Treatment effects beyond the mean using distributional regression: Methods and guidance

Area of Science:

• Econometrics
• Statistical Modeling

Background:

• Traditional regression models focus on mean effects, limiting analysis of treatment impacts on outcome distributions.
• Generalized Additive Models for Location, Scale and Shape (GAMLSS) offer a flexible framework to model all parameters of a distribution.

Purpose of the Study:

• Introduce GAMLSS as a powerful tool for analyzing treatment effects across the entire conditional distribution.
• Demonstrate the integration of GAMLSS with various program evaluation techniques.
• Apply GAMLSS to reanalyze data from the Mexican Progresa program to assess cash transfer impacts.

Main Methods:

• Utilized Generalized Additive Models for Location, Scale and Shape (GAMLSS) for distributional regression.
• Integrated GAMLSS with program evaluation methods: randomized controlled trials, panel data, difference-in-differences, instrumental variables, and regression discontinuity design.
• Reanalyzed Mexican Progresa program data.

Main Results:

• GAMLSS successfully modeled treatment effects on the entire conditional distribution, not just the mean.
• Application to the Progresa program revealed no statistically significant effects of cash transfers on conditional consumption inequality.
• The framework accommodates non-normal outcomes and complex effects like nonlinearity and spatial dependencies.

Conclusions:

• GAMLSS provide a flexible and comprehensive approach to analyzing treatment effects beyond the mean.
• The study highlights the importance of distributional analysis in program evaluation.
• No significant impact of the studied cash transfer program on consumption inequality was detected.