Hi there,

I have to report on the difference in estimates of 2SLS and OLS (same model) and the endogeneity of the variable of interest.

I have region-aggregate balanced panel data with N=22 and T=16

When using an LSDV specification, The coefficients are quite different and with -estat endogenous- I get evidence for endogeneity. However, when using a FD model, the coefficients are almost identical and -estat endogenous- gives no evidence for endogeneity.

How could this be? Also, what specification should I look at?


Below I've recreated my steps with the dataset nlswork to demonstrate my steps. The variable of interest is the LDV. The variables don't make any sense with this dataset, but it's just to show my thought process. I've had to replicate results from a paper, and the 2SLS estimates are identical to those in the paper, so I assume my commands are correct. (Also, the researcher preferred using the FD model and used 2SLS for his results).


Thank you in advance (:

Code:
clear all
webuse nlswork
xtset idcode year
bys idcode: egen count = count(_N)
keep if count==15
keep if inrange(year,68,73)
egen id = group(idcode)
xtset id year
keep if id<11

bys id: generate lagln_wage = ln_wage[_n-1]
bys id: generate laghours = hours[_n-1]


*2SlS
ivregress 2sls ln_wage (lagln_wage = laghours) hours tenure i.id year, small
estat endogenous

ivregress 2sls d.ln_wage (d.lagln_wage = d.laghours) d.hours d.tenure, small
estat endogenous

*OLS
regress ln_wage lagln_wage hours tenure i.id year
regress d.ln_wage d.lagln_wage d.hours d.tenure