Gender gap in math score: does teacher gender matter?

• Název práce v češtině: Rozdíl mezi pohlavími v matematickém skóre: záleží na pohlaví učitele?
• Název v anglickém jazyce: Gender gap in math score: does teacher gender matter?
• Klíčová slova: Genderová nerovnost v matematických výsledcích, Propensity score matching, TIMSS standardizované testy, Efekt pohlaví učitele, Mezinárodní analýza
• Klíčová slova anglicky: Gender gap in math achievement, Propensity score matching, TIMSS standardized tests, Effect of teachers’ gender, International analysis
• Akademický rok vypsání: 2020/2021
• Typ práce: diplomová práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: Mgr. Barbara Pertold-Gebicka, M.A., Ph.D.
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 15.06.2021
• Datum zadání: 15.06.2021
• Datum a čas obhajoby: 01.02.2023 09:00
• Místo konání obhajoby: Opletalova - Opletalova 26, O206, Opletalova - místn. č. 206
• Datum odevzdání elektronické podoby: 02.01.2023
• Datum proběhlé obhajoby: 01.02.2023
• Oponenti: doc. PhDr. Martin Gregor, Ph.D.

Seznam odborné literatury

ANGHEL, Brindusa, Núria RODRÍGUEZ-PLANAS and Anna SANZ-DE-GALDEANO. Is the math gender gap associated with gender equality? Only in low-income countries. Economics of Education Review. 2020, 79(C).
CARELL, Scott E., Marianne E. PAGE and James E. WEST. Sex and Science: How Professor Gender Perpetuates the Gender Gap. The Quarterly Journal of Economics. 2010 125(3), p. 1101-1144.
CORDERO, José M., Víctor CRISTÓBAL and Daniel SANTÍN. Causal Inference on Education Policies: A Survey of Empirical Studies Using PISA, TIMSS AND PIRLS. Journal of Economic Surveys. 2017, 32(3), p. 878-915.
CHO, Insook. The effect of teacher-student gender matching: Evidence from OECD countries, Economics of Education Review, 2012, 31(3), p. 54-67.
DEE, Thomas S. Teachers and the Gender Gaps in Student Achievement. The Journal of Human Resources, 2007, 42(3), p. 528–554.
DOSSI, Gaia, David FIGLIO, Paola GIULIANO and Paola SAPIENZA. Born in the Family: Preferences for Boys and the Gender Gap in Math. CEPR Discussion Papers 13504. 2019.
GUISO, Luigi, Ferdinando MONTE, Paola SAPIENZA and Luigi ZINGALES. Culture, Gender, and Math. Science. 2008, 320(5880), p. 1164-1165.
HERMANN, Zoltán and Alfa DIALO. Does teacher gender matter in Europe? Evidence from TIMSS data. Budapest Working Papers on the Labour Market 1702, Institute of Economics, Centre for Economic and Regional Studies. 2017.
HOGREBE, Nina a Rolf STRIETHOLT. Does non-participation in preschool affect children’s reading achievement? International evidence from propensity score analyses. Large-Scale Assessments in Education. 2016, 4(1), p. 1-22.
KIM, Doo Hwan a Helen LAW. Gender gap in maths test scores in South Korea and Hong Kong: Role of family background and single-sex schooling. International Journal of Educational Development. 2012, 32 (1), p. 92-103.
KRIEG, John M. Student gender and teacher gender: What is the impact on high stakes test scores?, Current Issues in Education. 2005, 8(9).
WINTERS, Marcus A., Robert HAIGHT, Thomas SWAIM and Katarzyna A. PICKERING. The effect of same-gender teacher assignment on student achievement in the elementary and secondary grades: Evidence from panel data,." Economics of Education Review. 2013 34(C), p. 69-75.

Předběžná náplň práce v anglickém jazyce

Motivation:
The difference between girls and boys in standardized test scores – commonly known as the gender gap has sparked the interest of researchers as hypothetically there should be none. In math and science, this gap historically favoured boys, while in reading or writing girls were more successful. As the prevailing ambition among policymakers is to provide equal educational opportunities, there is an acute need to explain these systematic differences. I am planning to focus mainly on the gap in Mathematics.
Various explanations of this phenomenon have appeared in the literature. For example, Guiso et al. (2008) found a link between test score differences and indicators of gender equality. On the contrary, later revisitation by Anghel et al. (2020) shows that this link vanishes once the country fixed effects are accounted for, yet the link still holds for poor countries. Apart from the societal inequality, some authors tried to explain the gap by cultural family background. For instance, Dossi et al. (2020) used fertility stopping rules to show that girls in families with boy preference score lower than girls in other families, as well as finding that maternal gender role attitudes have a similar impact. In summary, Dossi et al. (2020) claim that family background may explain part of the observed gap. Conversely, Kim and Law (2011) found little support for family background effect and also showed the non-trivial impact of single-sex schooling. I shall deviate from these explanations and instead, I shall add to another vast branch of literature that tried to evaluate the effect of teacher’s gender on student’s performance.
So far, the literature offers several mixed results. Krieg (2005) followed 3rd graders in the state of Washington for two years but found no significant impact of a same-sex teacher on student performance. Dee (2007) first exploited the matching pairs strategy to control for student fixed effects in longitudinal data (National Education Longitudinal Study of 1988) and found that assignment to the same-sex teacher significantly improves students results regardless of student gender. Carrell et al. (2010) examined the topic in college settings, the results suggest that although there is little effect of teacher gender on male students, female students are significantly affected. Winters et al. (2013) found no significant impact of a same-gender teacher on student performance in Florida panel. As for some cross-country comparisons, Cho (2012) utilized the data from the Trends in International Mathematics and Science Study (TIMSS) to evaluate the effect of the same-gender teacher in 15 OECD countries and found large heterogeneity and overall little support for the significance of this effect. The identification strategy is similar to the one used by Dee (2007) and accounts for student fixed effects. Diallo and Hermann (2017) also use TIMSS data on 20 European countries to evaluate the differences between Western and Eastern Europe. Their results suggest that same-gender teacher benefits mostly students in Western Europe. As mentioned above the results are still rather inconclusive and point to large cross-country heterogeneity.
The papers mentioned in the previous paragraph mainly use the first difference identifying strategy. An advantage of this method is that it identifies the causal effect, however, due to its’ nature, it is not applicable to be used for 4th graders in the TIMSS environment. Since results of TIMSS 2019 seem to point out to expansion of the gender gap for 4th grade I believe it is worth examining the teacher-student same gender effect on younger pupils. Specifically, if the student fixed effects are correlated with the assignment to a same-sex teacher (treatment) then the coefficient of this variable would be biased. Therefore, I plan to utilize (a different identification strategy that should reduce the selection bias) Propensity Score Matching to examine whether the gender of the teacher matters for the gender math gap among 4th grade students.

Hypotheses:
As mentioned above, the main goal of this thesis is to assess the effect of being taught by the same-gender teacher for girls and boys among 4th grade students using TIMSS data. To get an unbiased estimator using OLS, the selection to treatment should be random – uncorrelated with student characteristics. However, there exists a concern that this is not the case, as in Cho (2012): “For example, if students in lower academic tracks are more likely to be assigned to female teachers, this nonrandom assignment creates a negative correlation between teacher gender and unobservable student ability and causes a bias in the coefficient for the gender matching variable.” Hence, I plan to test the randomness of selection to treatment using observable data available at the time of assignment to treatment.
1. Hypothesis #1: Having a male teacher improves the math score for boys.
2. Hypothesis #2: Having a female teacher improves the math score for girls.
3. Hypothesis #3: Selection to treatment (having a male teacher) is non-random.

Methodology:
The intended identification strategy is Propensity Score Matching. This method aims to replicate the randomized experiment by balancing covariates. In the first stage, the probability of assignment to treatment, i.e. to having a male teacher, will be calculated using data from the Early Learning Survey. This survey is part of TIMSS and is intended to be completed by students’ parents. The questions are specifically aimed at children’s abilities and characteristics before they start school. For their timing, I believe that these data are exactly the ones that would determine the probability of being selected for treatment. I shall estimate this probability (or propensity score) using logistic regression. The first stage results should also provide me with a test of hypothesis 3. I expect the selection to be non-random, but in the unlikely case that all explanatory variables in the logistic model are jointly insignificant (selection to treatment is random) I could move to a simple OLS approach to test hypotheses 1 and 2.
After obtaining the propensity score from the fitted model, I move to the second stage. There are several approaches on how to proceed in the second stage. The most straightforward is a one-to-one matching of observations from the treatment and control group with the most similar propensity score using the nearest neighbour technique. Another option is to use stratification – dividing the propensity score distribution into several groups and weighting. Lastly, I may use a weighting algorithm (Hogrebe and Strietholt, 2016) – for example, kernel matching. I plan to perform leave-one-out validation to choose the best matching algorithm. Finally, to obtain standard errors bootstrap techniques will be used.
Preliminary inspection of TIMSS 2019 data showed large heterogeneity among countries when it comes to the proportion of students in 4th grade taught by a male teacher. While in Latvia it is less than 0.5% of students, in Saudi Arabia it is over 49%. Such heterogeneity suggests that the effect may not always be easily generalizable to a country’s population and so the results will probably be of higher value to countries where a larger proportion of the population is treated. Therefore, I may focus on countries where a greater proportion of students are being taught by a male teacher. Nonetheless, I intend to provide a cross-country comparison.

Expected Contribution:
The literature on teacher’s gender effect so far examined mainly teenage students and used the first difference estimating strategy to identify the effects. Such an approach does not allow researchers to estimate the effects for 4th graders, because 4th grade students are usually taught by one teacher in all subjects. The novelty of my work is a new identification strategy that should allow me to estimate the effect of being taught by a same-gender teacher for the TIMSS data for 4th grade. So far, this effect has not been evaluated for these students using TIMSS. Also, the latest data from TIMSS 2019 will be used. The results should add to the literature on the effect of teacher’s gender on student performance as well as to more general literature trying to explain gender gaps in standardized test scores. Finally, the results may provide useful implications for actual policy.

Outline:
The thesis will follow this structure.
1. Introduction – the motivation behind studying this topic will be disclosed as well as its’ relevance, moreover, possible contributions will be highlighted.
2. Literature review – both the literature on the gender gap and the effect of the teachers’ gender will be reviewed.
3. Data – description of the data used for testing each of the hypothesis.
4. First stage model and results – in this section, the approach to determine the probability of being treated will be described and the results will be presented.
5. Second stage model and results – based on the results from the previous section the model will be chosen and estimated, then the best model will be evaluated.
6. Discussion – the results, their meaning and implications for practice, and relevance of the employed method will be discussed.
7. Conclusion – the main findings of the thesis will be highlighted alongside key policy implications and possible avenues for future research.