Karteikarten
Matching estimators 9
Karten in diesem Stapel (3)
- 1Matching Basic idea: Multiple regression approach to causal inference „statistically controls“ for confounders X included in the regression equation by subtracting (BX) from each observation->is non-parametric, discards observations off common supportOne on one:Select (D= 0, D= 1) pairs with exactly the same values on X (ATE is interpreted as ATT)
One on many:Gain in efficiency by utilizing more than one direct counterparts in control state, in particular if control group-> treatment group - 2Propensity Score (PS) Tautology: Use X variables to calculate propensity scores, match cases according these scores , X values should be balanced between treatment & control group, if yes it is correct functional form/if no ->specify functional form in calculating the PSMatching can be used to find: logit prediction that person‘s polling place is closed, people without closed poll, similar people with and without closed poll, people that don‘t have similar counterpart and logit regression of turnout on closed poll, new distance, etc
- 3Strengths: matching focuses its attention on the heterogeneity of the causal effectWeaknesses: the inability to estimate the variance of most matching estimators with commonly accepted methods; diff matching estimators may lead to somewhat diff estimates of causal effects; perfect balance on imp covariates doesn't necessarily warrant casual inference
Informationen
- Beschreibung
- Karteikarten am Matching estimators 9, erstellt von tri_nana am 28/06/2015.
- Erstellt von
- tri_nana
- Aufrufe
- 0
- Bewertung