The Simple Geometry of Correlated Regressors and IV Corrections

In medical research, frequently other important determinants, correlated with the key treatment variable, are omitted from the analysis. This omission yields biased and inconsistent estimates. For example, leaving out correlated (with, say, ) determinants of Y from regressions yield biased estimates of key parameters (say ). Instrumental variable estimation solves this problem by constructing similar triangles to retrieve consistent estimates. This article illustrates the geometry of correlated regressor bias, and the simple IV geometric solution.