آیا کلیشه های جنسیتی نتایج سرمایه انسانی دختران و پسران را کاهش می دهد؟ مدارک و شواهد از یک آزمایش طبیعی

Schools and teachers are often said to be a source of stereotypes that harm girls. This paper tests for the existence of gender stereotyping and discrimination by public high-school teachers in Israel. It uses a natural experiment based on blind and non-blind scores that students receive on matriculation exams in their senior year. Using data on test results in several subjects in the humanities and sciences, I found, contrary to expectations, that male students face discrimination in each subject. These biases widen the female–male achievement difference because girls outperform boys in all subjects, except English, and at all levels of the curriculum. The bias is evident in all segments of the ability and performance distribution and is robust to various individual controls. Several explanations based on differential behavior between boys and girls are not supported empirically. However, the size of the difference is very sensitive to teachers' characteristics, suggesting that the bias against male students is the result of teachers', and not students', behavior.

Schools and teachers are often said to be a source of stereotypes that harm girls. Bernard (1979), Dusek and Joseph (1983), Madon et al. (1998), and Tiedemann (2000) are only a few of the many scholars who have claimed that teachers occasionally rely on stereotypes in forming perceptions about their students.1 Examples of such stereotypical perceptions are that boys excel in math and science and girls excel in other subjects, or that boys have more talent and that girls compensate by working hard (Deaux and LaFrance, 1998). Girls are then encouraged, on the basis of these stereotypes, to pursue traditionally female studies instead of mathematics, science, and other traditionally male subject areas, from as early as first grade (Carr et al., 1999) and women are steered toward certain occupations, as evidenced by studies of college students (Glick et al., 1995), PhD holding research students, (Rowsey, 1997) and others (Deaux and LaFrance, 1998). Another claim about stereotypes is that beliefs are manifested through teachers' evaluation of students. This claim is supported by evidence from a survey of 1st grade teachers (Fennema et al., 1990), the AAUW (1992) report, which surveyed girls from kindergarten to 12th grade, a survey of mothers and their 11–12 year old children (Jacobs and Eccles, 1992), and others (Ben-Zvi Mayer et al., Ben-Zvi Mayer et al., 1995 and Hildebrand, 1996).2 The bottom line of the literature on gender stereotypes is that they are partly responsible for one of the alleged forms of discrimination against women and that they may have far reaching implications and consequences regarding gender differences in human capital investment and outcomes. However, there is very little convincing evidence to date that substantiates these claims and this study attempts, using a unique empirical context, to fill some of this deficiency. This paper tests for the existence of gender stereotyping and discrimination by public high-school teachers in Israel. It uses a natural experiment based on blind and non-blind scores that students receive on matriculation exams in their junior and senior years. The natural experiment arises from rules that are used in Israel to determine scores in matriculation subjects (these rules are explained in detail in Section 3.1). This testing protocol elicits two scores, a blind and a non-blind score, both of which are meant to measure the student's knowledge of the same material. Due to this testing method, we may safely assume that the blind score is free of any bias that might be caused by stereotyped discrimination on the part of the external examiner. The non-blind score, however, may reflect biases occasioned by teachers' gender stereotypes. As long as the two scores are comparable, i.e., as long as they measure the same skills and cognitive achievements (grounds for this assumption are discussed further in Section 3.1), the blind score may be used as the counterfactual measure to the non-blind score, which may be affected (“treated”) by stereotyping and discrimination. This identification framework is similar to that used by Goldin and Rouse (2000) and by Blank (1991). I use the difference between the boys' and girls' gaps between the blind and the non-blind test scores as a measure of a potential gender bias. Using data on all matriculation scores of several cohorts of high-school seniors in Israel, I applied this natural-experiment framework to test for the existence of such a gender bias in nine subjects—four in the humanities (English, history, literature and biblical studies), mathematics, and four in science (biology, chemistry, computer science, and physics). The distributions of the blind and non-blind scores in many of these subjects are very similar and, in many cases, are identical. The basic results of these quasi-experiments show that, contrary to expectations, the potential bias is against male students. The sign of this difference is the same in all nine subjects examined and in all tests in cases where there is more than one exam per subject. The extent of the potential bias varies by subject and test, ranging from 0.05 to 0.25 of the standard deviation of the blind-score distribution. This gap against male students, on average, doubles the gender-score difference because female students outperform male students on state external exams in all subjects except English. The results are not sensitive to various student-level controls because the identification strategy is based on differences-in-differences at the student level, for which reason individual fixed effects are assumed away. In some subjects the difference is largest for low achievers and in some subjects it is actually largest for the most proficient male students. The basic results withstand several specification checks. Overall, they do not support the hypotheses that the gender difference in the non-blind score reflects statistical discrimination against male students. For example, limiting the sample to schools where boys outperform girls on average, overall or in specific subjects, leaves the results basically unchanged. The variance in performance of boys is higher on average and in every subject than that of girls, suggesting that statistical discrimination against boys may also occur due to “noisier” signals in boys' test scores. The data, however, do not support this interpretation because the gender potential bias is not different in schools where girls demonstrate more variability in performance on average. I also examined the possibility that the results mainly reflect the effect of the specific pattern in the timing of the exams where the state follows the school exam. Using data from a second chance state-level exam in English that was taken 4 months after the school-level exam led to almost identical estimates, suggesting that the short time interval between the state and school exams cannot explain our basic results. An interesting and obvious question in this context is whether this estimated gender difference represents students' behavior or teachers' behavior. An example of students' behavior that may explain this potential bias is differential pattern of mean reversion by gender, e.g., due to a time-varying gender difference in knowledge or due to girls not performing as well in the state exams because they may represent a more ‘pressured environment' for girls. The evidence suggests, if anything, stronger mean reversion for girls than for boys, namely girls tend to improve their scores more if they perform below average at the school exam (either relative to the class mean, the class mean by gender or the prior-self in earlier exams). Another possibility is that the state and school exams do not share the same content and/or do not measure the same skills. For example, some systematic gender differences in within-class behavior (discipline, attitude, absenteeism) may end up impacting grading at the school level and not at the state level. Unfortunately, the data do not include information that would help in directly addressing this source of concern but various pieces of indirect evidence do not support this hypothesis either. For example, the basic evidence is relatively robust to controlling for lagged school scores for other exams in the same or other subjects. The paper also examines explanations based on empirical insights gained from experiments in social psychology. One such important insight is that a “stereotype threat”–the threat of being perceived as a negative stereotype or the fear of a poor performance that would confirm the stereotype–may be powerful enough to shape the intellectual performance and academic identities of entire groups of people. The implication in our context is that the difference in favor of girls of the gap between the non-blind and blind-test scores reflects the inferior performance of girls in non-blind tests because it involves a stereotype threat and superior performance in blind tests that conceal their gender. The evidence provided in this paper does not allow us to state that this mechanism explains the negative male difference. Instead, the data support the hypothesis that teachers' behavior is responsible for this difference. I show that the gender potential bias in various subjects is very sensitive to teachers' characteristics such as gender, age, years of experience, and even family size. There is no reason to expect this pattern of sensitivity of the difference to teachers' characteristics to reflect students' behavior. Based on the evidence presented in the paper, I conclude that the estimated gender differences can be interpreted as an anti-male bias. The paper is organized as follows: The next section discusses the literature of stereotypes and discrimination, and in particular evidence about teachers' stereotypical behavior culled from experiments in social psychology. Section 3 explains the design of the study, the Israeli high-school matriculation exam system, and the comparability of school- and state-exam scores, and presents a graphical exposition of the blind–non-blind test score difference by gender. The basic results of the experiment are presented in Section 4. Sections 5 and 6 discuss alternative interpretations and explanations and Section 7 concludes.

Recent convincing evidence suggests that women face discrimination in the labor market in terms of both employment opportunities and wages (e.g., Goldin and Rouse, 2000). However, the question of whether discrimination against women is also partly responsible for the observed gender differences in human-capital investment and occupational choice has not been directly addressed.16 This paper confronted one important alleged form of discrimination against women that may cause some of the skewed gender pattern in productivity-enhancing investments: discrimination against girls in school by their teachers. This discrimination, it is argued, emanates from a stereotyping of cognitive ability that causes female students to under-perform in, and to shy away from, math and science subjects in secondary and post-secondary schooling—a state of affairs that also affects occupational choice, of course. Women continue to avoid college majors and occupations that entail moderate amounts of coursework in mathematics, even though they match or surpass men's performance in high-school math and science courses (AAUW, 1999).17 Gender-related differences in career choices also persist, especially in the fields of physical science, mathematics, and engineering, where women hold only about 10% of jobs.18 In 1995, women constituted about 46% of the U.S. labor force but only 22% of scientists and engineers in the labor force.19 In the U.K., only 6% of engineers in the labor force are women.20 In Israel in 2005 only 20% of civil engineers and architects are women despite making up 52% of the labor force.21 The evidence presented in this study confirms that the previous belief that schoolteachers have a grading bias against female students may indeed be incorrect. On the contrary: on the basis of a natural experiment that compared two evaluations of student performance–a blind score and a non-blind score–the difference estimated strongly suggests a bias against boys. The direction of the bias was replicated in all nine subjects of study, in humanities and science subjects alike, at various level of curriculum of study, among underperforming and best-performing students, in schools where girls outperform boys on average, and in schools where boys outperform girls on average. Lindahl (2006) adopts a similar methodology and finds supporting evidence from national tests in Sweden that boys seem to be discriminated against in final tests, compared to girls. The bias against boys persisted when other interactions with the non-blind score were allowed, e.g., interactions with parental schooling and other student characteristics. When students' ability was added as a control or as an interaction with the bias against male students, the average difference did not disappear and was revealed at all levels of cognitive ability. The anti-male bias among teachers widened the difference between male and female students because the latter, on average, outperformed the former in almost all subjects. The size of the bias varied from − 0.053 to − 0.180 in the first estimates (Table 2), to − 0.062 to − 0.272 when controlling for the identical distribution of exams, with the approximate average value hovering at around − 0.10 in most of the different specifications. It is difficult to quantify exactly the effect of this gap on matriculation rates, but given that boys already have lower average marks than girls, this gap will lower their marks further, and could decrease their chances of acceptance into various programs after leaving school. The anti-male bias may reflect differential behavior between male and female students, e.g., gender differences in the intrinsic relative ability level when performing in the school and state exams or gender differences in the ability to improve upon a bad signal. It may also reflect statistical discrimination against male students that emanates from the superior performance of female students in almost every subject or from the higher variance of performance of boys in every subject. The data do not support these two potential sources of the bias. On the other hand, variation of the estimated bias against male students in accordance with teachers' characteristics such as gender, age, and teaching experience strengthens the interpretation of the male bias as a form of discrimination resulting from teachers' behavior. The magnitude of the grading bias against male students is relatively large and may harm students firstly by causing their final scores on some matriculation exams to fall below the passing mark and, thereby, by disqualifying them for a matriculation certificate. This may bar students from post-secondary schooling, at least temporarily, implying lower schooling attainment. Secondly, since admission to most university departments in Israel is based solely on the average score on the matriculation exams, the bias against boys lowers their average matriculation score and, by so doing, may reduce boys' prospects of admission to their preferred fields of study. Both effects have negative future labor-market consequences. Finally, I should caution that the findings reported above that teachers are systematically giving lower grades to boys for the same quality of academic performance is not enough evidence to conclude that teaches are not engaging in other behavior that might reduce girls' desire to further human capital. In fact, it can be argued that the evidence of teachers male bias in grading tests could be consistent with teachers being tougher on boys because they have higher expectations for them. The tougher grades could also be viewed as a way to give male students stronger incentives to study harder in the future. Teachers may not use such an incentives device for girls if they do not have as high expectations for them. Such an alternative interpretation of the findings of this paper cannot be disproved since the evidence does not allow distinguishing between the hypotheses that teachers are lenient or merciful toward girls versus being tough or over demanding on boys.