Data & Methods
What are same-sex
role model effects?
We follow the existing literature and define the same-sex role model effect as the premium of having a same-sex teacher —on top of the general effect of having a female or male teacher (Eble and Hu, 2020; Hoffmann and Oreopoulos, 2009; Lim and Meer, 2017; Muralidharan and Sheth, 2016). A positive role model effect could be driven by female students benefitting more from female teachers than male students, male students benefitting more from male teachers than female students, or both. This effect is distinct from sex differentials in teacher effectiveness. For example, there would be no role model effect if girls and boys benefit equally from having a female teacher. However, there would be a positive role model effects if girls benefit more than boys from having a female teacher.
While we follow the literature and call this effect a role model effect, note that this effect could be driven by the behavior of teachers, students, or both. For example, we could observe role model effects because teachers use teaching styles that students of their own sex can more easily relate to. However, we could also observe role model effects because students behave differently with teachers of their own sex.
What data do we use?
To estimate same-sex role model effects, we build a large-scale multi-country dataset. We combine data from TIMSS & PIRLS for all available countries, waves and education levels. TIMSS and PIRLS are administered by the International Association for the Evaluation of Educational Achievement, which specializes in administering education assessments that allow for international comparisons. TIMSS measures the skills and knowledge in mathematics and sciences of 4th graders (9- to 10-year-old children) and 8th graders (13- to 14-year old children). PIRLS measures the reading skills of 4th graders.
For both studies, we use all waves as of December 2021, which is when we finished our data collection. These are seven waves of TIMSS (1995, 1999, 2003, 2007, 2011, 2015, and 2019) covering 86 different countries and four waves of PIRLS (2001, 2006, 2011, and 2016) covering 64 different countries.
How do we estimate
role model effects?
To measure the effect of same-sex role models on test scores, we estimate the following regression model:
Outcomeisj = β1Female Studenti + β2Female Teacherj + β3Female Studenti x Female Teacherj + γ 'Xisj + uisj
where Outcomeisj is one of the four relevant outcomes (job preferences, test scores, subject enjoyment, and subject confidence) for individual i in subject s that is taught by teacher j. Female Studenti is a dummy variable indicating the sex of the student, Female Teacherj is the share of female teachers in subject s (which is equivalent to a dummy variable when students only have one teacher in subject ), and Female Studenti x Female Teacherj is an interaction term of these two variables. Xisj is a vector of control variables that differ by specification and is the error term. The role model effect is captured by β3, which shows the additional premium or penalty from having a same-sex teacher, on top of the general effect of having a female teacher. We estimate Equation (1) via ordinary least squares regressions (OLS) and cluster our standard errors at the classroom level following the criteria outlined in Abadie et al. (2022).
Results shown on this website are based on our preferred specification. In our preferred specification, we include student fixed effects and teacher fixed effects. In this specification, we use within-student across-subject variation to hold constant all student characteristics that are the same across subjects. For example, we exploit that the same student may have a female science teacher and a male math teacher (or vice versa). By also including teacher fixed effects we address one main concern: that more-effective teachers could be assigned to a higher share of students of their own sex.