View in EDS HTML Full Text PDF Full Text

Discrimination against Students with Foreign Backgrounds: Evidence from Grading in Swedish Public High Schools

Saved in:
Bibliographic Details
Title: Discrimination against Students with Foreign Backgrounds: Evidence from Grading in Swedish Public High Schools
Language: English
Authors: Hinnerich, Bjorn Tyrefors, Höglin, Erik, Johannesson, Magnus
Source: Education Economics. 2015 23(6):660-676.
Availability: Routledge. Available from: Taylor & Francis, Ltd. 325 Chestnut Street Suite 800, Philadelphia, PA 19106. Tel: 800-354-1420; Fax: 215-625-2940; Web site: http://www.tandf.co.uk/journals
Peer Reviewed: Y
Page Count: 17
Publication Date: 2015
Document Type: Journal Articles
Reports - Research
Education Level: High Schools
Secondary Education
Descriptors: Foreign Countries, Public Schools, High Schools, Educational Discrimination, High School Students, Grading, Minority Group Students, Scores, National Competency Tests, Statistical Analysis, Regression (Statistics), Robustness (Statistics), Least Squares Statistics
Geographic Terms: Sweden
DOI: 10.1080/09645292.2014.899562
ISSN: 0964-5292
Abstract: We rigorously test for discrimination against students with foreign backgrounds in high school grading in Sweden. We analyse a random sample of national tests in the Swedish language graded both non-blindly by the student's own teacher and blindly without any identifying information. The increase in the test score due to non-blind grading is significantly higher for students with a Swedish background. This discrimination effect is sizeable, about 10% of the mean or 20% of the standard deviation of the blind test score.
Abstractor: As Provided
Number of References: 36
Entry Date: 2015
Accession Number: EJ1076235
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwHynOUuYNYNq5YN78rxUkCGAAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDAwjL7pDGJq-oi1YMgIBEICBm_gq8O1GhAhBhkoarUxOHTzqF2yGei1rKNwI7cc3vfz8Jw6GZRjeJWJUXMndT1JbIMr0DlUuYyYltRjtJKzICLTPcjQvenpkCqshfxKu6FEsez5cgG6lwV2MMO_ISWoyDeBH32xb66s71xHhNey_ZfTOh7ffQlRSE-U3ZRRYdMCqfwy-1UcJIn445UN9iXLKxWnwU5SGkyHUXdXf
Text:
  Availability: 1
  Value: <anid>AN0110071015;ede01dec.15;2019Feb20.14:06;v2.2.500</anid> <title id="AN0110071015-1">Discrimination against students with foreign backgrounds: evidence from grading in Swedish public high schools. </title> <p>We rigorously test for discrimination against students with foreign backgrounds in high school grading in Sweden. We analyse a random sample of national tests in the Swedish language graded both non-blindly by the student's own teacher and blindly without any identifying information. The increase in the test score due to non-blind grading is significantly higher for students with a Swedish background. This discrimination effect is sizeable, about 10% of the mean or 20% of the standard deviation of the blind test score.</p> <p>Keywords: discrimination; education; grading; teacher bias</p> <hd id="AN0110071015-2">1. Introduction</hd> <p>Several studies have suggested that minority groups are discriminated against in the labour market and elsewhere (see the review by Anderson, Fryer, and Holt [<reflink idref="bib2" id="ref1">2</reflink>]). Depending on the historical and country-specific context, different groups may suffer from discrimination. In the USA, race, and more specifically black and white outcome differentials, has been studied. In Sweden, segregation, unequal schooling, and labour market outcomes are related to country of origin or 'foreignness'.[<reflink idref="bib1" id="ref2">1</reflink>] For example, Szulkin and Jonsson ([<reflink idref="bib34" id="ref3">34</reflink>], 1) note that 'the development during the 1990s in combination with a strong and growing residential segregation based on country of birth and income ... have led to concentrated disadvantage in many local areas'. Moreover, there is evidence of discrimination against foreigners, mostly non-Europeans, in both the labour market and the housing market (see e.g. Ahmed and Hammarstedt [<reflink idref="bib1" id="ref4">1</reflink>]; Arai and Skogman Thoursie [<reflink idref="bib3" id="ref5">3</reflink>]; Carlsson and Rooth [<reflink idref="bib10" id="ref6">10</reflink>]). To what extent discrimination occurs within the education system is less established (see the review of Farkas [<reflink idref="bib15" id="ref7">15</reflink>]). More recent works show that achievement gaps can be explained by the minority–majority match of the student and the teacher (Dee [<reflink idref="bib11" id="ref8">11</reflink>], [<reflink idref="bib12" id="ref9">12</reflink>], [<reflink idref="bib13" id="ref10">13</reflink>]). However, discrimination in the direct or active form (e.g. same student performance but different grades) is only one potential explanation. Alternatively, the interaction between the majority teacher and minority students may diminish the students' performance because, for example, students may have to adjust their efforts to correspond to the stereotypes held by the teacher.[<reflink idref="bib2" id="ref11">2</reflink>] As Dee ([<reflink idref="bib12" id="ref12">12</reflink>], 164) states, 'the exact design of ... policies also require a clear understanding of the underlying structural mechanisms that make these student–teacher interactions relevant in the first place'.</p> <p>This paper aims to investigate whether the direct form of discrimination against students with foreign backgrounds exists in high school grading by comparing non-blind and blind grading of a test with the same content.[<reflink idref="bib3" id="ref13">3</reflink>] We find that students with foreign backgrounds are directly discriminated against. The point estimates are largest for non-Europeans, which resembles the result from labour market studies in Sweden (see e.g. Arai and Skogman Thoursie [<reflink idref="bib3" id="ref14">3</reflink>]). Since we only observe country of birth, our results are only indicative with respect to ethnic discrimination. Our study suggests that discrimination is one explanation for educational inequality across country of origin, and based on an efficiency perspective, we propose that a blind grading system be introduced, particularly since test scores on national tests, in practice, constitute a large portion of students' final grades, which determine admission to university.</p> <p>The methodology of varying the degree of 'blindness' to detect discrimination has previously been used in the economics literature by Blank ([<reflink idref="bib7" id="ref15">7</reflink>]), Goldin and Rouse ([<reflink idref="bib18" id="ref16">18</reflink>]), and Lavy ([<reflink idref="bib23" id="ref17">23</reflink>]). Of particular interest, Lavy ([<reflink idref="bib23" id="ref18">23</reflink>]) compares two different test scores for the same individuals. He finds statistically significant discrimination against men on all the examined tests. A limitation of Lavy's study, however, is that it does not compare blind and non-blind grading of the same test. Hence, we contribute to the literature by grading a test with the same content twice, bringing the method closer to the ideal test setting. Hinnerich, Höglin, and Johannesson ([<reflink idref="bib20" id="ref19">20</reflink>]) use the same student test score data that we use for the present paper to test for gender discrimination but did not find any significant effects. To test for discrimination against students with foreign backgrounds, we collected new data on country of birth.</p> <p>Our study was carried out in Swedish public high schools, where national tests are given in Swedish, English, and maths. These tests are graded non-blindly, usually by the student's own teacher. The grading <emph>should</emph> be based on the extensive written guidelines provided for each test, test date, and test topic, which state the prerequisites for each grade. To test for discrimination, we used a random sample of written tests in Swedish for the academic year 2005/2006. The students in our sample were classified by Swedish background (<emph>n</emph> = 1423) or foreign background (<emph>n</emph> = 199) according to the classification used by Statistics Sweden ([<reflink idref="bib29" id="ref20">29</reflink>]). After rewriting the tests on a word processor and removing the students' identities and teachers' notes, but nothing else, the tests were regraded blindly by a group of teachers who were hired from a teachers' agency. Since we typeset the exam, strictly speaking, our experiment involves a second manipulation, in addition to the external grading. Given the teachers' notes on the original essays, we believed that retyping the student's exams was the only practical option for duplicating the exams, and upon inspecting the original essays, we observed no major differences in handwriting quality across the groups, as discussed in Section 2.</p> <p>The difference between the non-blind grade and the blind grade is a measure of the bias induced by non-blind grading. In the absence of discrimination, the grades should be the same for the different groups. However, we find a sizeable and significant difference between students with Swedish backgrounds and students with foreign backgrounds. Non-blind grading increased the test score by 16% for students with Swedish backgrounds compared with 4% for students with foreign backgrounds.</p> <p>Our paper is related to the recent work of Hanna and Linden ([<reflink idref="bib19" id="ref21">19</reflink>]), who test for discrimination with respect to age, gender, and caste in India. They find that teachers assigned lower scores to exams from children in the lower caste. Our results are consistent with their findings in the sense that both caste and foreign background may be indicators of social status and/or ethnicity. Sprietsma ([<reflink idref="bib28" id="ref22">28</reflink>]) finds evidence of grading discrimination based on Turkish as compared with German names using a field experiment. However, recently, van Ewijk ([<reflink idref="bib14" id="ref23">14</reflink>]), using a similar approach, only finds a small insignificant discriminatory effect on the minority group. Moreover, Kiss ([<reflink idref="bib22" id="ref24">22</reflink>]), finds that second-generation immigrants with lower test scores were affected by grade discrimination in primary education. In a recent study, Feld, Salamanca, and Hamermesh ([<reflink idref="bib16" id="ref25">16</reflink>]) note that the literature on discrimination treats outcomes as relativistic and does not distinguish between favouritism and discrimination. They conduct a randomised field experiment in grading and find evidence of favouritism but not of discrimination. Our results may be consistent with this finding since, but given our design, we cannot determine whether students with Swedish backgrounds are being favoured or whether students with foreign backgrounds are being negatively discriminated against.</p> <p>It is common in economics to distinguish between statistical discrimination (Arrow [<reflink idref="bib4" id="ref26">4</reflink>], [<reflink idref="bib5" id="ref27">5</reflink>]; Phelps [<reflink idref="bib26" id="ref28">26</reflink>]) and taste-based discrimination (Becker [<reflink idref="bib6" id="ref29">6</reflink>]). In a strict sense, it should not be possible to observe statistical discrimination in our setting because the outcome to be evaluated (the test) is observable to the teacher. However, one cannot rule out that the grading is still affected by beliefs about ability or performance that are associated with particular foreign backgrounds.[<reflink idref="bib4" id="ref30">4</reflink>] Teachers who are unsure of how to grade a test may, for instance, be affected by the perceived ability of the student. Taste-based discrimination involves preferring one group over another group, and this type of discrimination is a potential mechanism for our observed effect. Discriminating based on taste would not be costly for the teachers in our setting, which is an important difference compared to the contexts of the labour market. It is also possible that our observed discrimination effect operates through other, more indirect channels if foreign background is associated with other factors that may bias the non-blind grading. However, the discrimination effect is robustly significant after controlling for a host of factors, such as gender, year of birth, and school fixed effects.</p> <p>Below, we first outline the design of our study and the methodology and data used to test for discrimination. Thereafter, we present our findings, and we then present extensions and further robustness tests and finally, we conclude.</p> <hd id="AN0110071015-3">2. Design of the study</hd> <p></p> <hd id="AN0110071015-4">2.1. Sample, foreign backgrounds, and blind grading</hd> <p>Compulsory national tests in the Swedish high school system are given in the core subjects of Swedish, English, and maths.[<reflink idref="bib5" id="ref31">5</reflink>] We collected data on the extensive written test in Swedish for the academic year 2005/2006. In this test, students are asked to write an essay based on one of nine topics. Each test is graded on a four-tiered scale, Fail, Pass, Pass with Distinction, and Excellent, and translated into the cardinal scale, 0, 10, 15, and 20, respectively (this scale is used to calculate the grade point average in the Swedish high school system).</p> <p>The national tests are graded by the student's own teacher. For every specific test and topic, the teachers are given extensive written guidelines stating the prerequisites for each grade, but they have great discretion in the actual grading since no measures are taken by the national authorities to ensure that the guidelines are followed. The test aims to help teachers measure the knowledge criteria that should determine students' final grades. Grades should be set according to absolute knowledge criteria, and the core subjects have nationally stipulated prerequisites for each grade. Hence, conditional on the level of knowledge, grades must not reflect participation, diligence, or ambition, but again no measures are taken by the national authorities to ensure compliance with the guidelines. Final grades are an important factor in university admission. It is also worth noting that there are no formal incentives for teachers when teaching or grading, since teachers are not reviewed by the authorities. That said, since the 1990s, students in Sweden have had the right to choose schools, and schools are paid for each student via a governmental voucher system. Thus, schools compete for students. Teachers may, therefore, be incentivised to inflate grades to increase the competitiveness of their schools (Jacob and Levitt [<reflink idref="bib21" id="ref32">21</reflink>]), and concerns have been raised about grade inflation due to this system (Wikström and Wikström [<reflink idref="bib36" id="ref33">36</reflink>]). Thus, school competition might trigger grade inflation.</p> <p>We used a random sample of 2880 students who were eligible to take the test. Of the 2880 students in the sample, we received complete information, which included the actual test, the test score, and the student's identity, for 1713 students (a response rate of 59%). The main reason for non-response was absenteeism on the test day (Swedish National Agency for Education [<reflink idref="bib31" id="ref34">31</reflink>]). For each student, we collected data on the country of birth of the student and both parents. We collected data manually from the Swedish Tax Agency for a very small number of individuals at a time by exercising the right of public access to official documents. To classify students into students with Swedish backgrounds and students with foreign backgrounds, we used the classification used in Statistics Sweden ([<reflink idref="bib29" id="ref35">29</reflink>]). If the student was born in Sweden and at least one of the parents was born in Sweden, the student was classified as having a Swedish background. If the student was born abroad or if both parents were born abroad, the student was classified as having a foreign background. We obtained data on the backgrounds of 1622 of the 1713 students in our sample. Our analyses were performed for the sample of 1622 students, of which 1423 were of Swedish background and 199 were of foreign background.</p> <p>Some caveats regarding the definition of foreign background in this study should be noted. First, a student born abroad was always classified as a student with foreign background, irrespective of the parent's background. For example, adopted children that are likely to share the adoptees' characteristics, except for physical appearance, were therefore classified into the treatment group. Since there is no historical racial conflict in Sweden, defining foreign background in this way might bias our result downwards. In the robustness section, we try to address this problem. Moreover, if the student is born in Sweden and has only one Swedish parent, she/he was classified as belonging to the control group; again this classification may affect our result. We address this possibility in the robustness section.</p> <p>To estimate the effect of discrimination based on foreign background, we estimated the difference between the non-blind and the blind grade. The non-blind grade is the grading of the test by the student's own teacher. To obtain the blind grade, we first rewrote all tests on a word processor (we hired 12 rewriters) and deleted the students' identities as well as their teachers' notes (the rewriters were not given any information regarding the purpose of the study). Nothing else was changed.</p> <p>The tests were then regraded blindly by 42 teachers who were hired from a teachers' agency. The external graders had no information regarding the purpose of the study. To our knowledge, external grading of national tests has not been carried out on a large scale before in Sweden.[<reflink idref="bib6" id="ref36">6</reflink>] To mimic a normal grading situation for national tests, we established some requirements. We required external graders to have graded national tests in Swedish before and to be representative of the composition of gender, certification, and country of birth. The external graders were provided with the official written guidelines, which stated the prerequisites for each grade and topic in the same manner as for the teacher who graded the original test. The external graders were told to correct the essays using the same effort and time that they otherwise would use when grading national tests in Swedish and were paid accordingly. Thus, we did our best to reproduce a similar environment for grading.</p> <hd id="AN0110071015-5">2.2. Identification and hypothesis</hd> <p>Because we had blind (B) and non-blind (NB) grades on tests with the same content (<emph>i</emph>), we perform a difference-in-difference analysis. The major advantage of such an analysis is that the unobserved quality of a test can be controlled for, and since the regrader does not know the students' foreign backgrounds, we can estimate the discrimination effect (<emph>β</emph>) by ordinary least squares (OLS) with</p> <p>(<reflink idref="bib1" id="ref37">1</reflink>)</p> <p>Graph</p> <p>where and Foreign = 1 if the student has a foreign background and equals zero otherwise.</p> <p>If no discrimination is present, then students of different backgrounds producing the same quality essays should receive the same grade. If the grades differ, one group was discriminated against. Thus, we interpreted <emph>β</emph> as a discrimination effect. If <emph>β</emph> is negative, then students with foreign backgrounds are discriminated against, and if <emph>β</emph> is positive, the opposite holds true.</p> <p>Our empirical set-up is based on comparing the grading of an essay by the student's own teacher to the grading of the same essay, but typeset, by an external examiner with no information about the student. Thus, differences in test scores can be attributed to whether the grading was blind, whether they essays were typeset, or differences in the framing of the grading situation.</p> <p>The external setting is likely to result in more conservative grading (see e.g. Lavy [<reflink idref="bib23" id="ref38">23</reflink>]), potentially owing to the framing of the situation. We tried to minimise the differences in framing by providing the official written guidelines. The external graders were told to correct the essays using the same effort and time that they otherwise would use when correcting national tests in Swedish, and they all had the experience of grading national tests under normal circumstances before. Moreover, for the difference in framing to be a threat to internal validity, this difference would furthermore need to be related to the unknown background of the student, which seems unlikely.</p> <p>If the essay carried some information about a foreign background, then our estimate is biased towards zero, given that the external graders are representative of the teaching population and should be interpreted as a lower bound. We tried to ensure the representativeness of the external grader sample by hiring external graders who work as teachers and requiring them to have graded national tests in Swedish before. Moreover, we ensured the representativeness of the external grader sample in terms of gender, certification, and country of birth.</p> <p>It should also be acknowledged that we cannot separate the effect of grading due to blindness from the effect of grading from the external grader. The bias may stem from the personal relation with the student in combination with non-blindness. However, Lavy ([<reflink idref="bib23" id="ref39">23</reflink>]) partly addresses this question with respect to gender. He finds that external graders grade non-blindly as conservative as they would when they grade blindly; however, the difference in grading across gender is diminished when the external examiner grades non-blindly, providing evidence of grader-induced discrimination due to the grader's knowledge of the student's gender.</p> <p>From a legal perspective, it is illegitimate to put weight on the quality of handwriting, given the quality of the test.[<reflink idref="bib7" id="ref40">7</reflink>] However, if teachers grade with a bias, teachers may punish inferior handwriting when grading. Reviewing the sample of the handwritten essays, we did not observe any differences in handwriting quality between the groups. However, there appears to be a clear difference in handwriting quality across gender.</p> <p>Our estimate of the discrimination effect could still be biased through selection. However, only 6 of 100 schools did not respond or submitted no information on the tests, which makes selection very unlikely to be problematic at the school level. For the absence of students from the test to affect the estimation of the discrimination effect, the potential difference in test scores would have to be related to the students' foreign background. In terms of internal validity, it is not a problem for our study if this group performed differently from the students taking the test. Moreover, students in Swedish high schools <emph>cannot</emph> choose teachers, but they can choose schools.</p> <p>We test the hypothesis that students with foreign backgrounds are discriminated against by testing whether the increase in grades due to non-blind grading is larger for students with Swedish backgrounds than for students with foreign backgrounds using a cardinal scale for grades.[<reflink idref="bib8" id="ref41">8</reflink>] Moreover, we further investigate the robustness of and the mechanism explaining our results. In the regression section, we follow the conventions in the testing literature by using scores standardised to a distribution with a zero mean and a unit standard deviation, meaning that the discrimination effect should be interpreted as the share of a standard deviation of the blind test score.</p> <hd id="AN0110071015-6">3. Results</hd> <p></p> <hd id="AN0110071015-7">3.1. Descriptive results and tests</hd> <p>The mean test scores are shown in Table 1.[<reflink idref="bib9" id="ref42">9</reflink>] On average, the non-blind test scores are 8% higher for students with Swedish backgrounds compared to students with foreign backgrounds. However, when examining the blind test scores, the entire difference between the two student groups disappears.[<reflink idref="bib10" id="ref43">10</reflink>]</p> <p>Table 1. Test scores and differences in test scores.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td>Sample statistics</td><td><italic>n</italic></td><td>Mean</td><td>SD</td></tr></thead><tbody><tr><td><italic>Non-blind test score</italic></td><td /><td /><td /></tr><tr><td>Students with Swedish background</td><td char=".">1423</td><td char=".">12.122</td><td char=".">4.975</td></tr><tr><td>Students with foreign background</td><td char=".">199</td><td char=".">11.181</td><td char=".">5.099</td></tr><tr><td>Difference</td><td /><td char=".">0.941</td><td /></tr><tr><td><italic>p</italic>-value of diff. (<italic>t</italic>-test)</td><td /><td char=".">0.015</td><td /></tr><tr><td><italic>p</italic>-value of diff. (Mann–Whitney test)</td><td /><td char=".">0.008</td><td /></tr><tr><td><italic>Blind test score</italic></td><td /><td /><td /></tr><tr><td>Students with Swedish background</td><td char=".">1423</td><td char=".">10.457</td><td char=".">5.509</td></tr><tr><td>Students with foreign background</td><td char=".">199</td><td char=".">10.754</td><td char=".">5.383</td></tr><tr><td>Difference</td><td /><td char=".">−0.297</td><td /></tr><tr><td><italic>p</italic>-value of diff. (paired <italic>t</italic>-test)</td><td /><td char=".">0.478</td><td /></tr><tr><td><italic>p</italic>-value of diff. (Mann–Whitney test)</td><td /><td char=".">0.651</td><td /></tr><tr><td><italic>Non-blind test score</italic> <italic>− blind test score</italic></td><td /><td /><td /></tr><tr><td>Students with Swedish background</td><td char=".">1423</td><td char=".">1.665</td><td char=".">5.743</td></tr><tr><td>Students with foreign background</td><td char=".">199</td><td char=".">0.427</td><td char=".">5.500</td></tr><tr><td>Difference</td><td /><td char=".">1.238</td><td /></tr><tr><td><italic>p</italic>-value of diff. (<italic>t</italic>-test)</td><td /><td char=".">0.003</td><td /></tr><tr><td><italic>p</italic>-value of diff. (Mann–Whitney test)</td><td /><td char=".">0.003</td><td /></tr></tbody></table> </ephtml> </p> <p>The equal test performance between students with foreign backgrounds and Swedish students in the blind grading might seem suspicious. However, to our knowledge, the only data on test results for Swedish high school students are provided in the 2008 Trends in International Mathematics and Science Study (TIMSS) report, which indicates that students with foreign backgrounds perform worse in blindly graded tests (Mullis et al. [<reflink idref="bib24" id="ref44">24</reflink>]). However, the report covers only maths and physics. Recently, the Swedish Schools Inspectorate ([<reflink idref="bib32" id="ref45">32</reflink>], [<reflink idref="bib33" id="ref46">33</reflink>]) has begun to partly evaluate schools. For 2011, but not for 2010, we were able to obtain similar information. We can replicate the results from the 2008 TIMS report in mathematics (i.e. students with foreign backgrounds perform worse in blindly graded tests). However, in Swedish, there is no significant difference between Swedish students and students with foreign backgrounds. However, as noted in Section 2.2, even though students' absence on the test day causes no threat to the internal validity of the study, the lack of a difference in blind grades between Swedish and foreign students might indicate a threat to external validity.</p> <p>Non-blind grading thus favours students with Swedish backgrounds. Non-blind grading increased the test score by 16% for students with Swedish backgrounds compared to only 4% for students with foreign backgrounds.</p> <hd id="AN0110071015-8">3.2. Main regression results</hd> <p>To examine our results further, we run OLS regressions with the difference between the non-blind and the blind grade as the dependent variable. We use test scores standardised to a distribution with a zero mean and a unit standard deviation. These results are shown in Table 2. In the first model, we only include a dummy variable for foreign background to measure the discrimination effect. To account for a possible correlation in observations for different schools and different external graders, we estimate standard errors based on two-way clustered standard errors at the school and external grader level (Cameron, Gelbach, and Miller [<reflink idref="bib9" id="ref47">9</reflink>]; Thompson [<reflink idref="bib35" id="ref48">35</reflink>]).[<reflink idref="bib11" id="ref49">11</reflink>] Students with a foreign background are discriminated against, and the point estimate shows a statistically significant discrimination effect of about 0.24 of a standard deviation. To check for the robustness of our result, we add controls sequentially. We add a dummy for gender and student year of birth; however, the discrimination effect is unchanged, as shown in Table 2, columns 2–3.</p> <p>Table 2. Regression results on the difference between the non-blind and blind test scores.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td>Variables</td><td>(1)</td><td>(2)</td><td>(3)</td><td>(4)</td><td>(5)</td><td>(6)</td><td>(7)</td><td>(8)</td></tr></thead><tbody><tr><td>Foreign background</td><td>−0.242***</td><td>−0.245***</td><td>−0.230***</td><td>−0.222***</td><td>−0.233***</td><td>−0.224***</td><td>−0.211***</td><td>−0.168***</td></tr><tr><td /><td>(0.069)</td><td>(0.069)</td><td>(0.066)</td><td>(0.056)</td><td>(0.054)</td><td>(0.051)</td><td>(0.056)</td><td>(0.058)</td></tr><tr><td>Male</td><td>No</td><td>−0.034</td><td>−0.030</td><td>0.002</td><td>0.002</td><td>0.002</td><td>−0.015</td><td>−0.022</td></tr><tr><td /><td /><td>(0.051)</td><td>(0.051)</td><td>(0.047)</td><td>(0.046)</td><td>(0.045)</td><td>(0.046)</td><td>(0.049)</td></tr><tr><td>Student birth year</td><td>No</td><td>No</td><td>0.146**</td><td>0.163**</td><td>0.163**</td><td>0.175**</td><td>0.163**</td><td>0.129**</td></tr><tr><td /><td /><td /><td>(0.071)</td><td>(0.070)</td><td>(0.071)</td><td>(0.071)</td><td>(0.069)</td><td>(0.062)</td></tr><tr><td>Regrader fixed effect</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Rewriter fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Programme fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Topic fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td></tr><tr><td>School fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td></tr><tr><td><italic>R</italic><sup>2</sup></td><td>0.005</td><td>0.006</td><td>0.009</td><td>0.102</td><td>0.115</td><td>0.116</td><td>0.133</td><td>0.240</td></tr></tbody></table> </ephtml> </p> <p>Notes: Dependent variables are standardised scores. A constant is always included. Two-way clustered standard errors are reported in parentheses at the school and regrader level (Cameron, Gelbach, and Miller [<reflink idref="bib9" id="ref50">9</reflink>]; Thompson [<reflink idref="bib35" id="ref51">35</reflink>]). <emph>n</emph> = 1622 in all specifications. The 'foreign background' variable measures the effect of discrimination. *Significance at the 10% level. **Significance at the 5% level. ***Significance at the 1% level.</p> <p>Our experimental manipulation was to rewrite and regrade the test. Thus, it is natural to include fixed effects for the rewriter and the external grader. As shown in columns 4–5, our results are robust to this inclusion. Moreover, if students with foreign backgrounds choose academic/vocational training to a larger extent, this might have influenced the results if there was a relatively more liberal way of grading in one of the programmes. Adding a fixed effect for academic or vocational programmes, as shown in column 6, however, does not change the results.</p> <p>In general, a major concern with any non-blind/blind set-up is that the blind grader also can at least partly observe the relevant characteristics of the student. The external grader may have been able to infer the background of the student based on the text of essay. Such an effect could lead to a bias towards zero in our estimated discrimination effect, making our estimate too small in absolute terms, as discussed in section 2. As the students chose from nine different topics, the choice of topic may have revealed some information about the student's foreign background, as indicated by van Ewijk ([<reflink idref="bib14" id="ref52">14</reflink>]).[<reflink idref="bib12" id="ref53">12</reflink>] Thus, including a control for the topic of the essay could increase the discrimination effect if the topic has informational value about the background of the student.[<reflink idref="bib13" id="ref54">13</reflink>] As shown in Table 2, column 7, we include topic-specific effects, but the estimate is left rather unchanged.</p> <p>Lastly, since the 1990s, students in Sweden have had the right to choose schools, and schools are paid for each student via a governmental voucher system. Thus, schools compete for students. Teachers may, therefore, be incentivised to inflate grades to increase the competitiveness of their schools. Moreover, the ability to choose schools and school competition over students has led to increased socioeconomic and ethnic sorting (Böhlmark and Holmlund [<reflink idref="bib8" id="ref55">8</reflink>]). If segregation and grade inflation are related, then adding school fixed effects should change our estimate. The last column presents the results, including school fixed effects. Indeed, including school fixed effects results in a drop in the discrimination effect of about four percentage points. This result suggests that part of the discrimination effect may be an indirect effect of different grading standards between schools with different fractions of students with foreign backgrounds. However, the point estimate of the discrimination effect with school fixed effects is not significantly different from the point estimate without school fixed effects.</p> <hd id="AN0110071015-9">3.3. Extensions and robustness</hd> <p>One major concern with OLS in combination with the use of the 0–20 GPA scale is the sensitivity to large differences in grading. For example, four students received the highest grade by the teacher but the lowest by the external grader. One way to investigate the robustness of our estimate is to redefine our outcome to an outcome that places equal weight on deviations. In Table 3, column 1, we use the same specifications as in Table 2, column 8, but we define the outcome to be one if the external grade was lower than the grade set by the teacher and zero otherwise. Students with a foreign background had about a six percentage point lower probability of receiving a lower grade by the external grader, which is consistent with previous findings.</p> <p>Table 3. Extension and robustness.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td /><td>1</td><td>2</td><td>3</td><td>4</td><td>5</td></tr><tr><td /><td>Probability of being downgraded by external grader</td><td>Interval regression</td><td>Check for ability discrimination</td><td>Full sample. Subdivision of treatment and control group</td></tr></thead><tbody><tr><td>Foreign background</td><td>−.058** (0.029)</td><td>−1.326** (0.578)</td><td>−.218*** (0.0510)</td><td>−0.185** (0.089)</td><td>−0.198** (0.087)</td></tr><tr><td>Adopted</td><td>No</td><td>No</td><td>No</td><td>−0.012 (0.280)</td><td>−0.028 (0.281)</td></tr><tr><td>Swede with one foreign parent</td><td>No</td><td>No</td><td>No</td><td>No</td><td>−0.137 (0.128)</td></tr><tr><td><italic>N</italic></td><td>1622</td><td>3272</td><td>1622</td><td>1622</td><td>1622</td></tr><tr><td><italic>R</italic><sup>2</sup></td><td>0.200</td><td /><td>0.280</td><td>0.240</td><td>0.241</td></tr></tbody></table> </ephtml> </p> <p>Notes: Dependent variables are standardised scores. A constant and the full set of controls in Table 2, column 8, are always included. Two-way clustered standard errors are reported in parentheses at the school and regrader level (Cameron, Gelbach, and Miller [<reflink idref="bib9" id="ref56">9</reflink>]; Thompson [<reflink idref="bib35" id="ref57">35</reflink>]). Regression results on the difference between the non-blind and blind test scores. The 'foreign background' variable measures the effect of discrimination. *Significance at the 10% level. **Significance at the 5% level. ***Significance at the 1% level.</p> <p>Our dependent variable is not continuous, as we only observed four possible grades. To test the importance of this assumption, we also estimate an interval regression, where we place the bounds at the midpoint between each of the grades. The estimate is shown in Table 3, column 2, and is measured in non-standardised test scores. Comparing the size of the effect (−1.326) with the standard deviations in Table 1, we find that the size of the effect amounted to about 25% of a standard deviation.[<reflink idref="bib14" id="ref58">14</reflink>]</p> <p>The estimate of interest might be biased owing to other uncontrolled factors related to foreign background. For example, if students with foreign backgrounds were less proficient in Swedish and non-blind graders tended to be generous to less-proficient students, our interpretation of discrimination against students with foreign backgrounds is flawed. One way of testing for this possibility is to assume that the blind test score is an unbiased measure of ability and to define dummies for the different blind grades. We then estimate (Table 3, column 3) the effect of foreign background on the <emph>non-blind</emph> test score when controlling for the categories of the blind grade. There is no evidence that our estimate of interest is sensitive to generosity to less-proficient students.</p> <p>Because of the nature of the field experiment, we cannot investigate the mechanisms that drive the results on to a very thorough level. However, adopted children with two parents born in Sweden are more likely to share the 'foreign' physical attribute with other students from foreign backgrounds, but not other attributes, such as religion, language, set of attitudes, and behaviours. We define adopted children (<emph>n</emph> = 38) as those born outside Sweden but whose parents were both born in Sweden. We separate this group from students with foreign backgrounds, including them within their own category in the regression. The point estimates in Table 3, column 4, show little evidence of discrimination against this group. This finding supports the hypothesis that the discrimination is not racial.</p> <p>Moreover, our control group – students with Swedish backgrounds – includes students who were born in Sweden but who had one parent born abroad (<emph>n</emph> = 115). The regression presented in Table 3, column 5, includes a dummy for this group as well. Although imprecisely measured, the point estimate is in line with a sizeable discrimination effect against this group as well.</p> <p>Bias may arise if the external graders did not take their task seriously. For example, assuming an extreme situation in which the external grader passed all tests and the teacher gave each a grade according to the quality of essay. Then, our point estimate would suggest that students with foreign backgrounds are discriminated against because they performed worse. Several indicators suggest that such potential bias is not a major problem. First, the correlation (0.41) between the original grade and the blind grade is high and significant at the 1% level. Second, the scores are identical for about 46% of the students, making it unlikely that external graders did not take the task seriously on a large scale. Finally, the lack of a finding of gender discrimination, consistent with Hinnerich, Höglin, and Johannesson ([<reflink idref="bib20" id="ref59">20</reflink>]), also speaks against this possibility.</p> <p>Other studies on the Swedish labour market provide some evidence that some groups (Arabs and Africans) face more discrimination than others (e.g. Arai and Skogman Thoursie [<reflink idref="bib3" id="ref60">3</reflink>]). In our study, the treatment group consisted of only 199 students with foreign backgrounds. Because of the limited sample size, we were unable to categorise members of the treatment group by country of origin. Moreover, we only categorised students by the country in which the student, the mother, and the father were born, not other variables of importance. We divide the students with foreign backgrounds into European (<emph>n</emph> = 87) and non-European (<emph>n</emph> = 112).[<reflink idref="bib15" id="ref61">15</reflink>] We would expect non-Europeans to experience higher levels of discrimination than Europeans – consistent with the results from studies on the labour market.</p> <p>Table 4 shows the OLS regression results with the same procedure of adding controls as in Table 2. The point estimate of the discrimination effect is consistently higher for students with non-European backgrounds in all specifications. For students with non-European backgrounds, the point estimate ranges between 0.219 and 0.302 standard deviations, and all estimates are significant at the 5% level. As expected, students with European backgrounds experienced lower levels of discrimination, and the point estimates are insignificant. Thus, there is evidence that non-Europeans are subject to more discrimination than Europeans, not only in the labour market, but also in Swedish schools. However, we cannot reject the null hypothesis of no difference between students with European and non-European backgrounds, which is not very surprising, given the small sample sizes of these two groups.</p> <p>Table 4. Regression results on the difference between the non-blind and blind test scores when the 'foreign background' variable is divided into non-European and European.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td>Variables</td><td>(1)</td><td>(2)</td><td>(3)</td><td>(4)</td><td>(5)</td><td>(6)</td><td>(7)</td><td>(8)</td></tr></thead><tbody><tr><td>Non-European background (<italic>β</italic>1)</td><td>−0.287***</td><td char=".">−0.290***</td><td char=".">−0.266***</td><td char=".">−0.286***</td><td char=".">−0.302***</td><td char=".">−0.292***</td><td char=".">−0.273***</td><td char=".">−0.219**</td></tr><tr><td /><td char=".">(0.097)</td><td char=".">(0.098)</td><td char=".">(0.096)</td><td char=".">(0.087)</td><td char=".">(0.083)</td><td char=".">(0.081)</td><td char=".">(0.079)</td><td char=".">(0.091)</td></tr><tr><td>European background (<italic>β</italic>1)</td><td char=".">−0.181</td><td char=".">−0.182</td><td char=".">−0.183</td><td char=".">−0.136</td><td char=".">−0.139</td><td char=".">−0.132</td><td char=".">−0.129</td><td char=".">−0.104</td></tr><tr><td /><td char=".">(0.116)</td><td char=".">(0.117)</td><td char=".">(0.113)</td><td char=".">(0.109)</td><td char=".">(0.110)</td><td char=".">(0.109)</td><td char=".">(0.116)</td><td char=".">(0.100)</td></tr><tr><td>Male</td><td>No</td><td char=".">−0.035</td><td char=".">−0.031</td><td char=".">0.001</td><td char=".">0.001</td><td char=".">0.001</td><td char=".">−0.016</td><td char=".">−0.024</td></tr><tr><td /><td /><td char=".">(0.051)</td><td char=".">(0.051)</td><td char=".">(0.046)</td><td char=".">(0.045)</td><td char=".">(0.045)</td><td char=".">(0.046)</td><td char=".">(0.049)</td></tr><tr><td>Student birth year</td><td>No</td><td>No</td><td char=".">0.144**</td><td char=".">0.159**</td><td char=".">0.160**</td><td char=".">0.171**</td><td char=".">0.160**</td><td char=".">0.126**</td></tr><tr><td /><td /><td /><td char=".">(0.071)</td><td char=".">(0.069)</td><td char=".">(0.071)</td><td char=".">(0.070)</td><td char=".">(0.069)</td><td char=".">(0.062)</td></tr><tr><td>Regrader fixed effect</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Rewriter fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Programme fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Topic fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td></tr><tr><td>School fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td></tr><tr><td>Test of if <italic>β</italic>1 = <italic>β</italic>2 (<italic>p</italic>-value)</td><td char=".">0.518</td><td char=".">0.510</td><td char=".">0.608</td><td char=".">0.353</td><td char=".">0.309</td><td char=".">0.318</td><td char=".">0.370</td><td char=".">0.450</td></tr><tr><td><italic>R</italic><sup>2</sup></td><td char=".">0.006</td><td char=".">0.006</td><td char=".">0.010</td><td char=".">0.103</td><td char=".">0.116</td><td char=".">0.117</td><td char=".">0.134</td><td char=".">0.240</td></tr></tbody></table> </ephtml> </p> <p>Notes: Dependent variables are standardised scores. A constant is always included. Two-way clustered standard errors are reported in parentheses at the school and regrader level (Cameron, Gelbach, and Miller [<reflink idref="bib9" id="ref62">9</reflink>]; Thompson [<reflink idref="bib35" id="ref63">35</reflink>]). <emph>n</emph> = 1622 in all specifications. *Significance at the 10% level. **Significance at the 5% level. ***Significance at the 1% level.</p> <hd id="AN0110071015-10">4. Conclusions</hd> <p>We find a sizeable and robust discrimination effect against students with foreign backgrounds in grading of Swedish national test. Whether this result carries over to other subjects such as mathematics still needs to be investigated. The estimate is based on a representative sample of students from public Swedish high schools for a core subject (Swedish). From a policy perspective, it would be relatively easy to reduce educational inequality as measured by test scores on national tests by introducing a blind grading system. Since this study suggests that discrimination is one explanation for educational inequality, the introduction of blind grading is motivated by an efficiency perspective, particularly since scores on national tests constitute a large portion of students' final grades, which determine admission to university. However, because final subject grades are not based solely on national tests, discrimination may nevertheless affect the final subject grades through other means. To further decrease the possibility of discrimination, a system in which final grades are based solely on blindly graded tests and external examiners would be desirable.</p> <hd id="AN0110071015-11">Acknowledgements</hd> <p>We thank George Farkas and Tore Ellingsen, the editor Colin Green and two anonymous referees for their helpful comments, the Institute for Labour Market Policy Evaluation (IFAU), the Jan Wallander and Tom Hedelius Foundation, the Swedish Research Council, and the Swedish Council for Working Life and Social Research for financial support.</p> <hd id="AN0110071015-12">Appendix</hd> <p>Table A1. Ordered logit and probit: marginal effects of foreign background.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td /><td>1</td><td>2</td><td>3</td><td>4</td><td>5</td><td>6</td><td>7</td><td>8</td><td>9</td></tr><tr><td>Variables</td><td>Pr(−20)</td><td>Pr(−15)</td><td>Pr(−10)</td><td>Pr(−5)</td><td>Pr(0)</td><td>Pr(5)</td><td>Pr(10)</td><td>Pr(15)</td><td>Pr(20)</td></tr></thead><tbody><tr><td /><td><italic>Panel A: ordered logit</italic></td></tr><tr><td>Foreign background</td><td>0.000 (0.000)</td><td>0.002** (0.001)</td><td>0.014** (0.007)</td><td>0.022** (0.010)</td><td>0.020** (0.009)</td><td>−0.025** (0.011)</td><td>−0.025** (0.011)</td><td>−0.008** (0.004)</td><td>−0.001 (0.001)</td></tr><tr><td /><td><italic>Panel B: ordered probit</italic></td></tr><tr><td>Foreign background</td><td>0.000 (0.000)</td><td>0.003** (0.002)</td><td>0.015** (0.007)</td><td>0.021** (0.009)</td><td>0.018** (0.007)</td><td>−0.022** (0.009)</td><td>−0.024** (0.010)</td><td>−0.009** (0.004)</td><td>−0.001 (0.001)</td></tr></tbody></table> </ephtml> </p> <p>Notes: Dependent variable is the difference in scores, not normalised () and ranges from −20 to 20. A negative value indicates that the teacher's score is lower than the external grader's score. A positive value indicates downgrading by the external grader. The controls (as in Table 2, column 8) are included. Clustered standard errors at the school level are reported in parentheses at the school. <emph>n</emph> = 1622 in all specifications.</p> <p>*Significance at the 10% level.</p> <p>**Significance at the 5% level.</p> <p>***Significance at the 1% level.</p> <p>Table A2. Regression results on the difference between the non-blind and blind test scores.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td>Variables</td><td>(1)</td><td>(2)</td><td>(3)</td><td>(4)</td><td>(5)</td><td>(6)</td><td>(7)</td><td>(8)</td></tr></thead><tbody><tr><td>Foreign background</td><td char=".">−0.243***</td><td char=".">−0.246***</td><td char=".">−0.235***</td><td char=".">−0.230***</td><td char=".">−0.239***</td><td char=".">−0.227***</td><td char=".">−0.211***</td><td char=".">−0.169***</td></tr><tr><td /><td char=".">(0.062)</td><td char=".">(0.063)</td><td char=".">(0.061)</td><td char=".">(0.051)</td><td char=".">(0.050)</td><td char=".">(0.050)</td><td char=".">(0.054)</td><td char=".">(0.053)</td></tr><tr><td>Male</td><td char="." /><td char=".">−0.048</td><td char=".">−0.045</td><td char=".">−0.015</td><td char=".">−0.016</td><td char=".">−0.017</td><td char=".">−0.030</td><td char=".">−0.046</td></tr><tr><td /><td char="." /><td char=".">(0.053)</td><td char=".">(0.053)</td><td char=".">(0.051)</td><td char=".">(0.049)</td><td char=".">(0.049)</td><td char=".">(0.048)</td><td char=".">(0.048)</td></tr><tr><td>Student birth year</td><td char="." /><td char="." /><td char=".">0.114</td><td char=".">0.133*</td><td char=".">0.133*</td><td char=".">0.148**</td><td char=".">0.136*</td><td char=".">0.119**</td></tr><tr><td /><td char="." /><td char="." /><td char=".">(0.077)</td><td char=".">(0.076)</td><td char=".">(0.077)</td><td char=".">(0.073)</td><td char=".">(0.073)</td><td char=".">(0.060)</td></tr><tr><td>Regrader fixed effect</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Rewriter fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Programme fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Topic fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td></tr><tr><td>School fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td></tr><tr><td><italic>R</italic><sup>2</sup></td><td char=".">0.005</td><td char=".">0.006</td><td char=".">0.008</td><td char=".">0.099</td><td char=".">0.113</td><td char=".">0.114</td><td char=".">0.133</td><td char=".">0.247</td></tr></tbody></table> </ephtml> </p> <p>Notes: Dependent variables are standardised scores based on a 0–3 scale. A constant is always included. Two-way clustered standard errors are reported in parentheses at the school and regrader level (Cameron, Gelbach, and Miller [<reflink idref="bib9" id="ref64">9</reflink>]; Thompson [<reflink idref="bib35" id="ref65">35</reflink>]). The 'foreign background' variable measures the effect of discrimination using 0–3 scale. <emph>n</emph> = 1622 in all specifications.</p> <p>*Significance at the 10% level.</p> <p>**Significance at the 5% level.</p> <p>***Significance at the 1% level.</p> <p>Table A3. OLS regression results on the difference between the non-blind and blind test scores.</p> <p> <ephtml> <table><thead valign="bottom"><tr><td>Variables</td><td>(1)</td><td>(2)</td><td>(3)</td><td>(4)</td><td>(5)</td><td>(6)</td><td>(7)</td><td>(8)</td></tr></thead><tbody><tr><td>Non-European background(<italic>β</italic>1)</td><td char=".">−0.297***</td><td char=".">−0.301***</td><td char=".">−0.282***</td><td char=".">−0.302***</td><td char=".">−0.319***</td><td char=".">−0.306***</td><td char=".">−0.283***</td><td char=".">−0.231**</td></tr><tr><td /><td char=".">(0.090)</td><td char=".">(0.091)</td><td char=".">(0.090)</td><td char=".">(0.081)</td><td char=".">(0.078)</td><td char=".">(0.080)</td><td char=".">(0.079)</td><td char=".">(0.095)</td></tr><tr><td>European Background(<italic>β</italic>1)</td><td char=".">−0.169</td><td char=".">−0.170</td><td char=".">−0.170*</td><td char=".">−0.132</td><td char=".">−0.132</td><td char=".">−0.122</td><td char=".">−0.115</td><td char=".">−0.093</td></tr><tr><td /><td char=".">(0.105)</td><td char=".">(0.106)</td><td char=".">(0.103)</td><td char=".">(0.106)</td><td char=".">(0.107)</td><td char=".">(0.106)</td><td char=".">(0.109)</td><td char=".">(0.088)</td></tr><tr><td>Male</td><td char="." /><td char=".">−0.048</td><td char=".">−0.045</td><td char=".">−0.016</td><td char=".">−0.018</td><td char=".">−0.019</td><td char=".">−0.032</td><td char=".">−0.048</td></tr><tr><td /><td char="." /><td char=".">(0.053)</td><td char=".">(0.053)</td><td char=".">(0.050)</td><td char=".">(0.049)</td><td char=".">(0.049)</td><td char=".">(0.048)</td><td char=".">(0.048)</td></tr><tr><td>Student birth year</td><td char="." /><td char="." /><td char=".">0.111</td><td char=".">0.129*</td><td char=".">0.129*</td><td char=".">0.144**</td><td char=".">0.132*</td><td char=".">0.115*</td></tr><tr><td /><td char="." /><td char="." /><td char=".">(0.077)</td><td char=".">(0.076)</td><td char=".">(0.076)</td><td char=".">(0.073)</td><td char=".">(0.073)</td><td char=".">(0.060)</td></tr><tr><td>Regrader fixed effect</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Rewriter fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Programme fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td><td>Yes</td></tr><tr><td>Topic fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td><td>Yes</td></tr><tr><td>School fixed effect</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>No</td><td>Yes</td></tr><tr><td>Test of if <italic>β</italic>1 = <italic>β</italic>2 (<italic>p</italic>-value)</td><td char=".">0.394</td><td char=".">0.383</td><td char=".">0.4588</td><td char=".">0.274</td><td char=".">0.227</td><td char=".">0.236</td><td char=".">0.271</td><td char=".">0.360</td></tr><tr><td><italic>R</italic><sup>2</sup></td><td char=".">0.006</td><td char=".">0.006</td><td char=".">0.009</td><td char=".">0.100</td><td char=".">0.114</td><td char=".">0.115</td><td char=".">0.134</td><td char=".">0.247</td></tr></tbody></table> </ephtml> </p> <p>Notes: Dependent variables are standardised scores based on a 0–3 scale. A constant is always included. Two-way clustered standard errors are reported in parentheses at the school and regrader level (Cameron, Gelbach, and Miller [<reflink idref="bib9" id="ref66">9</reflink>]; Thompson [<reflink idref="bib35" id="ref67">35</reflink>]). <emph>n</emph> = 1622 in all specifications. The 'foreign background' variable divided into non-European and European using 0–3 scale.</p> <p>*Significance at the 10% level.</p> <p>**Significance at the 5% level.</p> <p>***Significance at the 1% level.</p> <ref id="AN0110071015-13"> <title> Notes </title> <blist> <bibl id="bib1" idref="ref2" type="bt">1</bibl> <bibtext> For example, Statistics Sweden ([30]) reports that in about 60% of the Swedish localities, almost all children have a Swedish background. In the larger cities, areas where a majority of children have foreign backgrounds are common. In areas that are dominated by foreign children, the children are generally from Africa, Asia, or a non-EU-25 country.</bibtext> </blist> <blist> <bibl id="bib2" idref="ref1" type="bt">2</bibl> <bibtext> See Ouazad and Page ([25]) for some new experimental evidence.</bibtext> </blist> <blist> <bibl id="bib3" idref="ref5" type="bt">3</bibl> <bibtext> As in previous studies, we measure the relative difference in outcomes between two groups. We cannot separate between one group being positively discriminated (favouritism) from that the other being negatively discriminated, as discussed by Feld, Salamanca, and Hamermesh ([16]).</bibtext> </blist> <blist> <bibl id="bib4" idref="ref26" type="bt">4</bibl> <bibtext> For evidence, see Ferguson ([17]).</bibtext> </blist> <blist> <bibl id="bib5" idref="ref27" type="bt">5</bibl> <bibtext> For a detailed description, see Hinnerich, Höglin, and Johannesson ([20]).</bibtext> </blist> <blist> <bibl id="bib6" idref="ref29" type="bt">6</bibl> <bibtext> Recently, the Swedish Schools Inspectorate has begun to partly evaluate schools.</bibtext> </blist> <blist> <bibl id="bib7" idref="ref15" type="bt">7</bibl> <bibtext> See the regulatory curriculum in Skolverket ([27]).</bibtext> </blist> <blist> <bibl id="bib8" idref="ref41" type="bt">8</bibl> <bibtext> In the appendix, we also provide results from an ordered logit, not assuming cardinality, and the results when the grades have equal space using a scale of 0, 1, 2, and 3, not 0, 10, 15, and 20. The results are robust to these alternative specifications.</bibtext> </blist> <blist> <bibl id="bib9" idref="ref42" type="bt">9</bibl> <bibtext> In Table 1, we use the scale 0, 10, 15, and 20 for Fail, Pass, Pass with Distinction, and Excellent.</bibtext> </blist> <blist> <bibtext> The correlation between the non-blind and the blind grade is 0.41 (for both the Pearson and the Spearman correlation). The scores are identical for about 46% of the students.</bibtext> </blist> <blist> <bibtext> The smallest cluster is of reasonable size (42 external graders). Thus, small sample problems should be of little importance. Interestingly, the standard errors are somewhat larger when we cluster only on the school level, However, with respect to critical levels of significance (1% and 5%), there is no difference in Tables 2 and 4 when we use either schools as level of clustering or both schools and external graders. The results from clustering only on the school level can be received upon e-mailing the corresponding author.</bibtext> </blist> <blist> <bibtext> Since there was one test date in the spring and one in the fall, there were 18 topics in the data set. Moreover, there was one category in which the topic was unclear.</bibtext> </blist> <blist> <bibtext> A standard test of the same proportions shows that of 19 different topics, the proportions are significantly different from zero (5% level) in four topics between the groups, indicating some sorting depending on the student's background.</bibtext> </blist> <blist> <bibtext> Marginal effects from estimating Equation (1) as an ordered logit are presented in Table A1.</bibtext> </blist> <blist> <bibtext> A student with a foreign background could have two different classifications, being both non-European and European, if, for example, the mother was born outside Europe but the father was born in Europe (but not Sweden). We defined the groups by first using information about whether the student with a foreign background was born in Sweden, in Europe, or outside Europe. If she was not born in Sweden, she was classified according to her country of birth. If she was born in Sweden, we based our presented results on the country of birth of the mother. However, the results are similar when we use the country of birth of the father, as this distinction only changed the classification of two students from the European group to the non-European group.</bibtext> </blist> </ref> <ref id="AN0110071015-14"> <title> References </title> <blist> <bibtext> Ahmed, Ali M., and Mats Hammarstedt. 2008. "Discrimination in the Housing Market: A Field Experiment on the Internet." Journal of Urban Economics 64 (2): 362–372. doi: 10.1016/j.jue.2008.02.004</bibtext> </blist> <blist> <bibtext> Anderson, Lisa, Roland Fryer, and Charles Holt. 2006. "Discrimination: Experimental Evidence from Psychology and Economics." In Handbook on the Economics of Discrimination, edited by William M. Rodgers III, 97–118. Northampton, MA: Edward Elgar.</bibtext> </blist> <blist> <bibtext> Arai, Mahmood, and Peter Skogman Thoursie. 2009. "Renouncing Personal Names: An Empirical Examination of Surname Change and Earnings." Journal of Labor Economics 27 (1): 127–147. doi: 10.1086/593964</bibtext> </blist> <blist> <bibtext> Arrow, Kenneth J. 1972a. "Models of Job Discrimination." In Racial Discrimination in Economic Life, edited by Anthony H. Pascal, 83–102. Lexington, MA: D.C. Heath.</bibtext> </blist> <blist> <bibtext> Arrow, Kenneth J. 1972b. "Some Mathematical Models of Race Discrimination in the Labor Market." In Racial Discrimination in Economic Life, edited by Anthony H. Pascal, 187–204. Lexington, MA: D.C. Heath.</bibtext> </blist> <blist> <bibtext> Becker, Gary. 1957. The Economics of Discrimination. Chicago: University of Chicago Press.</bibtext> </blist> <blist> <bibtext> Blank, Rebecca M. 1991. "The Effects of Double-Blind Versus Single-Blind Reviewing: Experimental Evidence from the American Economic Review." American Economic Review 81 (5): 1041–1067.</bibtext> </blist> <blist> <bibtext> Böhlmark, Anders, and Helena Holmlund. 2011. 20 år med förändringar i skolan. Vad har hänt med likvärdigheten. Stockholm: SNS.</bibtext> </blist> <blist> <bibtext> Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2006. Robust Inference with Multi-way Clustering. National Bureau of Economic Research Working Paper T0327.</bibtext> </blist> <blist> <bibtext> Carlsson, Magnus, and Dan-Olof Rooth. 2007. "Evidence of Ethnic Discrimination in the Swedish Labor Market Using Experimental Data." Labour Economics 14 (4): 716–729. doi: 10.1016/j.labeco.2007.05.001</bibtext> </blist> <blist> <bibtext> Dee, Thomas S. 2004. "Teachers, Race, and Student Achievement in a Randomized Experiment." Review of Economics and Statistics 86 (1): 195–210. doi: 10.1162/003465304323023750</bibtext> </blist> <blist> <bibtext> Dee, Thomas S. 2005. "A Teacher Like Me: Does Race, Ethnicity, or Gender Matter?" American Economic Review 95 (2): 158–165. doi: 10.1257/000282805774670446</bibtext> </blist> <blist> <bibtext> Dee, Thomas S. 2007. "Teachers and the Gender Gaps in Student Achievement." Journal of Human Resources 42 (3): 528–554.</bibtext> </blist> <blist> <bibtext> van Ewijk, Reyn. 2011. "Same Work, Lower Grade? Student Ethnicity and Teachers' Subjective Assessments." Economics of Education Review 30 (5): 1045–1058. doi: 10.1016/j.econedurev.2011.05.008</bibtext> </blist> <blist> <bibtext> Farkas, George. 2003. "Racial Disparities and Discrimination in Education: What Do We Know, How Do We Know It, and What Do We Need to Know?" Teachers College Record 105 (6): 1119–1146. doi: 10.1111/1467-9620.00279</bibtext> </blist> <blist> <bibtext> Feld, Jan, Nicolás Salamanca, and Daniel S. Hamermesh. 2013. Endophilia or Exophobia: Beyond Discrimination. IZA Discussion Paper No. 7380.</bibtext> </blist> <blist> <bibtext> Ferguson, Ronald F. 2003. "Teachers' Perceptions and Expectations and the Black-White Test Score Gap." Urban Education 38 (4): 460–507. doi: 10.1177/0042085903038004006</bibtext> </blist> <blist> <bibtext> Goldin, Claudia, and Cecilia Rouse. 2000. "Orchestrating Impartiality: The Impact of Blind Auditions on Female Musicians." American Economic Review 90 (4): 715–741. doi: 10.1257/aer.90.4.715</bibtext> </blist> <blist> <bibtext> Hanna, Rema, and Leigh Linden. 2009. Measuring Discrimination in Education. National Bureau of Economic Research Working Paper 15057.</bibtext> </blist> <blist> <bibtext> Hinnerich, Björn T., Erik Höglin, and Magnus Johannesson. 2011. "Are Boys Discriminated in Swedish High Schools?" Economics of Education Review 30 (4): 682–690. doi: 10.1016/j.econedurev.2011.02.007</bibtext> </blist> <blist> <bibtext> Jacob, Brian A., and Steven D. Levitt. 2003. "Rotten Apples: An Investigation of the Prevalence and Predictors of Teacher Cheating." Quarterly Journal of Economics 118 (3): 843–877. doi: 10.1162/00335530360698441</bibtext> </blist> <blist> <bibtext> Kiss, David. 2011. "Are Immigrants and Girls Graded Worse? Results of a Matching Approach." Education Economics 21 (5): 447–463. doi: 10.1080/09645292.2011.585019</bibtext> </blist> <blist> <bibtext> Lavy, Victor. 2008. "Do Gender Stereotypes Reduce Girls' Human Capital Outcomes? Evidence from a Natural Experiment." Journal of Public Economics 92 (10–11): 2083–2105. doi: 10.1016/j.jpubeco.2008.02.009</bibtext> </blist> <blist> <bibtext> Mullis, Ina V.S., Michael O. Martin, David F. Robitaille, and Pierre Foy. 2009. TIMSS Advanced 2008 International Report: Findings from IEA's Study of Achievement in Advanced Mathematics and Physics in the Final Year of Secondary School. Chestnut Hills, MA: TIMSS & PIRLS International Study Center, Boston College.</bibtext> </blist> <blist> <bibtext> Ouazad, Amine, and Lionel Page. 2011. Estimating Perceptions of Discrimination: Experimental Economics in Schools. INSEAD Working Paper No. 2011/34/EPS.</bibtext> </blist> <blist> <bibtext> Phelps, Edmund S. 1972. "The Statistical Theory of Racism and Sexism." American Economic Review 62 (4): 659–661.</bibtext> </blist> <blist> <bibtext> Skolverket. 2011. Curriculum for the Upper Secondary School. Stockholm: Fritzes.</bibtext> </blist> <blist> <bibtext> Sprietsma, Maresa. 2013. "Discrimination in Grading: Experimental Evidence from Primary School Teachers." Empirical Economics 45 (1): 523–538. doi: 10.1007/s00181-012-0609-x</bibtext> </blist> <blist> <bibtext> Statistics Sweden. 2002. Statistics on Persons with Foreign Background: Guidelines and Recommendations. Reports on Statistical Co-ordination for the Official Statistics of Sweden 2002:3. Örebro: Statistics Sweden.</bibtext> </blist> <blist> <bibtext> Statistics Sweden. 2007. Barn, boendesegregation och skolresultat. Demografiska rapporter 2007:2. Örebro: Statistics Sweden.</bibtext> </blist> <blist> <bibtext> Swedish National Agency for Education. 2006. Gymnasieskolans Kursprov vt 2006: En Resultatredovisning. Stockholm: Swedish National Agency for Education.</bibtext> </blist> <blist> <bibtext> Swedish Schools Inspectorate. 2011. Lika eller olika? Omrättning av nationella prov i grundskolan och gymnasieskolan. Stockholm: Swedish Schools Inspectorate.</bibtext> </blist> <blist> <bibtext> Swedish Schools Inspectorate. 2012. Lika för alla? Omrättning av nationella prov i grundskolan och gymnasieskolan under tre år. Stockholm: Swedish Schools Inspectorate.</bibtext> </blist> <blist> <bibtext> Szulkin, Ryszard, and Jan O. Jonsson. 2006. Ethnic Segregation and Educational Outcomes in Swedish Comprehensive Schools: A Multilevel Analysis. Stockholm: Mimeo, Stockholm University.</bibtext> </blist> <blist> <bibtext> Thompson, Samuel B. 2010. "Simple Formulas for Standard Errors that Cluster by Both Firm and Time." Journal of Financial Economics 99 (1): 1–10. doi: 10.1016/j.jfineco.2010.08.016</bibtext> </blist> <blist> <bibtext> Wikström, Christina, and Magnus Wikström. 2005. "Grade Inflation and School Competition: An Empirical Analysis Based on the Swedish Upper Secondary Schools." Economics of Education Review 24 (3): 309–322. doi: 10.1016/j.econedurev.2004.04.010</bibtext> </blist> </ref> <aug> <p>By Bjorn Tyrefors Hinnerich; Erik Höglin and Magnus Johannesson</p> <p>Reported by Author; Author; Author</p> </aug> <nolink nlid="nl1" bibid="bib34" firstref="ref3"></nolink> <nolink nlid="nl2" bibid="bib10" firstref="ref6"></nolink> <nolink nlid="nl3" bibid="bib15" firstref="ref7"></nolink> <nolink nlid="nl4" bibid="bib11" firstref="ref8"></nolink> <nolink nlid="nl5" bibid="bib12" firstref="ref9"></nolink> <nolink nlid="nl6" bibid="bib13" firstref="ref10"></nolink> <nolink nlid="nl7" bibid="bib18" firstref="ref16"></nolink> <nolink nlid="nl8" bibid="bib23" firstref="ref17"></nolink> <nolink nlid="nl9" bibid="bib20" firstref="ref19"></nolink> <nolink nlid="nl10" bibid="bib29" firstref="ref20"></nolink> <nolink nlid="nl11" bibid="bib19" firstref="ref21"></nolink> <nolink nlid="nl12" bibid="bib28" firstref="ref22"></nolink> <nolink nlid="nl13" bibid="bib14" firstref="ref23"></nolink> <nolink nlid="nl14" bibid="bib22" firstref="ref24"></nolink> <nolink nlid="nl15" bibid="bib16" firstref="ref25"></nolink> <nolink nlid="nl16" bibid="bib26" firstref="ref28"></nolink> <nolink nlid="nl17" bibid="bib21" firstref="ref32"></nolink> <nolink nlid="nl18" bibid="bib36" firstref="ref33"></nolink> <nolink nlid="nl19" bibid="bib31" firstref="ref34"></nolink> <nolink nlid="nl20" bibid="bib24" firstref="ref44"></nolink> <nolink nlid="nl21" bibid="bib32" firstref="ref45"></nolink> <nolink nlid="nl22" bibid="bib33" firstref="ref46"></nolink> <nolink nlid="nl23" bibid="bib35" firstref="ref48"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1076235
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Discrimination against Students with Foreign Backgrounds: Evidence from Grading in Swedish Public High Schools
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Hinnerich%2C+Bjorn+Tyrefors%22">Hinnerich, Bjorn Tyrefors</searchLink><br /><searchLink fieldCode="AR" term="%22Höglin%2C+Erik%22">Höglin, Erik</searchLink><br /><searchLink fieldCode="AR" term="%22Johannesson%2C+Magnus%22">Johannesson, Magnus</searchLink>
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22Education+Economics%22"><i>Education Economics</i></searchLink>. 2015 23(6):660-676.
– Name: Avail
  Label: Availability
  Group: Avail
  Data: Routledge. Available from: Taylor & Francis, Ltd. 325 Chestnut Street Suite 800, Philadelphia, PA 19106. Tel: 800-354-1420; Fax: 215-625-2940; Web site: http://www.tandf.co.uk/journals
– Name: PeerReviewed
  Label: Peer Reviewed
  Group: SrcInfo
  Data: Y
– Name: Pages
  Label: Page Count
  Group: Src
  Data: 17
– Name: DatePubCY
  Label: Publication Date
  Group: Date
  Data: 2015
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Journal Articles<br />Reports - Research
– Name: Audience
  Label: Education Level
  Group: Audnce
  Data: <searchLink fieldCode="EL" term="%22High+Schools%22">High Schools</searchLink><br /><searchLink fieldCode="EL" term="%22Secondary+Education%22">Secondary Education</searchLink>
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Foreign+Countries%22">Foreign Countries</searchLink><br /><searchLink fieldCode="DE" term="%22Public+Schools%22">Public Schools</searchLink><br /><searchLink fieldCode="DE" term="%22High+Schools%22">High Schools</searchLink><br /><searchLink fieldCode="DE" term="%22Educational+Discrimination%22">Educational Discrimination</searchLink><br /><searchLink fieldCode="DE" term="%22High+School+Students%22">High School Students</searchLink><br /><searchLink fieldCode="DE" term="%22Grading%22">Grading</searchLink><br /><searchLink fieldCode="DE" term="%22Minority+Group+Students%22">Minority Group Students</searchLink><br /><searchLink fieldCode="DE" term="%22Scores%22">Scores</searchLink><br /><searchLink fieldCode="DE" term="%22National+Competency+Tests%22">National Competency Tests</searchLink><br /><searchLink fieldCode="DE" term="%22Statistical+Analysis%22">Statistical Analysis</searchLink><br /><searchLink fieldCode="DE" term="%22Regression+%28Statistics%29%22">Regression (Statistics)</searchLink><br /><searchLink fieldCode="DE" term="%22Robustness+%28Statistics%29%22">Robustness (Statistics)</searchLink><br /><searchLink fieldCode="DE" term="%22Least+Squares+Statistics%22">Least Squares Statistics</searchLink>
– Name: Subject
  Label: Geographic Terms
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Sweden%22">Sweden</searchLink>
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1080/09645292.2014.899562
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 0964-5292
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: We rigorously test for discrimination against students with foreign backgrounds in high school grading in Sweden. We analyse a random sample of national tests in the Swedish language graded both non-blindly by the student's own teacher and blindly without any identifying information. The increase in the test score due to non-blind grading is significantly higher for students with a Swedish background. This discrimination effect is sizeable, about 10% of the mean or 20% of the standard deviation of the blind test score.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: Ref
  Label: Number of References
  Group: RefInfo
  Data: 36
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2015
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1076235
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1076235
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1080/09645292.2014.899562
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 17
        StartPage: 660
    Subjects:
      – SubjectFull: Foreign Countries
        Type: general
      – SubjectFull: Public Schools
        Type: general
      – SubjectFull: High Schools
        Type: general
      – SubjectFull: Educational Discrimination
        Type: general
      – SubjectFull: High School Students
        Type: general
      – SubjectFull: Grading
        Type: general
      – SubjectFull: Minority Group Students
        Type: general
      – SubjectFull: Scores
        Type: general
      – SubjectFull: National Competency Tests
        Type: general
      – SubjectFull: Statistical Analysis
        Type: general
      – SubjectFull: Regression (Statistics)
        Type: general
      – SubjectFull: Robustness (Statistics)
        Type: general
      – SubjectFull: Least Squares Statistics
        Type: general
      – SubjectFull: Sweden
        Type: general
    Titles:
      – TitleFull: Discrimination against Students with Foreign Backgrounds: Evidence from Grading in Swedish Public High Schools
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Hinnerich, Bjorn Tyrefors
      – PersonEntity:
          Name:
            NameFull: Höglin, Erik
      – PersonEntity:
          Name:
            NameFull: Johannesson, Magnus
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 01
              Type: published
              Y: 2015
          Identifiers:
            – Type: issn-print
              Value: 0964-5292
          Numbering:
            – Type: volume
              Value: 23
            – Type: issue
              Value: 6
          Titles:
            – TitleFull: Education Economics
              Type: main
ResultId 1