Primary school test results and teacher assessments of refugee children in the Netherlands

Transcription

1 Primary school test results and teacher assessments of refugee children in the Netherlands Lex Borghans 1 and Cécile Magnée 2 * Abstract We document how test scores relate to how long refugee children are in the host country and how the assessment of the potential of the children by teachers deviates from the test scores. Our main finding is that refugees test scores and assessment levels increase the younger their age at arrival, indicating that refugees catch up over time. It turns out that teachers give a higher assessment to refugees than their test score would suggest, indicating that teachers take the expected catching up into account in their assessment. Additionally, we find that refugee students score and receive significantly lower test scores and assessments than Dutch students. Furthermore, boys score significantly better than girls for both refugee students and Dutch students. We see that refugee students from Iran and Afghanistan receive higher mean test scores and assessment levels than the other countries of origin, whilst Somalian refugees perform worse. Keywords: refugees, primary education, test results, assessment levels, the Netherlands JEL: I21, I24, J15, J24 1 Professor at the Economics Department of the School of Business and Economics, Maastricht University. 2 PhD student at the Research Centre for Education and the Labour Market (ROA), Maastricht University. * Corresponding author: c.magnee@maastrichtuniversity.nl 1

2 1. Introduction The increased flow of refugees into Europe during the past years is a challenge for education. Refugee children have to deal with language problems, a lack of education in recent years, trauma s and need to adjust to a new culture. Initially their school results might be below their potential due to these factors and the question rises whether such gaps are permanent or gradually disappear due to catching up. Teachers can play an important role in this catching up process. They can either assess refugees lower than their potential and assess them according to their current test scores or they can take the catching up of the refugees into account and give a higher assessment than the current refugees test scores would suggest. The aim of this paper is to investigate how test scores relate to how long refugee children are in the host country and how the assessment of the potential of the children by teachers deviates from the test scores. To answer this question, we take advantage of a feature of Dutch education. Almost all children take a standardized test at the end of primary school and receive an assessment by the teacher indicating their potential for secondary education. We use administrative data from and select refugee and native Dutch students for whom a test score and/or assessment level is known. Our sample includes test scores and assessments of about 2,000 refugee and 900,000 native Dutch students. We document how the test score gap between refugees and native children varies with the age of entrance into the country, and how teacher assessments deviate from these test results. Our main finding is that refugees test scores and assessment levels increase the younger their age at arrival, indicating that refugees catch up over time. Although, even after a longer stay in the Netherlands, a part of the gap remains. It turns out that teachers give a higher assessment to refugees than their test score would suggest, indicating that teachers take the expected catching up into account in their assessment. This is especially true for refugees who very recently entered the Netherlands. We also find variation between country of origin. Refugee students from Iran 2

3 and Afghanistan perform relatively good, whilst Somalian refugees on average perform worse than other refugees. The forced migration of refugees makes their situation different from the position of immigrants in general. First, refugees typically enter the host country after a longer period of stress and trauma, due to the circumstances in their home country and the sometimes long period between leaving their own home and finding a new place to live. For that reason, especially during the first years after entering the host country, their position is very vulnerable. Children attend a school using a language that they don t know, after a long period in which they received no or little education. This makes it hard to apply findings from studies about immigrants in general to refugees. The effect of immigrants on natives educational outcomes are documented by several authors (Brunello & Rocco, 2013; Chachasvili-Bolotin, Lissitsa, Shavit & Avalon, 2016; Diette & Uwaifo Ovelere, 2012; Gould, Lavy & Paserman, 2009; Hunt, 2017; Jensen & Würtz Rasmussen, 2011; Neymotin, 2009; Ohinata & van Ours, 2013; Ohinata and & Ours, 2016), but only one paper reports the effect of refugees on natives performances (Figlio & Ozek, 2017). In addition, there is a vast amount of literature on the educational outcomes of immigrants (Andriessen & Phalet, 2002; Barban & White, 2011; Betts & Lofstrom, 2000; Glick & White, 2004; McHenry, 2015). Papers about refugees outcomes have only recently been published (Dustmann, Fasani, Fattini, Minale & Schönberg, 2017; Evans & Fitzgerald, 2017; Maliepaard, Witkamp & Jennissen, 2017). This lack of literature on refugees educational outcomes (and their effect on natives outcomes) is mainly due to the impossibility to identify refugees from immigrant samples, a lack of long term data and availability and the small sample sizes of refugees in surveys when data are available (Evans & Fitzgerald, 2017). This paper reports new insights on the educational outcomes of refugees using administrative data and fills the gap in the current literature on refugees. The evidence from the Netherlands is relevant to a wider 3

4 audience, since the forced migration part of refugees is common to all countries that deal with the increased inflow of refugees the past years. 2. Data and methodology 2.1 Data and selection In this study, we use Dutch administrative data that provide information about all students enrolled in primary education in the Netherlands in the period These data record test scores and assessment levels, along with a set of student characteristics including demographics (e.g. country of origin, gender, birth year), the country of origin of the parents and date of arrival at the Netherlands. In our analysis we exclude students who emigrated (N=15,745) or died (N=758) and when a student is enrolled in multiple schools, we only keep the primary enrolment data (N=4,182). The data does indicate that a child recently migrated to the Netherlands, but does not indicator whether he has the status of a refugee. In order to identify refugees, we use data from yearly reports of the Central Agency for the Reception of Asylum Seekers (COA) and immigration data from the Central Bureau of Statistics (CBS). Yearly reports of COA report the top five or top three countries of origin of refugees that they house. These countries are displayed in Table 1. CBS data show the total inflow of immigrants and make a distinction in the reason for migration, e.g. economic motive or refugee. These data allow us to calculate how many immigrants per year were refugees, which enables us to indicate for which countries of origin the immigrants are most likely refugees. This ratio is shown in brackets in Table 1. This method is in line with a method proposed by Capps, Newland, Fratzke, Groves, Auclair, Fix & McHugh (2015) who identify refugees by looking at certain year/country combinations. If the inflow of refugees is very close to the total inflow of immigrants for a specific arrival year and country of origin, the chance is very high that someone with that 4

5 year/country combination is a refugee. The refugee to immigrant ratio is called Refugee Concentration Ratio (RCR) by Evans and Fitzgerald (2017) and is comparable to the ratio we display in brackets in Table 1. There is no rule of thumb about how high the refugee/immigrant ratio should be, so we choose to use a ratio of at least 0.5, indicating that at least half of the inflow of immigrants in that specific year/country combination were refugees. Selecting the countries from Table 1 with a ratio of at least 0.5, we can define refugee children as students from Iraq, Iran, Afghanistan, Somalia, Eritrea and Syria, who entered the Netherlands and enrolled in primary education between 2003 and Even though we cannot calculate the ratio in 2015, given the fact that all the countries in the top 5 of 2015 had high ratios in the past years and the fact that the inflow of refugees increased over 2015, we assume they will be relevant for our analysis and thus include them. In order to compare the results of these refugees with the national Dutch trend in test scores and assessment levels, we define native students as the ones who are born in the Netherlands, whose parents are born in the Netherlands and who haven t lived outside the Netherlands. We do this in order to eliminate comparisons with second generation immigrants (N= 631,114) or other newcomers (N= 80,559), which might bias the results. Second generation immigrants are defined as students who are born in the Netherlands, but their parents were not. Other newcomers, for example children who moved to the Netherlands due to the job of their parents, are defined as students who are not born in the Netherlands and whose parents are not born in the Netherlands as well. This is similar to the definition of refugees, only the country of origin of the students differs in this case. In the Netherlands it is common that students take a standardized test and receive an assessment from their teacher in the last grade of primary school. Both the test score and assessment level determine the secondary education level of the student. There are several standardized tests a school can choose from, but the Cito test is the most commonly used one. 5

6 The test consists of language and math questions and the scores range from , where 550 is the highest score possible. Since it is difficult to compare the results from different tests due to different scaling and test components, we do not take students into account who took a different standardized test than Cito (N=160,411). The assessment level for the secondary education level of a student is given by the teacher and is based on the standardized test score and the experience of the teacher with the child, e.g. previous test scores, motivation, attitude, learning difficulties etc. A distinction can be made between three main secondary education levels, S1 (vmbo in Dutch), S2 (havo) and S3 (vwo), where S1 is the lowest level and S3 the most challenging. On top of that, S1 has three different tracks. S1a is the lowest track and prepares students for a specific career with a very practical oriented education and S1c is the highest and more theoretical track. If a teacher is in doubt between two levels, it is possible to give a so called double advice. There are two more types of Dutch education levels, namely special education and practical education. We do not take these types into account, because special education is for students with severe behavioral, cognitive or physical disabilities and practical education for students with an extremely low IQ. In general, due to the extreme student characteristics of these two types of education, these two levels are not considered as the mainstream secondary school levels and it is therefore hard to compare them with the S1-S3 levels. Following this consensus, we exclude students with assessment levels for practical or special education (N=26,536). If a student repeats the last grade of primary school he or she has two test scores and assessment levels. In this case we use the first test score and assessment level. When we look at gender, both Dutch and refugee students have about 50% male students in their sample. We calculate the age at arrival of refugees by subtracting the birth year from the year of arrival. One is allowed to enter primary education up until 13 years old. When you are older upon entering the Netherlands, you immediately enter a secondary 6

7 education level. We exclude students who were very old (older than 13) upon entering the Netherlands (N=16) and thus who do not belong in primary education. The mean age at arrival is years. Lastly, we look at the age of the student at the time of the test. We calculate this by subtracting the birth year from the year of the test. Children who are relatively very young when taking the test have probably skipped a grade due to very good school results and children who are relatively old when taking the test probably repeated a grade due to below average results. The mean age when taking the test is higher for refugees (12.04 years old) than for Dutch students (11.35 years old), which is what we would expect given the language problems and trauma s refugees face. We exclude the categories of the age when taking the test which make up less than 1% of both refugee and Dutch students samples, namely the students who are younger than 10 years old or 14 years old at the time when the test is taken (N=211). After the selections on background characteristics we are left with a sample of refugee and native Dutch students who were in the last grade of primary school in However, for our analysis we only look at the students for whom a test score and/or assessment level is known. It is possible that due to learning difficulties or administrative issues some refugees do not take the test and do not get an assessment. In order to get a feeling of the magnitude of this issue, Figure 1 reports the numbers of refugees who took a test and/or an assessment and those who did not. As can be seen, the number of refugees for whom no information about test scores and assessment levels is available is rather small. What is more interesting is the fact that when we look at the kind of information on scores and assessments that is available per age at arrival category, we see that the older students upon arrival are more likely to not take a test or receive an assessment. In the methodology section we will discuss how this could influence our results. 7

8 Table 2 shows the descriptive statistics for the refugees and the native students after the selection and for whom a test score and/or assessment level is known. We have test scores for 1,080 refugees and 1,013,574 native Dutch students and assessment levels for 2,061 and 876,592 students respectively. 2.2 Methodology Our empirical approach treats test scores and assessment levels as dependent variables and a refugee dummy, gender dummy, country of origin and age at arrival categories as independent variables. We are interested in the test scores and assessment levels of refugees compared to native Dutch students and how these relations change given the age at arrival in the Netherlands. In order to do so we report and compare mean test scores and assessment levels for refugee and native Dutch students. We look at and compare mean scores and assessments for different categories as well. We compare boys and girls and different age at arrival categories using independent sample t-tests. Differences between countries of origin are compared using ordinary least square regressions. We regress the mean of a specific country of origin on the mean of all the other countries of origin of refugees, to see whether students from a specific country perform better or worse than the average of the other refugees. The test scores are standardized, which makes the interpretation of the mean test scores and comparison between categories intuitive. The standardized test scores are given on a scale of The interpretation of a difference in mean assessment level is more complex due to the ordinal scale of the assessment levels. To cope with the ordinality of the scale we assign mean test scores of Dutch students to each assessment level, so anchoring the assessments using the Cito test scores for Dutch students. This anchored scale shows an upward relation between test scores and assessment level, i.e. the higher your test score, the higher the assessment you receive. Table 3 states the mean test scores per assessment level for 8

9 Dutch students. Not only does this allow us to interpret differences in assessment levels in terms of standardized test scores, additionally we can use the anchored scale to predict test scores of refugees given their assessments. Doing this, we have the actual mean test scores of the refugees who took the test and the predicted mean test score for all refugees who received an assessment. When we compare these scores for different age at arrival categories, we can demonstrate how the teacher assessments expressed in predicted mean test scores deviate from the actual mean test scores. Important to note is the fact that there might be a selection effect. As can be seen in Figure 1.B the refugees who are younger upon arrival are more likely to receive an assessment level and take the test. The few students who enter the Netherlands at an older age and who are able to take a test are very likely the better performing students. To look into this selection issue, we also calculate the anchored assessments for (1) the refugees who both took the test and got an assessment and (2) the ones who only got an assessment. We compare these predicted test scores for these two groups with the actual mean test scores to see whether this selection effect is present. If the assumption holds that the better students take the test, our results underestimate the true effect of age at arrival on test scores and assessment levels. Our slope would be less steep than the true relation. Our results can be interpreted as causal when the assumption is made that age at arrival does not correlate with the cognitive abilities of children. This would include the assumption that there is no link between the socio-economic status of parents and the age at arrival of the children. 3. Results 3.1 Test scores 9

10 The box plots in Figure 2 provide a first insight into the test score distribution by refugee status and gender. They allow a comparison of the median, the 25 th and 75 th percentile, and provide insight to the span of the test scores. Differences between Dutch native students and refugees can be observed. The median refugee student performs worse than the median Dutch student. The same holds for the 75 th and 25 th percentiles. Furthermore, the median of the Dutch students is about the same as the 75 th percentile of the refugee student for both boys and girls. In turn, the median of the refugee student is comparable to the 25 th percentile of the Dutch students for both boys and girls. The span of the test scores is wider for refugees, which means that there are refugees who receive the highest possible test scores. Whether the differences between refugees and Dutch students and boys and girls are significant is analyzed by performing t-tests on the mean test scores. The significance levels of these t-tests are given in Figure 2 by the asterisks. The line under or above the asterisks indicate which groups are compared. All three comparisons Dutch students versus refugees, Dutch boys versus Dutch girls, refugee boys versus refugee girls are significantly different at a 1% level. Dutch students perform better than refugees and boys perform better than girls for both Dutch students and refugees. Table 4 reports the OLS regressions of the mean test scores of each country of origin of refugee students on the mean test scores of all the other countries of origin. This means that in the OLS regression of Iran, the dependent variable is the mean test score of Iran and the independent variable is the mean test score of Iraq, Afghanistan, Syria, Somalia and Eritrea combined. Results show that at a 1% significance level students from Iraq score and students from Afghanistan score standardized test score points higher than the average of the other countries and Somalian refugees score standardized test score points lower than the mean test score of the rest. 10

11 In order to see whether the refugees catch up with the native Dutch students over time, we calculate the mean test score for each age at arrival category. Figure 3 shows these mean test scores for the age at arrival categories 2-3 years old until years old. The very young and very old age categories are not displayed due to a N<100. We use t-tests to show which mean test scores differ significantly from another age category. We perform these tests on several age category levels. We start with comparing test scores from adjacent age categories, e.g. 4-5 years old and 5-6 year old. Then we expand the age category differences to 2 years, 3 years etc. The significance levels of these t-tests are shown on top the of the bars in Figure 3 and indicate that the older a refugee is at arrival, the lower the mean test score at a 1% significance level. The Dutch reference point is a mean test score of The descriptive results on test scores can be summarized as (1) Dutch students perform better than refugee students on the test, (2) boys perform better than girls, (3) refugee students from Iran and Afghanistan perform better than the mean test score of the other countries of origin, whilst Somalian refugees perform worse, and (4) the younger the age at arrival, the higher the test score. 3.2 Assessment levels Figure 4 shows the percent of students from several categories Dutch students, all refugees and separately for countries of origin that receive a certain assessment level. Almost 20% of the Dutch students receive the highest assessment level (S3) followed by Iran and Afghanistan. The share of Somalian refugees who receive an S3 assessment is the lowest, whilst around 30% of them receive the lowest assessment level. As was mentioned before, due to the ordinal scale of the assessment levels, it is hard to compare the results between groups. To overcome this issue, we calculate the mean test score for refugees given their 11

12 assessment levels, using the Dutch trend. This calculated mean test score is called the predicted mean test score. Table 5 reports the OLS regressions of the predicted mean scores of each country of origin of refugee students on the mean test scores of all the other countries of origin. Results show that at a 1% significance level students from Iraq score and students from Afghanistan score predicted standardized test score points higher than the average of the other countries and Somalian refugees score predicted standardized test score points lower than the mean test score of the rest. Figure 5A displays the actual (blue line) and predicted (red line) mean test scores of refugees per age at arrival categories. For the younger age at arrival categories, these two lines do not differ, suggesting that when a child is younger upon arrival, he or she has enough time to catch up with the Dutch students and a teacher thus gives an assessment according to the level of the child at the end of primary school. If we look at the older age at arrival categories, Figure 5A shows that the predicted mean test score is higher than the actual mean test score, indicating that teachers do take into account that these older students have some catching up to do. To see whether a selection effect takes place, we compare the actual mean test scores with the predicted mean test scores of the refugees who got an assessment and took the test and the ones who did not take the test. The rationale behind this is that only the refugees who arrive at an older age and are able to take the test are the more able ones. The results in Figure 5B show the meant test scores for these three categories per age at arrival category. Again, for the younger age at arrival categories, no significant differences can be observed. However, for the older age at arrival categories, the figure shows that the refugees who took the test receive significant higher predicted mean test scores than their actual mean test score would suggest, whilst the opposite holds for refugees who did not take the test. This suggest that there is 12

13 indeed a selection effect, with the students who arrive at an older age and do take the test being the better performing refugees. In turn, teachers realise this and give them a higher assessment than their test scores would suggest. The results on assessment levels can be summarized as (5) refugees from Afghanistan and Iraq receive higher assessment levels than the other countries, Somalian refugees receive lower assessments than the rest, (6) teachers give a higher assessment to refugees than their test score would predict, indicating that they take the catching up of refugees over time into account, and (7) there is a selection effect amongst the refugees who arrived at an older age, suggesting the refugees who take the test are the better performing ones. 4. Conclusion This paper documents new findings about the primary education outcomes of refugee students in the Netherlands. The main results show that refugee students perform better the younger their age at arrival, indicating they catch up over time. Teachers take this catching up into account, as can be seen by the fact that they give higher assessment levels than the test score of the refugee would suggest. Additionally, we find that Dutch students perform better than refugee students, boys perform better than girls for both Dutch and refugee students, refugees from Iran and Afghanistan perform better and refugees from Somalia perform worse than the rest. Moreover, a selection effect is observed amongst the refugees who were older upon arrival, indicating that the refugees of this group who take the test are the better performing ones. Therefore, our results might underestimate the true relation between test scores and assessment levels and the age at arrival of refugees. The results of this paper have some clear policy implications for refugees education. First, the finding that refugee children score so poor in the first years in the host country, indicates that they go through a very vulnerable period in their education. It would be 13

14 important to investigate in what way education can optimally support them. A lower score or assessment can be caused by a language problem, or it can be the consequence of learning disabilities related to physical phenomena or the reaction to a traumatic past (Ferfolja & Vickers, 2010). And even though they catch up, they do not reach the level of Dutch native children. Second, the fast catching up of refugee children shows that their long run potential is higher than their current performance at school. It is important to take this specific situation into account in teaching refugees. Thirdly, we see substantial differences in performance of children from different countries of origin. For example, the group of Somalian refugees score significantly lower on the standardized test and receive lower assessment levels. This is a clear sign that they are lagging behind and need more attention in, for example, the form of extra language classes or more intensive supervision. An extension of our paper is to look at data about secondary education as well. We use the assessment a student gets, but this does not have to correspond with the level that the students enrolls in at secondary education. Besides checking for the real situation of distribution over secondary education levels, it would be interesting to see how many and which type of students change levels over the course of the secondary education. 14

15 References Andriessen, I. & Phalet, K. (2002). Acculturation and school success: A study among minority youth in the Netherlands. Intercultural Education, Vol. 13 No. 1, p Barban, N. & White, M. (2011). Immigrants Children s Transition to Secondary School in Italy. International Migration Review. Vol. 45, No. 3, p Betts, J. R. & Lofstrom, M. (2000). The Educational Attainment of Immigrants: Trends and Implications. In Issues in the Economics of Immigration, ed. George J. Borjas, Chicago: University of Chicago Press. Brunello, G. & and Rocco, L.( 2013). The Effect of Immigration on the School Performance of Natives: Cross-Country Evidence Using PISATest Scores. Economics of Education Review, 32(C): Capps, R., Newland, K., Fratzke, S., Groves, S., Auclair, G., Fix, M. & McHugh, M. (2015). The Integration Outcomes of U.S. Washington DC: Migration Policy Institute Chachashvili-Bolotin, S., Lissitsa, S., Shavit, Y. and Ayalon, H. (2016). The Short Term Effects of Immigrant Students on the Educational Achievements of Native-Born Students. International Migration, 54: Diette, T. & Uwaifo Oyelere, R. (2012). Do Significant Immigrants Inflows Create Negative Education Effects? Lessons from the North Carolina Public School System. IZA Discussion Paper 6561 Dustmann, C., Fasani, F., Fattini, T., Minale, L. & Schönberg, U. (2017). On the economics and politics of refugee migration. Economic Policy, 32(91), p Evans, W. & Fitzgerald, D. (2017). The economic and social outcomes of refugees in the United States: Evidence from the ACS. NBER Working Paper No , National Bureau of Economic Research, Cambridge 15

16 Ferfolja, T., & Vickers, M. (2010) Supporting refugee students in school education in Greater Western Sydney, Critical Studies in Education, Vol. 51, Figlio, D. & Ozek, U. (2017). Unwelcome guests? The effects of refugees on the educational outcomes of incumbent students. NBER Working Paper No , National Bureau of Economic Research, Cambridge Glick, J. & White, M. (2004). Post-secondary school participation of immigrant and native youth: the role of familial resources and educational expectations. Social Science Research. Vol. 33, No. 2, p Gould, E. D., Lavy, V. & Paserman, M. D. (2009). Does Immigration Affect the LongTerm Educational Outcomes of Natives? Quasi-Experimental Evidence. Economic Journal, 119(540): Hunt, J. (2017). The Impact of Immigration on the Educational Attainment of Natives. The Journal of Human Resources. Vol. 52, No. 4, p Jensen, P. & Würtz Rasmussen, A. (2011). The Effects of Immigrant Concentration in Schools on Native and Immigrant Children s Reading and Math Skills. Economics of Education Review, 30: Maliepaard, M., Witkamp, B. & Jennissen, R. (2017). De economische en sociale participatie van asielzoekers. WODC, Den Haag McHenry, P. (2015). Immigration and the Human Capital of Natives. Journal of Human Resources, 50(1): Neymotin, F Immigration and its Effects on the College-going Outcomes of Natives. Economics of Education Review, 28: Ohinata, A. & van Ours, J. (2013). How immigrant children affect the academic achievement of native Dutch children. The Economic Journal, 123(August), p

17 TABLES AND FIGURES Table 1 Countries of origin of asylum seekers Arrival year 2015 Syria (n.a.) Syria (0.982) 2013 Somalia (0.874) 2012 Iraq (0.680) 2011 Afghanistan (0.806) 2010 Afghanistan (0.812) 2009 Somalia (0.904) 2008 Iraq (0.851) 2007 Iraq (0.755) 2006 Iraq (0.595) 2005 Iraq (0.760) 2004 Iraq (0.829) 2003 Iraq (0.901) Top 5 countries of origin of inflow of asylum seekers Eritrea Iraq Ethiopia (n.a.) (n.a.) (n.a.) Eritrea Somalia Iraq (0.988) (0.816) (0.417) Syria Iraq Afghanistan (0.934) (0.614) (0.716) Somalia Afghanistan Iran (0.671) (0.783) (0.614) Iraq Somalia n.a. (0.692) (0.824) Iraq Somalia Iran (0.731) (0.912) (0.558) Iraq Afghanistan Iran (0.855) (0.722) (0.450) Somalia Afghanistan Iran (0.859) (0.599) (0.394) Somalia Afghanistan Guinee (0.767) (0.559) (0.771) Somalia Iran Afghanistan (0.671) (0.374) (0.810) Somalia Burundi n.a. (0.545) (n.a.) Somalia Afghanistan n.a. (0.512) (0.942) Afghanistan Angola n.a. (0.969) (0.923) Afghanistan (n.a.) Afghanistan (0.675) Eritrea (0.989) Russia (n.a.) n.a. Russia (n.a.) Soviet Union (0.451) Soviet Union (0.433) Iran (0.372) Burundi (n.a.) n.a. n.a. n.a. Note: The Central Agency for the Reception of Asylum Seekers (COA) yearly reports the top 5 countries of origin of the inflow of asylum seekers that they house. Note that these data are not complete, since the asylum seekers that are not housed by COA are not included in these data. In 2011, 2005, 2004 and 2003, only the top 3 countries of origin are reported. The number in brackets indicates the share of immigrants with an immigration motive as refugee or family reunion over the total amount of immigrants, using data from the Central Bureau of Statistics (CBS). These data are not available for 2015 and not known for Russia and Burundi. The countries in bold are countries with a ratio of refugees to total immigrants of at least

18 Figure 1 Overview of the sample A. The build up of our refugee sample B. Division of refugees test and assessment categories over age at arrival 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Age at arrival Test score and assessment n.a. Test score known, assessment n.a. Test score n.a., assessment known Test score and assessment known Note: N<100 at age at arrival categories younger than 1 year old, 1-2 years, 2-3 years, 3-4 years and years old 18

19 Table 2 Descriptive statistics Refugees Test score Assessment level Gender Age when taking test (10.543) N=1, (2.501) N=2,061 49,38% male N=3, (0.686) N=1,895 Age at arrival (2.410) N=3,003 Natives (9.708) N=1,013, (2.452) N=876, % male N=1,293, (0.538) N=1,014,937 Note: This table reports the descriptive statistics of test scores, assessment levels, gender, age when taking test and age at arrival for refugee and Dutch students. Refugees are students who are not born in the Netherlands, whose parents aren t born in the Netherlands and who come from either Iran, Iraq, Afghanistan, Somalia, Syria or Eritrea. Native Dutch students are defined as students who are born in the Netherlands, whose parents are born in the Netherlands and who have never lived outside the Netherlands. The test scores, assessment levels, age when taking test and age at arrival are given with standard deviation in brackets. Test scores range from and are standardized. Assessment levels range from 3-11, age when taking test from years old and age at arrival from 0-13 years old. Age when taking test is calculated by subtracting the birth year from the year when the test is taken. Age at arrival is calculated by subtracting the birth year from the arrival year. 19

20 Table 3 Mean test score of native Dutch students per assessment level Assessment level Mean test score Dutch students S1a S1a/S1b S1b S1b/S1c S1c S1c/S S S2/S S Note: this table reports the mean test score of native Dutch students who received the same assessment level. E.g. the first number in the second column reports the mean test score of Dutch students who received an S1a assessment. 20

21 Figure 2 Boxplot of test scores Note: Box plots for Cito test results, displayed by type of student and gender. The boxes are drawn around the median, indicated by the line in the box, and show the interquartile range, from the 25th percentile to the 75th percentile. The whiskers show the span of the data points. Their maximum length is 1.5 times the interquartile range. Data points outside this span are displayed separately and represent outliers. The asterisks indicate the significance level of the t-test, where *: p<0.1, **: p<0.05, ***: p< The line above or below the asterisks indicates which two groups were compared with the t-test. 21

22 Table 4 OLS regressions of countries of origin on test scores (1a) Iran (1b) Iraq (1c) Afghanistan (1d) Somalia (1e) Syria (1f) Eritrea Iran 4.750*** (0.815) Iraq (0.525) Afghanistan 2.593*** (0.640) Somalia *** (0.505) Syria (1.121) Eritrea (1.418) Constant *** *** *** *** *** *** Note: This table reports the OLS regression outcomes of the mean test score of a country of origin on the mean test score of the other countries.. Each column represents a separate regression. E.g. column (1a) shows the regression outcome for when the mean test score of Iran is the dependent variable and the mean test score of Iraq, Afghanistan, Somalia, Syria and Eritrea is the independent variable. Standard errors are given in parentheses. The asterisks indicate the significance level, where *: p<0.1, **: p<0.05, ***: p<

23 Mean test score Figure 3 Mean test scores per age at time of arrival B* C*** A*** B*** C*** D*** B*** C*** D*** C*** D*** B*** C*** D*** C*** D*** B* D*** Age at arrival Note: The height of the bars indicate the mean test score of refugees within a specific age at arrival category. One-sample t-tests were conducted at four different levels: A) 1 year category difference (e.g. 4-5 years old and 5-6 years old), B) two years category difference (e.g. 4-5 years old and 6-7 years old), C) three years category difference (e.g. 4-5 years old and 7-8 years old) and D) four years category difference (e.g. 4-5 years old and 8-9 years old). The indication of A, B, C or D on top of the bar reports that the mean of that year category is significantly (one-sample t-test, p<0.01) different from the mean of respectively one, two, three or four year categories lower. The asterisks indicate the significance level of the t-test, where *: p<0.1, **: p<0.05, ***: p< Age categories younger than 4 years old and older than 12 years old are omitted due to a low number of observations for each category (N<100). All t-tests conducted have degrees of freedom>

24 Figure 4 Bar chart assessment levels for Dutch students and refugees per countries of origin Somalia Iraq Syria Eritrea Afghanistan Iran All refugee students Regular students S1a S1a/S1b S1b S1b/S1c S1c S1c/S2 S2 S2/S3 S3 0% 20% 40% 60% 80% 100% Note: this figure displays the stack bar charts of the division of students over assessment levels. Each row represents the division of another group. 24

25 Table 5 OLS regressions of predicted mean test score per country of origin (2a) Iran (2b) Iraq (2c) Afghanistan (2d) Somalia (2e) Syria (2f) Eritrea Iran 4.275*** (0.663) Iraq (0.418) Afghanistan 2.296*** (0.521) Somalia *** (0.389) Syria (0.771) Eritrea (1.050) Constant *** *** *** *** *** *** Note: This table reports the OLS regression outcomes of the predicted mean test score of a country of origin on the predicted mean test score of the other countries. Each column represents a separate regression. E.g. column (2a) shows the regression outcome for when the predicted mean test score of Iran is the dependent variable and the predicted mean test score of Iraq, Afghanistan, Somalia, Syria and Eritrea is the independent variable. Standard errors are given in parentheses. The asterisks indicate the significance level, where *: p<0.1, **: p<0.05, ***: p<

26 (Predicted) mean test score (Predicted) mean test score Figure 5 Mean test scores and predicted mean test scores per age at time of arrival A) Mean test scores and predicted mean test scores of all refugees Age at arrival Mean test score Predicted mean test score - all refugees Note: This figure displays the mean test scores (blue line) and predicted mean test scores of all refugees (red line) per age at arrival category. The predicted mean test scores are calculated with the mean test score per assessment level of Dutch students as displayed in Table 3. The categories 0-4 years and years old at arrival are not displayed due to N<100. For the predicted mean test score, the category 4-5 years is also not displayed due to a N<100. B) Mean test scores and predicted mean test scores of all refugees, refugees who got an assessment and took the test and refugees who only got an assessment years 5-6 years 6-7 years 7-8 years 8-9 years 9-10 years years years Age at arrival Mean test score Predicted mean test score - all refugees Predicted mean test score - refugees who took the test Predicted mean test score - refugees who did not take the test Note: This figure displays actual mean test scores of refugees who took the test per age at arrival (blue line) and predicted mean test scores for all refugees (red line), for refugees who took the test (green line) and for refugees 26

27 who did not take the test but who received an assessment (yellow line) per age at arrival category. The predicted mean test score is calculated using the mean test scores per assessment level of Dutch students as displayed in Figure 3. Results are only displayed for N>