Attanasio Selection-Bias Decomposition Framework

In Simple Words

Think of testing a new coaching program on marathon runners. If the coaching motivates slower runners to finish who would otherwise drop out, the average finishing time of all completers might look worse even if the coaching actually made every individual runner faster.

Research question

Measuring treatment effects on output quality strictly among program completers introduces selection bias because peer interaction changes the baseline propensity to finish or submit work.

Participants

N = 3,333 entrepreneurs in the Large-Country Sample (N = 1,322 submitted proposals evaluated by 15 professional judges).

Study Design

Non-parametric selection-bias decomposition assuming No Defiers (ND) and Better Always Submitters (BAS), separating participants into always-submitters, never-submitters, compliers, and defiers.

Results

Virtual-Within interaction yields a statistically significant intensive quality increase among motivated Milestone 0 completers ($\beta = 0.138, p < 0.05$ in Large-Country Sample; $\beta = 0.427, p < 0.05$ in Uganda Sample). Under the BAS assumption, this observed conditional effect serves as a strict lower bound for the true unconditional quality impact.

Takeaway

Decompose conditional quality metrics using the Better Always Submitters framework when program treatments induce differential submission or completion rates between treated and control groups.

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.

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

• Exogenous Failure Assignment →Methods & evidence

  Imagine testing how drivers handle an unexpected patch of black ice on a road. If drivers only go onto the icy road voluntarily, only risk-tolerant drivers are tested; but if a road detour forces a random selection of drivers onto the ice patch, researchers can measure true handling ability and recovery without selection bias.

• 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.