Flashcards
Instrumental variables 8
Cards in this deck (3)
- 1Instrumental Variables (IV) Main Idea: To split the variation of D into 2parts. Find an instrumental variable (Z) which explains treatment but does not directly determine outcome and is not related to the error term. -Exogenous: part of D explained by the IV -Endogenous: part of D not explained by the IVRequirements of Z to become an IV:
Independence: Corr(Z, e | X) = 0; Z is required not to be related to the error term ; Z is randomly (as-if-random) assigned
Exclusion restriction: Z has no effect on Y except through D
First stage/ Relevance: Corr(Z, D | X) ≠ 0; Z needs to be (strongly) related to D; E.g. determine treatment status Monotonicity: No defiers - 2Z cannot fully explain D; Ín both cases we observed Compliers, Always-takers, Never-takers, Potentially also defiers; and we can therefore estimate LATE, i.e. the average causal effect of D for compliersStruggles with IV: Assumptions are ultimately untestable; Compliers are not always the sub population of greatest interest; Curse of strong instruments“: the closer Z is related to D, the more relevant Z is, and the more likely that Z suffers from the same endogeneity problems as D
- 3IVs can provide an imp tool because they help confront the issue of confounding. However whether IVs are only valid and useful depends on the RQ and setting. They often require strong assumptions, a priori reasoning and detailed knowledge of the empirical context.Issues: are often connected to the use of multiple regression models such as endogeneity of a single T variable leads researchers to use instrumental variables regression. Instruments that are truly rand may not be related to the endogenous T which leads to small-sample bias.
Information
- Description
- Flashcards on Instrumental variables 8, created by tri_nana on 28/06/2015.
- Created by
- tri_nana
- Views
- 1
- Rating