Does Classroom Gender Composition Affect School Dropout? Bulent Anil Tuba Toru Delibaşı Gokce Uysal 2/19/20161Labor Market Network Meeting
Introduction Does the gender composition of peers affect academic achievement? – Achievement: Drop-out decision of students. Transition from 8th to 9th grade. – Gender composition of peers: the ratio of female students to class size. What do we find? It is nice to have girls in your class! 2/19/2016Labor Market Network Meeting2
What we do We can disentangle academic peer effects and gender peer effects. – Academic peer effects: if your classmates are doing well, you do better. – Gender peer effects: the gender composition has an effect on your academic success. We study the transition from compulsory to non-compulsory education. 2/19/2016Labor Market Network Meeting3
What we find Even though girls are more likely to drop out, having more girls in the section decreases the probability that students drop out. Possible channels: – Academic peer effects (not in our case) – Competition driven by traditional gender roles (boys should do better than girls) – Potential detrimental effects of single gender – Socioeconomic background effects 2/19/2016Labor Market Network Meeting4
Literature Papers: Confounded academic peer effects and gender peer effects – Girls are doing better academically. If you have more girls in the classroom, you do better (Hoxby, 2000;Levy and Schlosser, 2011; Gotfried et al., 2013; Hu, 2015) Single-sex education is better (Whitmore, 2005) Higher share of the opposite gender among school friends reduces the academic achievement (Hill, 2015) Nothing on gender-peer effects in Turkey 2/19/2016Labor Market Network Meeting5
Literature Dropout: – Student’s preference, i.e., lower expectation from graduation (Eckstein and Wolpin, 1999) – Myopic behavior (Oreopoulos, 2007) – Family structure (Astone and McLanahan, 1994) – Family income (Belley, Frenette and Lochner, 2008) – Parents’ education and their valuation of education (Folley, Gallipoli and Green, 2014) Tansel (2002) – Determinants of the gender gap in enrollment after mandotary school – Income, parental education, location and migration Kirdar (2009) – Ethnic disparities on enrollment – Gender gap 2/19/2016Labor Market Network Meeting6
Institutional framework Transition from 8 th to 9 th is voluntary Students assigned to the closest school (except for private schools, 2-3%) Random assignment across sections Students take all of their classes in the same section, with the same classmates. 2/19/2016Labor Market Network Meeting7
Data e-Okul / e-School: universal More than 1 million students, many have missing data: 672 thousand students About 18 thousand schools Data on students’ academic background, family background and schools 2/19/2016Labor Market Network Meeting8
Estimation strategy Reduced form Baseline model: logit (also tried probit) Different specifications: – Males vs. females – Public vs. private – Non-linear effects 2/19/2016Labor Market Network Meeting9
Model specification Dependent variable: School Dropout Independent variables: – School & Class Share of females in class, average SBS scores (class & school), school type, school resources (students per teacher, number of teacher (FT), class size, labs, etc.) – Individual Age, gender, SBS score, academic performance indicators (math, science, Turkish and GPA), – Family Parent’s age, employment, education, income, number of siblings, CCT – Geographic Province, urban/rural 2/19/2016Labor Market Network Meeting10
Descriptive statistics 2/19/2016Labor Market Network Meeting11 VariableObsMean VariableObsMean School Dropout Middle Income Share of Females in class Upper middle income Class Average of SBS Score High income School Average of SBS score Conditional Cash Transfer Female GPA (SD) Age Maths (SD) Number of siblings Science (SD) Mother's Age Turkish (SD) Father's Age Student SBS Grade Mother Primary Private Mother Middle School Dorm Mother High School Minority Mother Higher Educ Public Father Primary Student per Teacher Father Middle School Number of Teacher (FT) Father High School Science Lab Father Higher Educ Multipurpose Lab Father works Class size Mother works Village Low income County Lower middle income Urban
Estimation Results I: Benchmark Model 2/19/2016Labor Market Network Meeting12 VARIABLESBenchmarkAverage Marginal Effect Share of Females in class-0.420*** Class Average of SBS Score *** School Average of SBS score *** Female1.020*** Number of siblings0.135*** Mother's Age0.008*** Father's Age *** Father works ** Mother works ** Conditional Cash Transfer GPA (SD)-0.298*** Maths (SD)-0.205*** Science (SD)-0.271*** Turkish (SD)-0.379*** Student SBS Grade-2.308*** Private Boarding school0.371*** Minority Student per Teacher *** Number of Teacher (Permanent contract)-0.368*** Average class size *** Village0.777*** County0.232*** n *** p<0.01, ** p<0.05, * p<0.1
Benchmark results Gender peer effects – 1 percent increase in the share of females in a student’s class decreases their probability of dropping out by 3.75 pp. – Even though girls are 9.1 pp more likely to drop out. Academic peer effects Usual suspects: – Students less likely to drop out if parents work – CCT – Academic background is controlled for – School resources 2/19/2016Labor Market Network Meeting13
Estimation Results II: Females&Males 2/19/2016Labor Market Network Meeting14 VARIABLESInteractionFemaleMaleFemale NonlinearMale Nonlinear Share of Female in class-0.397***-0.305***-0.864***-0.603*-1.954*** Share of Female*Female Square Share of Female in class *** Female1.043*** Class Average of SBS Score *** *** *** *** *** School Average of SBS score *** *** *** *** *** Number of siblings0.135***0.161***0.115***0.161***0.115*** Mother's Age *** *** *** *** *** Father's Age *** *** *** *** *** Father works ** * * Mother works ** ***0.0245* *** Conditional Cash Transfer GPA (SD)-0.298***-0.191***-0.390***-0.191***-0.390*** Maths (SD)-0.205***-0.314***-0.109***-0.314***-0.110*** Science (SD)-0.271***-0.293***-0.281***-0.293***-0.282*** Turkish (SD)-0.379***-0.344***-0.409***-0.344***-0.409*** Student SBS Grade-2.308***-2.648***-2.115***-2.648***-2.117*** Private Boarding school0.373***0.339***0.411***0.333***0.379*** Minority Student per Teacher *** *** *** Number of Teacher (FT)-0.368***-0.591***-0.157***-0.590***-0.158*** Student per Class *** *** *** *** *** Village0.777***0.889***0.707***0.889***0.702*** County0.232***0.358***0.124***0.358***0.124*** n *** p<0.01, ** p<0.05, * p<0.1
Results: Females & Males Do the effects differ by student’s gender? Interaction term suggests otherwise. Try for females and males: they seem to be different. The coefficients on gender peer effects are different, i.e. higher for males. Effects seem to be linear for females and non-linear for males. 2/19/2016Labor Market Network Meeting15
Estimation Results III: Other Specifications 2/19/2016Labor Market Network Meeting16 VARIABLESRuralUrbanPrivate Share of Female in class-0.540***-0.425***0.436 Class Average of SBS Score *** *** * School Average of SBS score *** *** * Female1.309***0.923***0.143 Number of siblings0.118***0.140***0.125 Mother's Age *** Father's Age *** *** *** Father works ***0.452 Mother works *** Conditional Cash Transfer GPA (SD)-0.232***-0.311*** Maths (SD)-0.282***-0.191***-0.292* Science (SD)-0.252***-0.286***-0.267* Turkish (SD)-0.307***-0.403***0.246* Student SBS Grade-2.299***-2.324***-3.011*** Private-0.208** Boarding school0.147***0.473*** Minority * Student per Teacher *** *** Number of Teacher (FT)-0.636***-0.161***0.432*** Student per Class *** *** n *** p<0.01, ** p<0.05, * p<0.1
Other specifications Gender peer effects somewhat stronger in rural areas. We cannot explain much in private schools. 2/19/2016Labor Market Network Meeting17
Conclusion Turkey vs. developed world: – Females are more likely to drop out – Still, it helps to have girls in your class. – We can disentangle gender peer effects and academic peer effects. Policy implication: single-gender schools may not be as beneficial. The question remains: How does this work? What are the channels? 2/19/2016Labor Market Network Meeting18