Initial Findings from Equity in Refugee Education: an RCT in Kenyan - PowerPoint PPT Presentation

Initial Findings from Equity in Refugee Education: an RCT in Kenyan refugee camps Education Evidence for Action December 5, 2017 Nisha Rai Timothy Kinoti Economic Researcher M&E Manager American Institutes for Research World University Service of Canada

Motivation • Despite increased global attention to refugee population, education access remains out of reach for many • Only 50% of all refugee children attend primary school (as compared to more than 90% of all children globally) (UNHCR, 2016) • Fraction of refugee children out of school is similar in Kakuma and Dadaab, refugee camps in northern Kenya – Children number 98,861 in Kakuma (representing 60.3% of camp) – Children number 163,442 in Dadaab (representing 59.3% of camp) • Remedial education programs provide additional tutoring in the form of tailored classes to small groups of children, designed to meet students’ needs, and often target lower achieving or at-risk students (Banerjee, Cole, Duflo, & Linden, 2007) 2

Equity in Education in Refugee Camps in Kenya (EERCK) • Funded by the U.S. Bureau of Population, Refugees, and Migration • Implemented by WUSC and its local partner, Windle International Kenya (WIK) • Provides remedial education to medium- and high-performing at- risk girls in Grades 7–8. 3

This Study Two types of research questions 1. Impact evaluation questions to increase understanding about the effects 2. Process evaluation questions to increase understanding of delivery and scaling model Longitudinal Study Description of HEA program The Humanitarian Education Accelerator, which was set up by the Department for International Development (DFID), the United Nations’ Children’s Fund, and the United Nations High Commissioner for Refugees (UNHCR), aims to generate rigorous evidence to understand how to transform high-potential pilot projects into scalable education initiatives for refugees and displaced communities worldwide. 4

Selection Criteria • Medium- and High-Performing Girls. – Average mark of the three terms of the previous year. – In Kakuma, girls scored higher on average, criteria of 200+ – In Dadaab, girls scored lower on average, criteria 170+ • Vulnerable Girls (with one or more of the following) – Teen mothers – Girls from child-headed households – Girls from single-parent families – Girls from foster families – Girls identified by teachers or other caregivers as at-risk of getting married – Girls with physical disabilities. • More girls met selection criteria in Kakuma than Dadaab, so we implemented two different designs 5

Baseline approach to the design Kakuma Kakuma RCT: Treatment and Control Group Baseline • 568 Treatment Characteristics Treatment Group Control Group Average Fraction Values 1 • 555 Control 0.8 0.6 • Stat. Sig Diff 5% level • Age (T older) 0.4 • Walk to school (T 0.2 more likely to walk) 0 • Stat. Sig Diff 10% level • Access water on plot (T more likely to Treatment group: 567 students Control group: 555 students access) Demographic Characteristics • # People in household (T has The above graph shows the balance between treatment and control groups in the Kakuma RCT on different baseline characteristics. As you can there are minimal higher #) significant differences in these characteristics between our treatment and control groups. 6

Baseline approach to the design Dadaab • 478 Treatment • 182 Comparison • Stat. Sig Diff 5% level • Age (T older) Eligibility Cut • Access water on plot Off (170) (T less likely to access) • T has higher fraction Control Control Treatment Treatment Group Group Group Group from DRC, lower fraction from Somalia • T higher fractions participate in WUSC, school club, mentorship • Stat. Sig Diff 10% level The above graph illustrates the regression discontinuity design in Dadaab. Those • Live with parent or who scored above 170 were eligible and invited to the EERCK remedial classes. The guardian (T group control group is comprised of students who scored between 151-169 just missing the eligibility cut off. less likely to) 7

Empirical Strategy • Intent-to-Treat measured by Difference in Differences Post Baseline 1st difference (Feb 2017) (Oct 2017) Treatment (T) Y T Y T ΔY T =(Y T Oct_2017 -Y T Feb_2017 ) Feb_2017 Oct_2017 Control / Y C Y C ΔY C =(Y C Oct_2017 -Y C Feb_2017 ) Feb_2017 Oct_2017 Comparison (C) Difference-in-difference = (ΔY T – ΔY C ) • Dependent Variables related to Aspirations, Resilience, and (eventually) Achievement • Baseline Control Variables of Age, Household size, Walk to school, Participates in non-WUSC program, score (for Kakuma) 8

Attrition Period Assignment Dadaab Kakuma Treatment group at Baseline 478 568 Baseline Control group at Baseline 182 555 Backup group at Baseline 0 109 Treatment group at Endline 373 444 Endline Control group at Endline 113 439 Backup at Endline 0 12 Difference in Treatment Group 105 124 Baseline Difference in Control Group 69 116 vs. Difference in Backup N/A 97 Endline Overall Difference (without backup) 174 240 Overall Difference (with backup) 174 337 9

Attrition Analysis • Differential Attrition: Test difference in baseline characteristics between the treatment and control households that remain in the study at follow up – No evidence of differential attrition in Kakuma or Dadaab – Statistically significant differences correspond to differences at baseline • Overall Attrition: Test difference in baseline characteristics between the remaining sample at follow-up and the sample at baseline – No evidence of overall attrition in Kakuma or Dadaab 10

Results: Kakuma (Aspirations) Index Like Look Think Balance Believe Beileve Want Doesn't School forward Learns school will will scholar want to at and finish finish -ship marriage school school other grade 2 nd -ary Time -0.107 -0.019 -0.017 -0.025 -0.007 -0.012 -0.004 0.000 -0.001 (0.060) (0.007) (0.016) (0.011)* (0.029) (0.014) (0.016) (0.000) (0.018) ** Treat -0.073 -0.015 -0.007 -0.023 -0.037 -0.002 0.003 0.000 0.015 (0.058) (0.006) (0.015) (0.011)* (0.030) (0.014) (0.015) (0.000) (0.019) * Diff- -0.025 0.009 0.006 0.017 -0.062 0.020 -0.025 0.000 -0.000 in-Diff (0.088) (0.012) (0.023) (0.017) (0.043) (0.020) (0.023) (0.000) (0.027) Cons. -0.203 0.925 0.981 0.901 1.092 0.883 0.702 1.000 0.296 (0.342) (0.064) (0.068) (0.066) (0.141) (0.077) (0.095) (0.000) (0.100)* ** ** ** ** ** ** * R 2 0.02 0.01 0.01 0.01 0.02 0.01 0.01 . 0.01 N 1,633 1,629 1,615 1,627 1,624 1,624 1,624 1,380 1,588 Notes: Heteroskedasticity robust standard errors in parenthesis. * p <0.10; ** p <0.05; *** p <0.01. Controls include age, number in household, walks to school indicator, participates in non-WUSC program, academic score. 11

Results: Kakuma (Resilience) Index Have Proud Treated Like my Getting Feel Friends Friends people of my fairly culture better belong- me support- in my eth- marks ing support ive life I can nicity im- in respect portant difficult Time -0.038 -0.042 0.019 -0.056 -0.015 0.006 0.019 -0.010 0.007 (0.065) (0.015)* (0.025) (0.033) (0.025) (0.016) (0.025) (0.027) (0.029) * Treat -0.054 -0.035 -0.003 -0.032 0.036 -0.022 -0.005 -0.010 0.020 (0.062) (0.015)* (0.026) (0.033) (0.023) (0.018) (0.025) (0.027) (0.029) Diff- -0.076 0.027 -0.055 -0.003 -0.004 -0.015 -0.064 -0.089 -0.100 in- (0.093) (0.024) (0.036) (0.047) (0.034) (0.025) (0.036) (0.040)* (0.042)* Diff Cons -0.395 0.847 0.579 0.705 0.690 0.686 0.868 0.584 0.672 (0.342) (0.092)* (0.133)* (0.158)* (0.109)* (0.106)* (0.127)* (0.135)* (0.138)* * * * * * * * * R 2 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.02 0.01 N 1,633 1,599 1,569 1,578 1,611 1,600 1,589 1,608 1,595 Notes: Heteroskedasticity robust standard errors in parenthesis. * p <0.10; ** p <0.05; *** p <0.01. Controls include age, number in household, walks to school indicator, participates in non-WUSC program, academic score. 12

Results: Kakuma (Resilience) I try to Can People Aware Know Can act Apply Have Share finish solve think I of where respons abilities family feelings activitie problem am fun strength to seek ibly support with s I start s s help in family/p commu artner nity Time -0.018 -0.034 0.017 -0.024 0.009 0.020 -0.056 -0.042 -0.010 (0.024) (0.032) (0.031) (0.030) (0.031) (0.027) (0.027)* (0.029) (0.029) Treat 0.019 -0.050 0.031 -0.024 0.004 0.036 -0.022 -0.009 -0.018 (0.024) (0.032) (0.031) (0.030) (0.031) (0.027) (0.027) (0.029) (0.029) Diff- -0.029 -0.005 -0.029 0.022 -0.015 -0.063 0.074 -0.005 -0.027 in- (0.034) (0.044) (0.044) (0.043) (0.044) (0.039) (0.038)* (0.040) (0.041) Diff Cons 0.942 0.594 0.709 0.611 0.539 0.657 0.826 1.139 0.866 (0.114)* (0.152)* (0.150)* (0.146)* (0.149)* (0.132)* (0.127)* (0.136)* (0.142)* * * * * * * * * * R 2 0.01 0.01 0.00 0.00 0.00 0.00 0.01 0.01 0.00 N 1,602 1,592 1,585 1,585 1,598 1,593 1,604 1,600 1,607 Notes: Heteroskedasticity robust standard errors in parenthesis. * p <0.10; ** p <0.05; *** p <0.01. Controls include age, number in household, walks to school indicator, participates in non-WUSC program, academic score. 13