Hi everyone,
I'm estimating the following model via IV regressions:
Y = b1*W + b2*X +u,
where Y is a non-negative integer variable; W is a binary endogenous regressor; X are control variables, including continuous and categorized variables, as well as regional and time fixed effects. X, Y, W are individual-level.
Following Wooldridge (2010) and the previous post, https://www.statalist.org/forums/for...nous-regressor, I tried the three models:
In the first and third models, the IV estimates are significantly positive, consistent with the original OLS estimation; however, in the second model, the IV estimate is insignificant. In all three IV regressions, instruments are strong enough that the first-stage F statistics are above 10.
Questions: why results are inconsistent across these three models? What instruments shall I use?
Further question: if Y is an ordinal independent variable and the main model is an ordinal logistic regression, how to apply IV in this case?
Many thanks for your time and help.
Sincerely,
Yupu
I'm estimating the following model via IV regressions:
Y = b1*W + b2*X +u,
where Y is a non-negative integer variable; W is a binary endogenous regressor; X are control variables, including continuous and categorized variables, as well as regional and time fixed effects. X, Y, W are individual-level.
Following Wooldridge (2010) and the previous post, https://www.statalist.org/forums/for...nous-regressor, I tried the three models:
- just run the 2SLS regression using Z as an instrument for W. Z is continuous and regional-level. i.e., ivregress 2sls Y (W=Z) X
- use the 2-step IV regression: 1. estimate W = a1* Z + a2* X + v using a logit model and get the fitted value, W_hat; 2. estimate the main equation by IV using W_hat as an instrument for W. i.e., ivregress 2sls (W=W_hat) X
- use the 2-step IV regression: 1. estimate W = a1*Z + a2*X + v using a logit model and get the fitted value, W_hat; 2. estimate the main equation by IV using both Z and W_hat as instruments for W. i.e., ivregress 2sls (W = Z W_hat) X
In the first and third models, the IV estimates are significantly positive, consistent with the original OLS estimation; however, in the second model, the IV estimate is insignificant. In all three IV regressions, instruments are strong enough that the first-stage F statistics are above 10.
Questions: why results are inconsistent across these three models? What instruments shall I use?
Further question: if Y is an ordinal independent variable and the main model is an ordinal logistic regression, how to apply IV in this case?
Many thanks for your time and help.
Sincerely,
Yupu
