Causality, Ceteris Paribus, and Experiments (cont’d)
14
we can distinguish between
experimental data: “created” in a laboratory experiment
non-experimental / observational data: researcher = passive collector of the data
a large part of econometrics deals with how to get “correct” results despite working with non-experimental data.
Causality & Econometrics Sum-Up:
Econometric tools cannot be used to find causal links; these have to be found in economic theory. Econometrics can help us quantify causal effects and/or verify their presence. The challenge in here consist in dealing with non-experimental data where ceteris paribus conditions cannot be established.
A Note on Randomized Experiments
15
sometimes, even in experiments related to natural sciences, it
impossible to enforce ceteris paribus conditions
example (crop yields): assessing the effect of a new fertilizer on soybeans
ceteris paribus = ruling out other yield-affecting factors such as
rainfall, quality of land, presence of parasites etc.
experimental design:
Choose several one-acre plots of land.
Apply different amounts of fertilizer to each plot.
Use statistical methods to measure the association between yields and
fertilizer amounts.
drawback: some of yield-affecting factors are not fully observed →
impossible to choose “identical” plots of land
solution: statistical procedures still work correctly, if fertilizer amounts are independent of the other factors1 – e.g., if we choose fertilizer amounts completely at random → hence randomized
