Dear Md Azizur Rahman ,
As a general rule, in those situations, the most suitable strategy is just to employ 2SLS. Consistency of estimates in this framework only requires no omitted relevant variable/endogeneity, although your standard errors can be higher than if you use a model that takes into acount the exact nature of the variable (e.g., a negative binomial model for count data or a logit or a probit for binary variables). In the latter cases, if there are additional problems in the model (e.g., disturbance does not follow the assumed distribution), this can result in inconsistent estimators. Therefore, a conservative approach is just to use 2OLS. You can find a reference backing this argument in Angrist and Pischke (2008) Mostly harmless Econometrics.
I hope it helps.
José-Ignacio Antón
Thank you so much Mr. José-Ignacio Antón for your kind suggestions.
Thanks Mr. Brilliant Professor. You are really and truly economics guy.... would you please suggest any reading materials?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression