ECON 112 Lecture Notes - Lecture 9: Exploratory Data Analysis, External Validity, Propensity Score Matching
Document Summary
Comes from double blind trial in medicine. Because allocation into groups is independent of any characteristics of group members, the difference in means gives the causal impact of the treatment. Causal because people were randomly split into group. Another variable that correlates with both dependent variable and independent variable. Omission of a confounding variable leads to omitted variable bias. College attendance, effect on wages: academic ability. Agricultural education program with self selection, effect on income: motivation and diligence of farmers. Rcts: any confounding factor is distributed equivalently in both groups and hence, the effect is nullified - no need to even know what the factor is. If there is large enough number and randomized well. Well designed studies gives results that can be interpreted as causal. Can reduce publication bias - more upfront cost to conducting an experiment than just doing some exploratory data analysis of existing data.