Data: cross section Firm level data
Countries 40
Industries 25
Dependent Variables Y1 Y2 Y3 (Binary)
Indp Variable Firm Level
Import (Continous)
Importer (binary 0,1)
Instruments (Continuous Variables)
IV1 (No empirical evidence I found on IV1 as instrument variable), IV2 (Reported in past studies)
ivprobit Y (import=IV1) Controls i.industry i.country, vce(cluster country/Industey)
This model suggest that replacing Y with Y1 Y2 Y3 and clustering on country except for Y3 where Wald test is significant at 10% , Y1 Y2 are significant at less than 5%.However clustering on industry, Wald test is insignificant for Y1 and Y2 and Y3 are significant at <5%.
IV2 (Reported in past studies)
ivprobit Y (import=IV2) Controls i.industry i.country, vce(cluster country/Industey)
Clustering on country, Wald test for Y1 is insignificant while Y2 and Y3 are significant, however, clustering on industry then except for Y3 both Y1 and Y2 are insignificant.
Q1-My Question is what are the decision criteria for clustering on Country or Industry. Can I keep the results from clustering on country because it’s much better than clustering on industry?
Q2- I don’t have empirical evidence on IV1 but in my case results are much better than IV2 which is reported in previous studies, should I report both of them? if so then how should I explain this contradiction of both IVs?
Q3- Remember I have another endogenous regressor “importer” which is also Binary (0,1) and outcome Y is also binary,” biprobit seemingly unrelated equation” is it the correct approach, while instruments are continous?
biprobit (Y=importer controls i.industry i.country) (importer=IV1/IV2 Controls i.industry i.country)
Or maybe I should use
ivregress 2sls Y (import=IV1/IV2) controls i.industry i.country (cluster country) first
Which ignores the binary nature of outcome,
Guide please……….sorry for this long post because I wanted to make clear the situation wht I have……….THANKS IN ADVANCE