DiD Impact Analysis | WBR Hwange Endline Evaluation

What is Difference-in-Difference?

Difference-in-Difference (DiD) compares the change in outcomes over time between a treatment group (wards that received bicycles in 2020) and a comparison group (wards that received bicycles later, in 2022). By comparing changes rather than levels, DiD controls for time-invariant differences between groups and common time trends.

-4.2 days

DiD Estimate

Causal effect on absenteeism

p < 0.01

Treatment Effect

Statistically significant

0.8 days

Pre-trend Difference

Not significant (p = 0.41)

346

Sample Size

198 early · 148 late treatment

Parallel trends

Mean days absent by treatment group - lines should track together before each group receives bicycles

DiD estimation

Comparing pre-treatment (2019) to post-treatment (2021) outcomes across groups

Group	Pre (2019)	Post (2021)	Change
Early treatment (bikes 2020)	12.8	4.8	-8.0
Late treatment (bikes 2022)	12.0	9.2	-2.8
Difference-in-Difference			-5.2

The DiD estimate of -5.2 days suggests that bicycle distribution caused a 5.2-day reduction in absenteeism beyond any secular trend.

Effect by subgroup

DiD effect size (days of absenteeism reduced) by student subgroup

Robustness checks

Tests supporting the validity of the causal estimate

Parallel trends test: Pre-treatment trends were not statistically different between groups (β = 0.8, p = 0.41)
Placebo test: Running the DiD model on a pre-treatment period (2018–2019) yields a null result (β = -0.3, p = 0.72)
Covariate robustness: The effect is robust to controlling for age, gender, distance to school, and ward-level fixed effects

Methodology note: This analysis uses a staggered DiD design exploiting the phased rollout of WBR's Hwange programme. The early treatment group (198 students in 10 wards) received bicycles in 2020; the late treatment group (148 students in 7 wards) received bicycles in 2022. Attendance data is from school administrative records. All data is anonymised. This page demonstrates quasi-experimental impact evaluation capability. See the R pipeline for full regression tables and sensitivity analyses.