Instrumental-variables Estimators.doc

PAGE  PAGE 13 4. Instrumental-variables Estimators The fundamental assumption for consistency (A consistent estimator converges toward the parameter being estimated as the sample size increases; see Wooldridge p.170-171) of least-squares estimators is that the model error term is unrelated to the independent variables or regressors, . The only effect of x on y is a direct effect via the term (think about ). x y u where there is no association between x and u. So x and u are independent causes of y. If this assumption fails, the OLS estimator is inconsistent and the OLS estimator can no longer be given a causal interpretation. In particular, the OLS estimate can no longer be interpreted as estimating the marginal effect on the dependent variable y of an exogenous change in the i-th regressor variable . For example, consider a regression of log-earnings (y) on years of schooling (x). The error term u embodies all factors other than schooling that determine earnings such as ability. Suppose a person has a high level of u, as a result of high (unobserved) ability. This increases earnings, since , but it may also lead to higher levels of x, since schooling is likely to be higher for those with high ability. A more appropriate path diagram is then the following: x y u where now there is an association between x and u. Now higher levels of x have two effects on y. There is both a direct effect via and an indirect effect via u affecting x, which in turn affects y. The goal of regression is to estimate only the first effect, yielding an estimate of . We have with total derivative The OLS estimate will combine these two effects, giving (for example) where both effects are positive (). The OLS estimator is therefore biased and inconsistent for , unless there is no association between x and u. Examples of such failure include omitted variables, simultaneity, measurement