0)

56.1 (1.4) 48.4 (1.2)

58.5 (0.6) 64.5 (0.7)

55.5 (0.8) 59.3 (0.9)

56.3 (0.6) 62.1 (0.7)

56.0 (0.7) 56.2 (0.9)

Note: Standard errors are given in parentheses. Table 5 Difference-in-differences estimation results. Dependent variable: test performance Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2007, 1=2008) The effect of news

Male (0=no, 1=yes) Constant

School fixed effects Subject fixed effects Observations

Kauhajoki region

Jokela region

Rest of the country

Rest of the country

3.6*** (1.3) [1.2] −1.5 (1.1) [1.3] −1.0 (1.8) [1.7] −5.1*** (0.9) [1.2] 57.3*** (1.5) [2.0] yes no 2122

2.2** (1.1) [1.0] −0.7 (1.0) [1.0] −4.1*** (1.6) [1.3] −1.7* (0.8) [0.9] 62.9*** (1.1) [1.4] yes no 3067

4.4*** (0.7) [0.6] −0.9 (0.6) [0.6] −2.1** (0.9) [0.9] −1.9*** (0.5) [0.6] 53.6*** (2.1) [3.0] yes no 7549

25.8*** (5.6) [4.5] −0.3 (0.5) [0.5] −2.2*** (0.7) [0.7] −1.7*** (0.4) [0.5] 46.3*** (1.8) [2.3] yes yes 7549

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses and (individual) clustered standard errors are given in square brackets. Significance levels are based on clustered standard errors. *** Denotes significance level 1%. ** Denotes significance level 5%. * Denotes significance level 10%.

a standard deviation. Aaronson et al. (2007) find that a one standard deviation improvement in math teacher quality for one semester raised student test scores in Chicago public high schools by 0.13 grade equivalents. A two-semester improvement corresponds to a tenth of a standard deviation of ninth grade math scores.12 In the last column of Table 7, we report the results for the rest of the country with subject and school fixed effects.13 The results are even more striking. The performance-reducing effect of the news on males is now statistically significant at the 0.1 percent level. The point estimate of the performance-reducing effect for females is still statistically insignificant. Our conclusions remain robust when we included individual fixed effects into the regressions. The results for these regressions are provided in Table A3 in the Appendix. As we observed in Table 1 some exams are taken earlier in 2007 than in 2008 and vice versa. This might influence our results if the performance is dependent on how late some exam is taken. In order to study this possibility we provide in Table 8 the results

12 13

The standard deviation of the ninth grade math scores (2.64) employed in Table 11 in Aaronson et al. (2007) is given in their Table 1. We perform this analysis only outside the Kauhajoki and Jokela regions, due to the small sample sizes in several subjects in the Kauhajoki and Jokela regions.

7

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table 6 Difference-in-differences estimation results. Dependent variable: test performance Dependent variables Test belongs to the treatment group (0=no, 1=yes) …X Male (0=no, 1=yes) Test year (0=2007, 1=2008)

…X Male (0=no, 1=yes) The effect of news

…X Male (0=no, 1=yes) Constant

…X Male (0=no, 1=yes) School fixed effects Subject fixed effects Observations

Kauhajoki region

Jokela region

Rest of the country

Rest of the country

3.9*** (1.6) [1.4] −1.3 (2.7) [2.6] −0.8 (1.5) [1.6]

4.3*** (1.6) [1.4] −4.9** (2.3) [2.0] −0.8 (1.3) [1.4]

5.6*** (0.9) [0.8] −3.3** (1.4) [1.3] −1.6* (0.8) [0.8]

23.6*** (6.0) [5.1] 10.4* (16.8) [5.8] −1.0 (0.6) [0.7]

−2.0 (2.4) [2.7] −1.2 (2.2) [2.1] 0.44 (3.8) [3.5] 54.3*** (1.7) [2.3] 4.8 (3.1) [3.8] yes no 2122

<0.1 (1.9) [2.1] −2.8 (2.1) [1.8] −3.6 (3.2) [2.7] 60.3*** (1.4) [1.7] 5.2* (2.2) [2.7] yes no 3067

2.0 (1.2) [1.3] −0.6 (1.2) [1.1] −3.8** (1.9) [1.8] 55.3*** (2.6) [2.8] −6.8 (4.4) [6.9] yes no 7549

1.7 (0.9) [1.1] −1.0 (1.0) [0.9] −3.1** (1.5) [1.5] 49.6*** (2.2) [2.4] −9.6* (3.7) [5.1] yes yes 7549

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses and (individual) clustered standard errors are given in square brackets. Significance levels are based on clustered standard errors. *** Denotes significance level 1%. ** Denotes significance level 5%. * Denotes significance level 10%. Table 7 Difference-in-differences estimation results. Dependent variable: test performance

Dependent variabls Test belongs to the treatment group (0=no, 1=yes) Test year (0=2007, 1=2008) The effect of news

Constant

School fixed effects Subject fixed effects Observations

Kauhajoki region

Jokela region

Females

Males

Females

3.9*** (1.5) [1.4] −0.8 (1.4) [1.6] −1.2 (2.2) [2.1] 54.3*** (1.7) [2.3] yes no 1367

2.6 (2.4) [2.2] −2.8 (2.0) [2.2] −0.8 (3.3) [2.8] 59.1*** (2.8) [3.0] yes no 755

4.3*** (1.5) [1.4] −0.8 (1.2) [1.4] −2.8 (2.0) [1.8] 60.3*** (1.3) [1.7] yes no 1660

Rest of the country

Rest of the country

Males

Females

Males

Females

Males

−0.6 (1.8) [1.4] −0.8 (1.5) [1.6] −6.4*** (2.5) [2.0] 65.5*** (1.8) [2.1] yes no 1407

5.6*** (0.8) [0.8] −1.6* (0.8) [0.8] −0.6 (1.2) [1.1] 55.3*** (2.6) [2.8] yes no 4430

2.3** (1.1) [1.1] 0.3 (0.9) [1.0] −4.3*** (1.6) [1.4] 48.5*** (3.6) [6.3] yes no 3119

23.6*** (5.1) [5.1] −1.0 (0.6) [0.7] −1.0 (1.0) [0.9] 49.6*** (2.3) [2.4] yes yes 4430

34.0*** (15.6) [2.9] 0.8 (0.7) [0.8] −4.1*** (1.2) [1.1] 39.9*** (2.9) [4.5] yes yes 3119

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses and (individual) clustered standard errors are given in square brackets. Significance levels are based on clustered standard errors. The results are also derived when clustering is performed at the school level. Those results are provided in Table A2 in the Appendix. *** Denotes significance level 1%. ** Denotes significance level 5%. * Denotes significance level 10%.

8

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Fig. 1. The effects of the news in the rest of the country. Notes: Results are presented as mean effect ± standard error of mean effect. The scale on the vertical axis is measured as percentage points of theoretical maximum points.

Table 8 Difference-in-differences estimation results with only exams that were at the same time in both years. Dependent variable: test performance

Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2007, 1=2008) The effect of news

Constant

School fixed effects Subject fixed effects Observations

Kauhajoki Region

Jokela Region

Rest of the country

Rest of the country

Females

Males

Females

Males

Females

Males

Females

Males

9.5*** (1.9) [1.6] −0.6 (1.7) [1.9] −2.6 (2.7) [2.4] 56.8*** (1.9) [2.4] yes no 920

5.5*** (2.9) [2.6] −6.1** (2.4) [2.6] 0.6 (4.2) [3.5] 62.0*** (3.3) [3.2] yes no 510

11.6*** (2.0) [1.7] −1.3 (1.6) [1.8] −7.9*** (2.7) [2.3] 63.8*** (1.6) [2.0] yes no 1039

9.8*** (2.6) [1.8] <0.1 (1.8) [2.0] −12.3*** (3.7) [2.8] 69.3*** (2.3) [2.4] yes no 851

10.9*** (1.0) [0.9] −2.2* (0.9) [1.0] −1.4 (1.4) [1.3] 60.2*** (3.1) [3.4] yes no 3024

8.6*** (1.4) [1.2] 0.2 (1.0) [1.1] −5.5*** (2.0) [1.6] 52.6*** (4.4) [7.8] yes no 2110

23.2*** (6.1) [5.0] −1.3 (0.7) [0.9] −2.5** (1.2) [1.1] 51.4*** (2.7) [3.0] yes yes 3024

33.6*** (15.6) [3.3] 0.4 (0.8) [0.9] −5.0*** (1.5) [1.4] 40.4*** (3.5) [4.9] yes yes 2110

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses and (individual) clustered standard errors are given in square brackets. Significance levels are based on clustered standard errors. *** Denotes significance level 1%. ** Denotes significance level 5%. * Denotes significance level 10%.

corresponding to those for the rest of the country in Table 7, but where we employ only observations on those exams that have not changed their place between 2007 and 2008. The table shows that our results are not driven by the changes in the order of the exams. The results remain statistically significant. The latter result for women also becomes statistically significant, yet remains smaller than that for men. Meyer (1995) lists potential threats to both the internal and external validity of studies using natural experiments. Most of the potential concerns, such as omitted variables, misspecified variances and mismeasurement, should not affect our results. Even if there was an omitted variable, its effect on the estimation results should be negligible as we examine the same control and treatment groups of tests in each year. If there had been any differences between various tests driven by an omitted variable in 2008, then they would also have shown up in 2007. It also seems unlikely that our study would suffer from mismeasurement of the variables, because there have not been any changes in the definitions of the variables. Furthermore, the tests were planned by the Matriculation Examination Board in order to provide a reliable performance measure to serve as a nationwide standardized exam. The most important assumption regarding the DID estimation is that there are no differences in the pretreatment trends between the treatment and control groups before the treatment. The difference in these trends would threaten the validity of the DID 9

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table 9 Difference-in-differences estimation for placebo results (2006 vs 2007). Dependent variable: test performance

Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2006, 1=2007) The effect of placebo news Constant School fixed effects Subject fixed effects Individual fixed effects Observations

Rest of the country

Rest of the country

Rest of the country

Females

Males

Females

Males

Females

Males

10.1*** (1.4) 4.4** (1.8) −3.1 (2.0) 47.8*** (4.2) yes no no 2246

9.0*** (1.7) 2.2 (2.0) −1.5 (2.3) 35.5*** (5.1) yes no no 1405

20.0*** (3.1) 3.9** (1.5) −1.8 (1.7) 50.3*** (4.1) yes yes no 2246

23.1*** (4.4) 0.8 (1.6) 0.8 (1.9) 40.0*** (4.6) yes yes no 1405

14.0*** (4.0) −3.1 (2.8) 0.7 (2.0) 44.0*** (8.3) yes yes yes 2246

15.5*** (6.0) −2.6 (3.3) 2.4 (2.6) 48.7*** (11.7) yes yes yes 1405

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Exams on mother tongue, health education, mathematics and English (long syllabus) are excluded in the placebo estimates due to the changes from year 2006 to 2007, which in turn makes their results incomparable between these years. ***Denotes significance level 1%. **Denotes significance level 5%. *Denotes significance level 10%.

estimation results. We test the parallel trends assumption in Table 9 and find no evidence for the differences in the pretreatment trends.14 Further, we present placebo results for the interruptions in Table A4. The placebo effects remain statistically insignificant. 6. Conclusions In this paper, we have examined how students' performance changed as a result of the shocking news of a school shooting. Our paper extends recent economic research on how the human mind works. There are no earlier studies in economics on how individual cognitive performance reacts to shocking news. We analyzed the effect of shocking news separately in the region in which the shooting took place, in the region that had suffered from a school shooting in the previous year, and in the rest of the country. Our main result is that the performance of males declined as a result of the news of the school shooting outside the regions in which the shootings had taken place. We did not find any systematic effect for females. One explanation for this intriguing gender difference is that females have, on average, wider social networks that could protect them against the adverse effects of shocking news. Another explanation is biological. Teenage females mature earlier than males. Greater maturity could allow females to cope with a shock more effectively. Our finding that males react more strongly to negative circumstances resembles the finding by Bertrand and Pan (2013) that boys' noncognitive skills react more strongly to parental inputs than girls'. As matriculation examinations affect admission to universities, our findings imply a reduction in the average years of schooling among males. Kemptner et al. (2011) studied the causal effect of years of schooling on health and showed that an increase in years of schooling decreases long-term illness in males, but not females. Therefore, for males, shocking news might have an indirect negative health effect via education, in addition to the direct negative effect on performance. According to Cutler and Lleras-Muney (2010), this may be mediated by change in cognitive ability as they report that education seems to influence cognitive ability, and cognitive ability in turn leads to healthier behaviors. Our result that males are more strongly affected is in line with Kuhn et al. (2009) and Lavy et al. (2014). Kuhn et al. (2009) use register data to study the implications of job loss on public health care expenditure. They find that job loss causes significant health problems for males, whereas for females this was not the case. Lavy et al. (2014) show that transitory increases in air pollution cause a sharper drop in young men's performance in Israel's high-stake Bagrut tests. 14 Due to the major reforms in the Finnish matriculation examination system we conduct our pretreatment trends tests using data only from 2006 and 2007. The biggest reform for the present-day matriculation examination system in Finland took place in 2006. The most notable piece of the reform related to our study is that a subject called “Arts and Sciences” was split into ten separate exams. This split induced the change in the number of subjects from 28 to 37, an increase of more than 30%. In addition to having new subjects without corresponding results from prior to the reform, this major change is also likely to have significant implications on the overall allocation of the preparation time between different exams. Therefore, this matriculation examination reform makes observations before the year 2006 incomparable to the observations after the reform. Thus our pretreatment years are restricted to include at most the observations from 2006 and 2007. In addition, three other changes in the matriculation examinations took place between 2006 and 2007. Health education was introduced as a new subject (the 38th subject), the exam on mother tongue was divided in two parts taking place on different days and two more difficult questions were added into the long syllabus mathematics exam. When health education was introduced, it appears that students who chose to write it often postponed their English (long syllabus) exams (in Finland, it is possible to take exams in different subjects in different terms). (Detailed analysis available upon request.) In order to reduce the possibility that our placebo results would be driven by these changes in the decomposition, we exclude these exams from our placebo regressions.

10

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Our study has two important implications. First, it suggests that in a crisis assistance must be provided not just where the shocking event occurred but also elsewhere, because the media carries the news elsewhere as well. Second, our findings suggest that in countries in which matriculation exams play a major role in university admissions, policy makers should consider allowing young people to resit exams the following year if a shocking event, such as a death of a family member, occurs just before exams. Although our analysis is restricted to test performance in schools, it has potentially important implications also more generally. News of a shocking event could increase the risk of mistakes in various organizations by taking up part of limited attention of workers and managers. Future research should test whether this conjecture receives support. Our paper suggests several topics for further research. First of all, our analysis of short-term effects calls for a complementary study of long-term effects. Secondly, it would be interesting to examine responses to shocking news in different age groups; in our sample, most people were aged 18 to 20. Finally, data from other countries and settings beyond education would allow comparisons to be made between countries and different organizations. Conflict of interest The authors declare that they have no conflict of interest. Acknowledgements An earlier version of this paper was circulated under the title “School shootings and student performance”. We are grateful for the helpful comments by Niclas Berggren, David Cutler, Christina Fong, David Jaeger, Henrik Jordahl, Markus Jäntti, Katarina Keller, Minna Magnusson, Rinat Mukminov, Maximilian Schwefer, Darius Scurtu, Thomas Stratmann, Janne Tukiainen, Madhinee Valeyatheepillay, Deirdre Weber, the participants at the HECER lunch seminar in Helsinki, the CESifo Employment and Social Protection Area Conference in Munich and the EEA-ESEM conference in Malaga and two anonymous reviewers of this journal. We thank the Matriculation Examination Board for providing us with the data. All the views expressed are our own.

Appendix A. Additional results Tables A1–A4.

Table A1 Difference-in-differences estimation results for the interruption rates. Dependent variable: test performance

Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2007, 1=2008) The effect of news Constant School fixed effects Subject fixed effects Observations

Kauhajoki region

Jokela region

Females

Males

Females

0.0008 (0.0036) −0.0003 (0.0033) 0.0108** (0.0051) 0.0040 (0.0039) yes no 1367

0.0002 (0.0088) −0.0004 (0.0074) −0.0062 (0.0124) 0.0009 (0.0103) yes no 755

0.0106* (0.0059) 0.0036 (0.0048) 0.0012 (0.0079) 0.0009 (0.0051) yes no 1660

Rest of the country

Rest of the country

Males

Females

Males

Females

Males

−0.0021 (0.0063) −0.0002 (0.0053) 0.0050 (0.0090) 0.0047 (0.0064) yes no 1407

0.0026 (0.0030) 0.0041 (0.0027) −0.0056 (0.0042) −0.0027 (0.0091) yes no 4430

−0.0024 (0.0039) −0.0026 (0.0032) 0.0040 (0.0054) 0.0321** (0.0126) yes no 3119

−0.0156 (0.0263) 0.0043 (0.0027) −0.0058 (0.0042) 0.0097 (0.0099) yes yes 4430

−0.0003 (0.0709) −0.0028 (0.0031) 0.0043 (0.0053) 0.0300 (0.0133) yes yes 3119

Notes: a test is considered as interrupted if a student has not filled in all the required parts for the particular test. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses. ***Denotes significance level 1%. **Denotes significance level 5%. *Denotes significance level 10%.

11

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table A2 Difference-in-differences estimation results with school level clustered standard errors. Dependent variable: test performance

Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2007, 1=2008) The effect of news

Constant

School fixed effects Subject fixed effects Observations

Kauhajoki region

Jokela region

Females

Males

Females

3.9* (1.5) [2.3] −0.8 (1.4) [1.1] −1.2 (2.2) [3.0] 54.3*** (1.7) [0.7] yes no 1367

2.6 (2.4) [4.3] −2.8* (2.0) [1.5] −0.8 (3.3) [3.6] 59.1*** (2.8) [0.8] yes no 755

4.3** (1.5) [1.8] −0.8 (1.2) [0.6] −2.8** (2.0) [0.9] 60.3*** (1.3) [0.4] yes no 1660

Rest of the country

Rest of the country

Males

Females

Males

Females

Males

−0.6 (1.8) [2.4] −0.8 (1.5) [1.7] −6.4** (2.5) [2.6] 65.5*** (1.8) [1.4] yes no 1407

5.6*** (0.8) [1.6] −1.6 (0.8) [1.3] −0.6 (1.2) [1.4] 55.3*** (2.6) [1.1] yes no 4430

2.3 (1.1) [2.0] 0.3 (0.9) [1.0] −4.3*** (1.6) [1.4] 48.5*** (3.6) [0.8] yes no 3119

23.6*** (5.1) [4.3] −1.0 (0.6) [1.0] −1.0 (1.0) [1.0] 49.6*** (2.3) [2.1] yes yes 4430

34.0*** (15.6) [1.0] 0.8 (0.7) [0.8] −4.1*** (1.2) [1.3] 39.9*** (2.9) [1.5] yes yes 3119

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses and (school) clustered standard errors are given in square brackets. Significance levels are based on clustered standard errors. *** Denotes significance level 1%. ** Denotes significance level 5%. * Denotes significance level 10%.

Table A3 Difference-in-differences estimation results with individual fixed effects. Dependent variable: test performance

Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2007, 1=2008) The effect of news Constant School fixed effects Subject fixed effects Individual fixed effects Observations

Kauhajoki region

Jokela region

Rest of the country

Females

Males

Females

Males

Females

Males

4.9 (11.6) −8.0*** (2.2) −0.8 (1.7) 41.4*** (6.7) yes yes yes 1367

5.6 (8.1) −11.9*** (3.0) 2.6 (2.5) 38.5*** (8.7) yes yes yes 755

13.6*** (4.1) −6.3*** (1.7) −5.5*** (1.6) 37.7*** (8.7) yes yes yes 1660

41.2*** (11.4) −5.3*** (1.9) −5.0*** (1.8) 76.4*** (9.4) yes yes yes 1407

5.1 (6.0) −9.2*** (1.2) −1.4 (0.9) 64.0*** (8.2) yes yes yes 4430

28.2* (16.7) −4.4*** (1.5) −2.8** (1.2) 62.8*** (12.0) yes yes yes 3119

Notes: individual performance in a test is measured as a percentage of theoretical maximum points. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Clustered standard errors are not identified in the specifications and therefore cannot be used here. *** Denotes significance level 1%. ** Denotes significance level 5%. * Denotes significance level 10%.

12

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Table A4 Difference-in-differences estimation results for the interruption rates (placebo results, 2006 vs 2007). Dependent variable: test performance

Dependent variables Test belongs to the treatment group (0=no, 1=yes) Test year (0=2006, 1=2007) The effect of news Constant School fixed effects Subject fixed effects Observations

Rest of the country

Rest of the country

Females

Males

Females

Males

0.0042 (0.0042) <0.0001 (0.0054) 0.0039 (0.0060) −0.0017 (0.0129) yes no 2246

0.0663*** (0.0048) 0.0012 (0.0056) −0.0011 (0.0065) −0.0025 (0.0145) yes no 1405

−0.0008 (0.0115) −0.0005 (0.0056) 0.0050 (0.0062) −0.0021 (0.0150) yes yes 2246

0.0012 (0.0154) 0.0014 (0.0057) 0.0002 (0.0066) −0.0040 (0.0161) yes yes 1405

Notes: a test is considered as interrupted if a student has not filled in all the required parts for the particular test. The effect of news is the coefficient of the interaction term between the variables Test belongs to the treatment group and Test year. Standard errors are given in parentheses. Exams on mother tongue, health education, mathematics and English (long syllabus) are excluded in the placebo estimates due to the changes from year 2006 to 2007, which in turn makes their results incomparable between these years. ***Denotes significance level 1%. **Denotes significance level 5%. *Denotes significance level 10%.

References Aaronson, D., Barrow, L., Sander, W., 2007. Teachers and student achievement in the Chicago public high schools. J. Labor Econ. 25, 95–135. Abouk, R., Adams, S., 2013. School shootings and private school enrollment. Econ. Lett. 118, 297–299. Addington, L.A., 2003. Students' fear after Columbine: findings from a randomized experiment. J. Quant. Criminol. 19, 367–387. Almond, D., 2006. Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 U.S. population. J. Polit. Econ. 114, 672–712. Banerjee, A.V., Mullainathan, S., 2008. Limited attention and income distribution. Am. Econ. Rev.: Pap. Proc. 98, 489–493. Beland, L.-P., Kim, D., 2016. The effect of high school shootings on schools and student performance. Educ. Eval. Policy Anal. 38, 113–126. Bertrand, M., Pan, J., 2013. The trouble with boys: social influences and the gender gap in disruptive behavior. Am. Econ. J.: Appl. Econ. 5, 32–64. Besley, T.J., Burgess, R., 2002. The political economy of government responsiveness and evidence from India. Q. J. Econ. 117, 1415–1451. Bhattacharyya, S., Hodler, R., 2015. Media freedom and democracy in the fight against corruption. Eur. J. Polit. Econ. 39, 13–24. Blanchard, E.B., Rowell, D., Kuhn, E., Rogers, R., Wittrock, D., 2005. Posttraumatic stress and depressive symptoms in a college population one year after the September 11 attacks: the effect of proximity. Behav. Res. Ther. 43, 143–150. Blau, F.D., Kahn, L.M., 2000. Gender differences in pay. J. Econ. Perspect. 14, 75–99. Bozzoli, C., Müller, C., 2011. Perceptions and attitudes following a terrorist shock: evidence from the UK. Eur. J. Polit. Econ. 27, S89–S106. Brandt, P.T., George, J., Sandler, T., 2016. Why concessions should not be made to terrorist kidnappers. Eur. J. Polit. Econ. 44, 41–52. Brunetti, A., Weder, B., 2003. A free press is bad news for corruption. J. Public Econ. 87, 1801–1824. Brück, T., Llussa, F., Tavares, J.A., 2011. Entrepreneurship: the role of extreme events. Eur. J. Polit. Econ. 27, S78–S88. Brück, T., Schneider, F., 2011. Terror and human insecurity: editorial introduction. Eur. J. Polit. Econ. 27, S1–S4. Collogan, L.K., Tuma, F., Dolan-Sewell, R., Borja, S., Fleischman, A.R., 2004. Ethical issues pertaining to research in the aftermath of disasters. J. Trauma. Stress 17, 363–372. Croson, R., Gneezy, U., 2009. Gender differences in preferences. J. Econ. Lit. 47, 448–474. Cutler, D.M., Lleras-Muney, A., 2010. Understanding differences in health behaviors by education. J. Health Econ. 29, 1–28. Deschênes, O., Greenstone, M., Guryan, J., 2009. Climate change and birth weight. Am. Econ. Rev.: Pap. Proc. 99, 211–217. Di Tella, R., Schargrodsky, E., 2004. Do police reduce crime? Estimates using the allocation of police forces after a terrorist attack. Am. Econ. Rev. 94, 115–133. Draca, M., Machin, S., Witt, R., 2011. Panic on the streets of London: police, crime and the July 2005 terror attacks. Am. Econ. Rev. 101, 2157–2181. Eagly, A.H., 1995. The science and politics of comparing women and men. Am. Psychol. 50, 145–158. Eccleston, M., 2011. In utero exposure to maternal stress: Effects of 9/11 on birth and early schooling outcomes in New York city. Job market paper, Harvard University. Galea, S., Ahern, J., Resnick, H., Kilpatrick, D., Bucuvalas, M., Gold, J., Vlahov, D., 2002. Psychological sequelae of September 11 terrorist attacks in New York City. N. Engl. J. Med. 346, 982–987. Grant, K.E., Compas, B.E., Thurm, A.E., McMahon, S.D., Gipson, P.Y., Campbell, A.J., Krochock, K., Westerholm, R.I., 2006. Stressors and child and adolescent psychopathology: evidence of moderating and mediating effects. Clin. Psychol. Rev. 26, 257–283. Hernandez-Julian, R., Rotthoff, K.W., 2014. The impact of college football on academic achievement. Econ. Educ. Rev. 43, 141–147. Johnston, D.W., Propper, C., Shields, M.A., 2009. Comparing subjective and objective measures of health: evidence from hypertension for the income/health gradient. J. Health Econ. 28, 540–552. Kearney, M.S., Levine, P.B., 2015. Media influences on social outcomes: the impact of MTV's 16 and Pregnant on teen childbearing. Am. Econ. Rev. 105 (12), 3597–3632. Kemptner, D., Jürges, H., Reinhold, S., 2011. Changes in compulsory schooling and the causal effect of education on health: evidence from Germany. J. Health Econ. 30, 340–354. Kendler, K.S., Myers, J., Prescott, C.A., 2005. Sex differences in the relationship between social support and risk for major depression: a longitudinal study of opposite-sex twin pairs. Am. J. Psychiatry 162, 250–256. Klick, J., Tabarrok, A., 2005. Using terror alert levels to estimate the effect of police on crime. J. Law Econ. 68, 267–279. Kollias, C., Manou, E., Papadamou, S., Stagiannis, A., 2011. Stock markets and terrorist attacks: comparative evidence from a large and a small capitalization market. Eur. J. Polit. Econ. 27, S64–S77. Kuhn, A., Lalive, R., Zweimüller, J., 2009. The public health costs of job loss. J. Health Econ. 28, 1099–1115. Lavy, V., Ebenstein, A., Roth, S., 2014. The long run human capital and economic consequences of high-stakes examinations. NBER WP 20647. Li, T., Han, L., Zhang, L., Rozelle, S., 2014. Encouraging classroom peer interactions: evidence from Chinese migrant schools. J. Public Econ. 111, 29–45. Lindo, J.M., Swensen, I.D., Waddell, G.R., 2012. Are big-time sports a threat to student achievement? Am. Econ. J.: Appl. Econ. 4 (4), 254–274.

13

European Journal of Political Economy xxx (xxxx) xxx–xxx

P. Poutvaara, O. Ropponen

Lyle, D.S., 2006. Using military deployments and job assignments to estimate the effect of parental absences and household relocation on children's academic achievement. J. Labor Econ. 24, 319–350. Marshall, R.D., Bryant, R.A., Amsel, L., Suh, E.J., Cook, J.M., Neria, Y., 2007. The psychology of ongoing threat: relative risk appraisal, the September 11 attacks, and terrorismrelated fears. Am. Psychol. 62, 304–316. Metcalfe, R., Powdthavee, N., Dolan, P., 2011. Destruction and distress: using a quasi-experiment to show the effects of the September 11 attacks on mental well-being in the United Kingdom. Econ. J. 121, F81–F103. Meyer, B.D., 1995. Natural and quasi-experiments in economics. J. Bus. Econ. Stat. 13, 151–161. Mobbs, D., Petrovic, P., Marchant, J.L., Hassabis, D., Weiskopf, N., Seymour, B., Dolan, R.J., Frith, C.D., 2007. When fear is near: threat imminence elicits prefrontalperiaqueductal gray shifts in humans. Science 317 (5841), 1079–1083. Montalvo, J.G., 2011. Voting after the bombings: a natural experiment on the effect of terrorist attacks on democratic elections. Rev. Econ. Stat. 93, 1146–1154. Muschert, G.W., 2009. Frame-changing in the media coverage of a school shooting: the rise of Columbine as a national concern. Social. Sci. J. 46, 164–170. Oswald, Y., Backes-Gellner, U., 2014. Learning for a bonus: how financial incentives interact with preferences. J. Public Econ. 118, 52–61. Poutvaara, P., Priks, M., 2009. The effect of police intelligence on group violence: evidence from reassignments in Sweden. J. Public Econ. 93, 403–411. Rud, I., Van Klaveren, C., Groot, W., Maassen van den Brink, H., 2014. The externalities of crime: the effect of criminal involvement of parents on the educational attainment of their children. Econ. Educ. Rev. 38, 89–103. Schlenger, W.E., Caddell, J.M., Ebert, L., Jordan, B.K., Rourke, K.M., Wilson, D., Thalji, L., Dennis, J.M., Fairbank, J.A., Kulka, R.A., 2002. Psychological reactions to terrorist attacks: findings from the national study of Americans' reaction to September 11. J. Am. Med. Assoc. 5, 581–588. Silver, R.C., 2004. Conducting research after 9/11 terrorist attacks: challenges and results. Fam. Syst. Health 22, 47–51. Silver, R.C., Holman, E.A., McIntosh, D.N., Poulin, M., Gil-Rivas, V., 2002. Nationwide longitudinal study of psychological responses to September 11. J. Am. Med. Assoc. 288, 1235–1244. Simeonova, E., 2011. Out of sight, out of mind? Natural disasters and pregnancy outcomes in the USA. CESifo Econ. Stud. 57, 403–431. Su, J.C., Tran, A.G.T.T., Wirtz, J.G., Langteau, R.A., Rothman, A.J., 2009. Driving under the influence (of stress): evidence of a regional increase in impaired driving and traffic fatalities after the September 11 terrorist attacks. Psychol. Sci. 20, 59–65. Swee, E.L., 2015. On war intensity and schooling attainment: the case of Bosnia and Herzegovina. Eur. J. Polit. Econ. 40, 158–172. Wike, T.L., Fraser, M.W., 2009. School shootings: making sense of the senseless. Aggress. Violent Behav. 14, 162–169.

14