Effect of Beliefs and Gender Roles on Girls' Math Education

We conduct an RCT in Ghana to investigate how providing information about potential abilities and returns to math education in terms of labor market opportunities and family outcomes affect investments in math education in terms of time, schooling expenditures, girls' aspirations, secondary school enrollment rate and the field of study.

(See Experimental Design)

2018-06-11

2019-01-01

Beliefs about Child's Performance:
Beliefs on relative performance of girls and boys by subjects*
Gap in actual - expected performance by gender and subject (math/reading)
Choice of prize for follow-up test (low, medium or high stakes)

Beliefs/Aspirations for education attainment:
Do you hope that your child (you) will enter senior secondary school?
If yes: What do you hope your child (you) will study?
Beliefs about whether son/daughter will enter senior secondary school
Beliefs about the field of study son/daughter will choose

Beliefs about labor market opportunities:
Beliefs on returns to secondary school by gender
Beliefs on returns to secondary school field by gender
Beliefs on likelihood of having a good career by gender and field

Beliefs about gender roles:
Beliefs on importance of different subjects for girls and boys*
Beliefs on whether it is a priority for a woman to be a good wife and mother

Beliefs about marital returns to education:
Beliefs on whether son/daughter will get married on the basis of secondary school field
Beliefs on age at which son/daughter will get married on the basis of secondary school field
Beliefs on fertility of son/daughter on the basis of secondary school field
Beliefs on earnings of son/daughter's spouse on the basis of secondary school field

Investment in math education:
Choice of prize for follow-up test (math textbooks, English textbooks or phone minutes)
Choice of weekly planner for child
Performance on math assessment
Completed exercises in distributed weekly planners
Children's perception of parental encouragement
Expected and realized purchase of textbooks by subject and gender
Expected and realized time spent on studying different subjects
Expected and realized time discussing education and future careers with parents

Education outcomes:
Enrollment in senior secondary school
Applied to senior secondary school
Applied to math-related program for senior secondary school
Wanted to apply to math-related program for senior secondary school
School attendance

Other outcomes:
Did parents inform child of their performance (only above median)

To address our research question, we will conduct an RCT in Ghana in June 2018. Our intervention is developed in the context of the Wave 3 of the Ghana Socioeconomic Panel Survey, a nationally representative household survey administered by University of Ghana with the next planned survey in 2022 (Wave 4). Our study will focus on the sample of about 2,000 households with at least one boy or one girl between ages 9 and 18 in 2018 who has not entered senior secondary school. In total, there will be about 3,000 children in our sample.
Because our survey questions will be integrated into the panel survey, we will collect endline data during Wave 4. Prior to our endline, we will return to collect follow-up data in June 2019.

We will randomly assign households to three information treatments. Enumerators will give each household a calendar with the relevant treatment information. In the first treatment, we provide parents and children with information about the grade-specific distribution by gender on the math and English reading assessment conducted during the survey. We will provide parents with the actual performance of their children if they scored above the median. Using previous waves of the survey, we find no significant differences in the math and English reading performance across gender. However, if parents are more likely to underestimate the performance of girls relative to boys, then there may be an under-investment in these subjects for girls.

In the second treatment, we present the same information as in the first treatment as well as potential labor market opportunities of math-related education for girls. In the third treatment, we show the information as in the first treatment and information on the benefits of math education for outcomes related to family well-being and parenting. As before, we will provide parents with the actual performance of those who score above the median. By also providing the information on overall performance by gender and subject area, the last two treatments will capture the additional effects of information on career or family well-being outcomes on educational decisions.

Moreover, in a cross-cutting design, we will randomly provide the performance information to either the mother, the father or both parents. This will allow us to evaluate whether the identity of the information recipient affects the final outcomes.
Additionally, we will also able to measure if there are any spillover effects across family members; for example, in the group where only the mother was informed, we can see if the father also updated his beliefs relatively more or less than when both parents were informed. We will also measure spillover effects onto younger siblings.

Note: Placebo calendars are provided to the control group.

Randomization done in office by a computer.

We will conduct our experiment with a double randomization. We start by randomly assigning the 2,200 households with at least one child into either the control group or one of three possible treatment arms (with equal probability).
In all treatment groups, all the eligible children in the same family will be treated in the same way.
In a second cross-randomization, we will randomize with 1/3 probability over which parent will receive the information on performance: mother only, father only, or both parents.
To achieve balance, we stratify using parents' education and region.

Households

Yes

2200 Households

3000 Children Age 9-18 not in Senior Secondary School

550 households control, 550 households information about performance, 550 households information about performance and labor market, 550 households information about performance and family

Cross randomization into three groups:
733 Mothers and eligible children receive information (if control, they are shown placebo calendar without information)
733 Fathers and eligible children receive information (if control, they are shown placebo calendar without information)
734 Both parents and eligible children receive information (if control, they are shown placebo calendar without information)

We are powered to detect an effect size of at least 0.2 standard deviations with power of 0.8 and significance level of 0.05. For our power calculation, we considered the sample of boys and girls separately. We assumed there is a correlation between baseline and follow-up data of 0.15, which corresponds to the actual correlation between Wave 1 and Wave 2 standardized math test scores. This correlation is measured with a four-year gap, which is a conservative approach, because our follow-up test score data will be collected one year after the collection of the Wave 3 data. To be conservative, we also assume clusters of two because there are about 1.26 girls per household. The intraclass correlation is assumed to be 0.38, which is the intrahousehold correlation of math test scores in Wave 2. Under these conservative assumptions, we require a sample size of 268 households per treatment arm, which is about 1/4 of the total households in our sample with at least one daughter. For household level analysis (e.g., outcomes for parents that are not gender-specific), we can detect an effect of 0.13 s.d. under the same assumptions.