Conservative Attrition Bounding and Imputation

In Simple Words

Imagine testing a health program over four years where some participants move away and stop answering calls—to prove the program works even under the worst-case scenario, researchers assume all missing control participants suddenly got high-paying jobs while missing treated participants got low-paying jobs, verifying if the positive effect still holds true.

Research question

Longitudinal panel studies over multi-year horizons face attrition risks that could bias long-term earning estimates if unobserved dropouts systematically differ across treatment and control arms.

Participants

**N = 3,052 panel participants** tracked across a 4-year follow-up period in Addis Ababa, Ethiopia.

Study Design

Sensitivity analysis imposing differential attrition bounds (Lee bounds and Karlan-Valdivia/Blattman imputation up to $\pm 0.5$ standard deviations of control outcomes).

Results

Long-run wage gains remained statistically significant and economically meaningful (>10% of control mean) across all differential attrition scenarios, retaining a positive point estimate even under the most conservative assumption of imputing a full 0.5 SD penalty to missing treated units.

Takeaway

Apply conservative differential mean and standard deviation imputation bounds to longitudinal trial data to confirm treatment effect robustness against attrition.

Read alongside the research

Plain-English glossary

Original explanations, informed by the references below. Examples are illustrations, not data from the studies above. The research text stays unchanged.

Longitudinal study

A study that checks the same group of people at different times. This lets researchers examine how things change over time.

ExampleResearchers ask the same workers about their wellbeing at the start, after three months and after six months.

NICE · Longitudinal study ↗Read more about longitudinal study →

Follow-up

Collecting more information about participants later in a study to see what has happened to them. The term can also mean the length of time researchers keep track of them.

ExampleResearchers interview the same workers three months after a programme ends. That later interview is a follow-up.

GET-IT glossary · Follow-up ↗Read more about follow-up →

Statistically significant

A label used when a result passes a chosen statistical threshold, often p < 0.05. It does not show how large or useful an effect is. A result that is not statistically significant does not prove there is no effect.

ExampleA study reports p = 0.03 against a 0.05 threshold. It passes that test, but the effect could still be too small to matter.

American Statistical Association · Statement on p-values ↗Read more about statistically significant →

Related findings

• Active Tracking Attrition Mitigation →Methods & evidence

  Dispatching field teams directly to children's homes when they miss school testing operates like a home delivery service tracking down uncollected mail to ensure every package is accounted for.

• Commercial Ad Platform Algorithmic Randomization Imbalance →Methods & evidence

  Allowing a commercial ad algorithm to handle randomization is like letting a profit-driven tournament host pick matching opponents. The platform optimizes for click rates rather than making sure both starting teams are equally balanced.

• Multi-Source Administrative-Survey Data Triangulation →Methods & evidence

  Think of monitoring a patient's health by combining heart rate monitors and blood tests with direct interviews about how they feel. Combining automated medical readings with verbal feedback provides a complete, accurate picture of health rather than relying on either measurement alone.

• Adversarial Non-Stationarity Bounds in Multi-Armed Bandits →Methods & evidence

  Imagine playing a card game where an opponent can secretly alter the deck rules between rounds. If you assume the deck never changes, early luck might trick you into a permanently failing strategy.