5 and N=1519). The survey periods are 2007, 2010,2013,2016,and 2017. The panel data is unbalanced, but because I use Fixed-Effect model at least I mitigate some attrition problem. Using an IV (village mean of land ownership with documents) and testing for the endogeneity of the main regressor, I found out that land certificate is indeed endogenous and the IV is not a weak instrument. Here is the result using -xtivreg,endog()- command
Code:
global y1 hh_simpson_crop_area global y22 document global x1 ln_cultivatedarea irrigation /// female_head age_head age2 dep_ratio mean_education /// tractor moto /// assets ln_remit credit /// phone hh_spo /// envir_shock /// made_road ln_distance villassets /// ln_offfarm ln_selfemp ln_livestock ln_envir_inc tab time, gen(t)
Code:
xtivreg2 $y1 ($y22=villdoc_excl) $x1 t1-t5, fe endog($y22)
Code:
-------------------------------------------------------------------------------- Underidentification test (Anderson canon. corr. LM statistic): 312.274 Chi-sq(1) P-val = 0.0000 ------------------------------------------------------------------------------ Weak identification test (Cragg-Donald Wald F statistic): 335.118 Stock-Yogo weak ID test critical values: 10% maximal IV size 16.38 15% maximal IV size 8.96 20% maximal IV size 6.66 25% maximal IV size 5.53 Source: Stock-Yogo (2005). Reproduced by permission. ------------------------------------------------------------------------------ Sargan statistic (overidentification test of all instruments): 0.000 (equation exactly identified) -endog- option: Endogeneity test of endogenous regressors: 10.963 Chi-sq(1) P-val = 0.0009
Code:
---------------------------------------------------- (1) (2) Simpson Di~x Simpson Di~x ---------------------------------------------------- Share of Documente~d -0.0141* -0.0997*** (-1.92) (-3.64) Constant 0.0721 0.110 (0.68) (1.02) ---------------------------------------------------- Observations 5865 5863 Instrument No Yes Control Variables Yes Yes Time Control Yes Yes Robust No No ---------------------------------------------------- t statistics in parentheses * p<0.10, ** p<0.05, *** p<0.010
However, I have a problem with testing the assumption of the fixed-effect model. I am not sure whether there is a strict exogeneity between the variables. And the variance distribution could possibly be homoscedastic too. I cannot produce a postestimation using -xttest- command for -xtivreg- command. However, after using the option -vce(cluster var)- command I see moderate difference in the t-stat between the normal standard errors and the robust standard errors, the result looks like following
Code:
---------------------------------------------------- (1) (2) Simpson Di~x Simpson Di~x ---------------------------------------------------- Share of Documente~d -0.0997*** -0.0997** (-3.64) (-2.41) Constant 0.110 0.110 (1.02) (0.81) ---------------------------------------------------- Observations 5863 5863 Instrument Yes Yes Control Variables Yes Yes Time Control Yes Yes Robust No Yes ---------------------------------------------------- t statistics in parentheses * p<0.10, ** p<0.05, *** p<0.010
1.Can I have a valid inference with the normal standard error or should I go with the clustered-robust standard error?
2.Since I cannot use the -vif- command for the estimation using-xtivreg- , should I do to test for the multicollinearity one by one?
3.Other than giving arguments for the exogeneity assumption of the IV, should I worry about other things for my IV estimation after seeing this result?
Thank you,
Raihan
0 Response to XTIVREG: Fixed-effect model with IV robust or non-robust? IV choice and normal se assumptions for FE
Post a Comment