MG107: Management and Analytics in
the Age of Big Data
Lecture 8: Causal inference and
experimentation
Noam Yuchtman
July 19, 2023
Course overview:
Part 4: Experiments and causality
In many comparisons you make, you want to
attribute differences to some causal factor
How to evaluate causal claims
Experiments are a uniquely powerful (and
transparent) method of testing for causal
effects
Also some important limitations (e.g., external validity)
Outline
In-lecture exercise: interpreting patterns in data
Discussion and take-aways
Outline
In-lecture exercise: interpreting patterns in data
Discussion and take-aways
Exercise: the effects of tax rates
on business activity
Key Challenge: how to measure the “business
environment” across many very different countries
Andrei Shleifer and co-authors constructed a firm on
paper (it makes ceramic pots)
Then, they had PwC employees in 85 countries
prepare tax returns for the firm as if it was
domestically located
The authors use the information from PwC to
construct the “first-year effective tax rate” that the firm
would face (the units are percentages)
Exercise: the effects of tax rates
on business activity
The authors then try to tell a story about how
tax rate differences across countries affect
entrepreneurship
Graphical results plot different outcomes
against the effective tax rate, and show the
best-fit “OLS line” (which we’ll study soon)
describing the average relationship between
the tax rate and outcomes
Storytelling with graphs
Outline
In-lecture exercise: interpreting patterns in data
Discussion and take-aways
The challenge of identifying causal
effects
Generally, after observing a pattern in data, you need to ask:
1. What causal effect is suggested
2. What alternative stories are there for the observed pattern
The selection problem:
Reverse causality
Omitted variables
Want credible design up front: experiments! But think
through their limitations
