Keeping Girls on the STEM Track: an Exploration of Role Model Effect and its Mechanisms

The two randomizations originate three groups of students:
- Pure Control Group
- Only Role Model Videos
- Role Model Videos + Personal Interaction with Role Models

Control group: Students in the control group will not receive any interventions. However, like the rest of the students in the project, they will answer a baseline and an endline survey.
Only role model videos: Students in the only role model videos groups receive several videos of role models through a digital curriculum. Each student logs into a platform with their username and password, and is presented with the materials in sequence. The digital curriculum consists of 5 sessions where students are exposed to videos with role models that guide students through the process of choosing electives.
Personal Interaction with Role Models: in addition to receiving role model videlos The personal interaction component consists of four synchronous online sessions held after school. In each session, a STEM college student will guide a conversation with a group of five female students, randomly selected from the top-5 math-performing female students of the schools that received the role model curriculum.

2024-08-01

2024-11-30

- Interest in math and science themes
- Choice of electives in 11th and 12th grade
- Aspiration to study in higher education
- Interest in STEM majors
- Post- secondary education trajectories:
- Enrolls in higher education.
- Enrolls in STEM higher education.
- Enrolls in STEM higher education at a top university.

Longer-Term Outcomes
- Post-secondary education trajectories:
-- Applies to higher education
-- Registers to take the university admission exam.
-- Takes the university admission exam.
-- Characteristics of the institutions and programs in which they enroll: peers' quality, graduation rates, over-duration, accreditation, expected earnings, expected employability.
- Higher Education Aspirations
-- Desire to continue studying after high school graduation.
-- Self-efficacy beliefs
-- Characteristics of the institutions and programs in which they wish to enroll: peers' quality, graduation rates, over-duration, accreditation, expected earnings, and expected employability.

To evaluate the interventions, high schools were randomly assigned (at the school level) into two groups: a pure control group and a web-based role model curriculum treatment group. Top-math female students of the web-based treatment schools were randomly assigned to a personal role-model treatment. This originates three arms of the main experiment: a pure control group, a group of students who are exposed to role models only through a web-based curriculum, and a group of students who are exposed to role models through the web-based curriculum and through personal interaction.

To study the underlying mechanisms of the role model effect, I randomized the gender of role models in the web-based curriculums to test if there is any heterogeneity of the treatment effect due to gender match between the role model and the student.

Not available

Randomization done in the office by a computer

The role model video provision was clustered at the school level, and personal interaction with role model was randomized at the individual level.

Yes

The randomization at the school level was done within 320 schools. At the individual level, it was done within 1.055 students.

25,600 students

109 school control schools, 211 schools in the role model video treatment, 528 students in the video + personal interaction

Only video intervention (320 schools, 80 student per school) - College graduation expectation: 3.14 [mean, slider 1 (no expectation) to 4 (certain)], 0.073308494 [MDE (SD)], 0.037116405 (ICC) - Interest in STEM elective for 11th and 12th grade: 0.659342178 [mean, 1 is interested in STEM elective, 0 in other case], 0.075367442 [MDE (SD)], 0.037116405 (ICC) - Language self-efficacy: 7.369636513 [mean, slider 1 (I am the person with the lowest language outcomes compared to my peers) to 10 (I am the person with the highest language outcomes compared to my peers)], 0.070813233 [MDE (SD)], 0.033785632 (ICC) - Math self-efficacy: 6.709677419 [mean, slider 1 (I am the person with the lowest math outcomes compared to my peers) to 10 (I am the person with the highest math outcomes compared to my peers)], 0.074549484 [MDE (SD)], 0.038815871 (ICC) - Mental well being: 2.261184841 [mean, likert scale of agreement with ``I feel sad'' 1 (I strongly disagree) - 4 (I strongly agree)], 0.056169114 [MDE (SD)] 0.016562741 (ICC) - STEM interest: 0.37924425 [mean, 0 no STEM interest - 1 interest in STEM], 0.050031644 [MDE (SD)], 0.010525806 (ICC) Video + Personal interaction (213 schools, 5 students per school) - College graduation expectation: 3.483805668 [mean, slider 1 (no expectation) to 4 (certain)], 0.086 [MDE (SD)] - Interest in STEM elective for 11th and 12th grade: 0.786885246 [mean, 1 is interested in STEM elective, 0 in other case], 0.086 [MDE (SD)] - Language self-efficacy: 7.534274194 [mean, slider 1 (I am the person with the lowest language outcomes compared to my peers) to 10 (I am the person with the highest language outcomes compared to my peers)], 0.158717017 [MDE (SD)] - Math self-efficacy: 8.10483871 [mean, slider 1 (I am the person with the lowest math outcomes compared to my peers) to 10 (I am the person with the highest math outcomes compared to my peers)], 0.086 [MDE (SD)] - Mental well being: 2.342741935 [mean, likert scale of agreement with ``I feel sad'' 1 (I strongly disagree) - 4 (I strongly agree)], 0.086 [MDE (SD)] - STEM interest: 0.388663968 [mean, 0 no STEM interest - 1 interest in STEM], 0.086 [MDE (SD)]