Do teachers’ college majors affect students’ academic achievement in the sciences? A cross-subfields analysis with student-teacher fixed effects

ABSTRACT

We examine whether and how teachers’ major fields in college affect students’ achievement, exploiting within-student variation across subfields in natural science (i.e. physics, chemistry, biology, and Earth science). Using middle-school students’ data from the TIMSS and controlling for student-teacher fixed effects, we find that teachers improve students’ achievement in subfields of natural sciences correspond to their college majors. Teaching practices explain about half of the effect, mostly accounted for by teachers’ preparation for teaching science topics. The results are robust to potential endogenous matching between students and teachers.

KEYWORDS:

• Teacher
• natural science
• college major
• middle school
• TIMSS

JEL Classification:

• H75
• I21
• J24

Acknowledgement

This study is conducted as a part of the Project ‘Microeconometric Analysis of Education Policy with Large Administrative Data’ undertaken at the Research Institute of Economy, Trade and Industry (RIETI). The authors are grateful to valuable comments from the editor, two anonymous referees, Hideo Akabayashi, Masayuki Morikawa, Daiji Kawaguchi, the participants of research meeting of the project and Discussion Paper seminar participants at RIETI. The authors acknowledge financial support for this project from the Japan Society for the Promotion of Science (Grant Numbers 20H05629 for Ryuichi Tanaka). The usual disclaimer applies. Any errors are our own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Several studies reported that the effect of subject-related knowledge or degrees is greater in STEM subjects than in non-STEM subjects (e.g. Clotfelter, Ladd, and Vigdor 2010; Goldhaber and Brewer 1997; Metzler and Woessmann 2012).

2 We excluded TIMSS 2015 and 2019 surveys because the published datasets do not include enrollment information used for a robustness check in this paper.

3 TIMSS uses a matrix-sampling booklet design, in which each student is administered only a subset of the entire TIMSS item pools to cover a wide range of topics. Therefore, not all participating students took the same test.

4 PVs are multiple imputations of the unobservable latent achievement for each student (Wu 2005). For more detailed information about the methods and procedures in TIMSS, see Martin and Mullis (2012).

5 For example, Woessmann (2011) and Hanushek et al. (2015) used international survey data such as PISA and PIAAC and conducted weighted analyses with same weights for each country.

6 When we include student samples taught by more than one science teacher, the number of observations is 2,513,282 student-subject and 64 countries are included. Note that we exclude science teachers who cannot identify which subfield they teach or who have missing values for the major field of study. When the student sample taught by more than one science teacher is excluded, the number of observations is 1,812,276 and 51 countries are included. Furthermore, when we exclude the student sample with missing values in the variables used for estimation, the number of observations is 876,880. We confirm that even after sample restrictions, a wide range of countries is included in the analysis, from developed to developing countries. In addition, we checked how the share of type of communities changes through the sample selection processes, and found that they are stable through our sample selection process indicating that our samples are not concentrated in special regions such as rural areas.

7 Details of science teachers’ major fields are gathered in the questionnaire for teachers. Teachers are asked to choose ‘yes’ or ‘no’ for each major field, including physics, chemistry, biology, and Earth science.

8 While the ratio of science teachers who majored in Earth science is small and may have unique characteristics that cannot be observed, robust results are obtained when Earth science is excluded from the data set. We can provide the results upon request.

9 While the TIMSS 2007 and 2011 included test scores for the following four subfields in science: biology, physics, chemistry, and Earth science, the TIMSS 2003 included following five subfields test scores: life science, physics, chemistry, Earth science, and environmental science. We regard the scores in life science in the TIMSS 2003 as the scores in biology. All results in this paper are robust when the samples of TIMSS 2003 are excluded.

10 Regarding teachers’ preparation to teach science topics, when more than one-third of the question items in a subfield are missing or ‘Not applicable,’ we put a missing value in the indicator of teachers’ preparation. The number of items in each subfield is as follows: in the order of the 2003, 2007, and 2011 TIMSS survey, 5, 7 and 7 in biology; 5, 5, and 4 in chemistry; 5, 6, and 5 in Physics; 3, 5, and 4 in Earth science.

11 While the TIMSS 2007 and 2011 have five science subfields in this question: biology, physics, chemistry, Earth science, and others, the TIMSS 2003 has six: life science, physics, chemistry, Earth science, environmental science, and others. We regard the answer in the life science area in the TIMSS 2003 as the answer in the biology area.

12 The minimum value of the variable ‘Time allocation’ is 0 and the maximum value is 100, indicating that some of the teacher sample teach only one specific subfield in a year. In the robustness check section, we only present results of teachers who teach all subfields in science in a year and thus, show that our results are robust.

13 Teachers are asked to choose ‘Mostly taught before this year’ if a topic was in the curriculum before the eighth grade. If a topic was partially taught in this year but not yet completed, they are asked to choose ‘Mostly taught this year.’ If a topic is not in the curriculum, they are asked to choose ‘Not yet taught or just introduced.’ When more than one-third of the question items in a subfield are missing, we put a missing value in the indicator. The number of the items in each subfield as follows: In the order of the 2003, 2007, and 2011 TIMSS survey, 12, 14 and 7 in biology; 8, 8, and 4 in chemistry; 10, 10, and 5 in Physics; 11, 14, and 4 in Earth science.

14 The robustness check here uses an approach similar to the one used by Bietenbeck, Piopiunik, and Wiederhold (2018).

15 While TIMSS does not include information about the number of classes in the grade, it obtains enrollment in the grade from school questionnaires and class size from teacher questionnaires. We, therefore, define the student sample who attend schools where the enrollment in the grade does not exceed the class size as students who attend schools with only one class in the grade.

16 When we restrict the sample to students who are in schools with only one class in the grade, we find that 39% is through the mechanisms that cannot be explained. The ratio is smaller than that in Fig.1, which is the result when we do not consider the within-school sorting of students based on students’ subfields specific unobservables. This suggests that part of the 50%, that is, the mechanism through unexplained factors in Fig.1, may include the influence of student sorting within schools.

17 Teachers who have broader major fields may indicate greater competence. In fact, the average number of major fields in the natural sciences for teachers who hold a master’s degree or higher is 1.39, which is higher than the average of 1.24 for teachers who do not. The result is still robust when we estimate the model with the interaction term of ‘Major in fields taught’ and the possession of a master’s degree or higher added to Table 6, Column 2. Therefore, it is plausible to interpret the results in Table 6, Column 2 as indicating that the effects of major fields vary based on differences in the breadth of teacher expertise, rather than differences in teacher ability.

18 Instead of the variable of years of teaching, we also add the interaction terms with dummy variables for years of teaching, but no statistically significant results are obtained.

19 On the heterogeneity in students’ gender, Sancassani (2021) found that while teacher expertise has a positive effect on female students’ test scores, it is positive, but not statistically significant, on male students’ test scores. On the heterogeneity in students’ abilities, Metzler and Woessmann (2012) found that teachers’ rich subject knowledge influences the academic performance of high-achieving students. Sancassani (2021) found that teachers with expertise have positive and significant effects only on students with high socioeconomic status.

Additional information

Funding

This work was supported by Japan Society for the Promotion of Science (JSPS): [Grant Number 20H05629].