The Fundamental Problem of Causal Inference

2. The Fundamental Problem of Causal Inference • Confounds and The Fundamental Problem of Causal Inference • Probabilistic vs. Deterministic Causality • Four Criteria for Showing Causality

3. Confounding Variable • A confounding variable: • causes changes in the dependent variable • is correlated with one of the independent variables • is “causally prior” to that independent variable. Chronologically or logically, it comes first. Wealth Prior Current Revolution Health Revolution

4. The Fundamental Problem of Causal Inference - 1 • Problem. We cannot rerun history to see whether changing the value of an independent variable would have changed the value of the dependent variable. • Solution #1. Give up.

5. The Fundamental Problem of Causal Inference - 2 • Solution #2. Design your research in a way that comes as close as possible to rerunning history. • Observe the effects of changes in one independent variable when all other independent variables remain the same, or • Measure other independent variables, then use statistical techniques to hold them constant.

6. Probabilistic vs. Deterministic Causality (Definitions) • “Probabilistic” means that when the values that an IV takes on increase, this usually results in the values of the DV increasing (or, usually, decreasing) • “Deterministic” means that when the values that an IV takes on increase, this always results in the values of the DV increasing (or, always, decreasing)

7. Why Political Science is Satisfied with a Probabilistic Notion of Cause • Like many other sciences that study complex systems, we care about necessary or sufficient causal factors that make an effect more likely, not just iron laws. • More education more likely to vote. • Cities that rely more on sales tax more likely to subsidize WalMarts

8. Four Criteria for Showing Causality • #1 Temporal Ordering • #2 Correlation • #3 Causal Mechanism • #4 Rule out Confounds

9. Criterion #1. Temporal Ordering • The hypothesized CAUSE (IV) must come BEFORE the EFFECT (DV). • Students decide whether or not to sit in the front of class before the get their final grade. • Campaign contributions on the eve of an election can’t cause a Congresswoman’s voting record in the previous session. • Political science has lots of tricky “chicken-and-egg” situations.

10. Criterion #2. Correlation • Two variables are “correlated” when changes in one variable occur TOGETHER with changes in the other • Correlation is roughly synonymous with association and co-variance. • A correlation between two variables can be positive or negative.

11. Criterion #3. Causal Mechanism • You have to be able to tell a plausible story that CONNECTS the IV to the DV • This story often includes an “intervening variable” that gets us from the IV to the DV • Students who sit up front are able to hear better, see better, and better comprehend the lecture (plausible story)

12. Criterion #4. Rule Out Confounds • If there is a confound that is causally PRIOR to BOTH an IV and a DV, then the correlation we observe between the IV and the DV may be SPURIOUS. • A possible confound is that more dedicated students are more likely to: a. sit up front, and b. perform well on the test. The observed correlation between their seating choice and their performance may be spurious.