Research Questions in Econometrics

Summary

Good econometrics starts with a good research agenda. The four “Frequently Asked Questions” provide a framework for organizing any empirical research project.

The Four FAQs

1. What is the causal relationship of interest?

The most interesting research in social science is about cause and effect. Example: the causal effect of schooling on wages (~40% higher wages for a college degree).

2. What experiment could ideally capture this effect?

Imagine yourself with no budget constraint and no IRB. What experiment would you run? Questions that cannot be answered by any experiment are FUQ’d (Fundamentally Unidentified Questions).

3. What is your identification strategy?

How does your research design use observational data to approximate the ideal experiment? Chapters 3-6 of MHE provide the conceptual frameworks.

4. What is your mode of statistical inference?

What population, sample, and assumptions underlie your standard errors? Covered in Chapter 8.

FUQ'd Questions

Questions about the causal effect of immutable characteristics (race, gender) seem FUQ’d but can be reframed: labor market discrimination turns on whether someone believes you are a certain race/gender, which can be manipulated (e.g., audit studies with fake resumes).

See Also

• Mostly Harmless Econometrics - Overview — full book overview
• The Experimental Ideal — the ideal benchmark for FAQ #2
• The Selection Problem — the fundamental identification challenge
• Conditional Independence Assumption — selection on observables strategy (FAQ #3)
• Instrumental Variables — instrument-based identification (FAQ #3)
• Differences-in-Differences — panel data identification (FAQ #3)
• Regression Discontinuity Designs — threshold-based identification (FAQ #3)
• Standard Errors and Clustering — mode of statistical inference (FAQ #4)
• Bayesian Workflow - Overview — iterative Bayesian research design paralleling the FAQ framework
• Omitted Variables Bias — the specific failure mode that FAQ #3 (identification strategy) aims to rule out
• Activity Bias in Advertising — a worked example showing all four FAQs applied to an advertising measurement problem