Intent-to-Treat (ITT) Estimation Framework
RCTClinical Trial
In field experiments involving opt-in digital products, evaluating treatment impact solely among active feature users introduces non-random self-selection bias. An intent-to-treat framework measures the overall policy effect of providing access to the feature across all assigned individuals regardless of individual take-up rates.
Picture this
Imagine testing a free gym membership program to see if it improves overall health in a town. Instead of only measuring the health of people who actually went to the gym every day, you compare everyone who was offered the free pass against everyone who was not offered one, giving a realistic picture of what happens when you launch the program town-wide.
What the evidence says
Estimated overall ITT remittance probability increase of +0.007 (p < 0.10) across all treatment-assigned migrants, preventing self-selection bias from confounding the experimental estimates.
- Who was studied
- 4,451 analyzed migrant study participants (2,213 treatment, 2,245 control) across 137,927 panel observations in the UAE.
- How
- OLS panel regression estimating treatment assignment indicator $T_{it}$ with individual and week fixed effects, controlling for time-invariant unobservables and common shocks.
What to do
Estimate treatment impact across all assigned subjects using intent-to-treat OLS panel specifications to preserve experimental randomization validity.
From the source
"This regression estimates intent-to-treat (ITT) effects of the labeling treatment... The coefficient on treatment is the estimate of the causal impact of the treatment on remittances."
A Field Experiment among Filipino Migrant Workers in the UAE