Different results estimated by SAS and STATA for Fixed and Random Effects

[deleted_user]

Different results estimated by SAS and STATA for Fixed and Random Effects? The Hausman Test is also give different answers

is there any differences in estimation method employed by these software? I will be very thankful for any comments

These are the commands and Results I got from the two soft.


a- For Fixed Effect:
SAS:
proc reg data=chapter3.all_area;
Model Ln_qdt = Ln_qdt2 Ln_vkm Ln_income Ln_F deregulation_dummy Time_Trend London_dummy Mets_ dummy Scotland_ dummy Wales_ dummy;
test London_dv = Mets_dv = Scotland_dv = Wales_dv = 0 ;
run;

STATA:
regress Ln_qdt Ln_qdt2 Ln_vkm Ln_income Ln_F deregulation_dummy Time_Trend London_dummy Mets_ dummy Scotland_ dummy Wales_ dummy




b- For Random Effect:
SAS:

proc panel data=chapter3.all_area;
ID area year;
Model Ln_qdt = Ln_qdt2 Ln_vkm Ln_income Ln_F deregulation_dummy Time_Trend / RANONE BP VCOMP=WK
;
run;




STATA:
iis area
xtreg qdt Ln_qdt2 Ln_vkm Ln_income Ln_F deregulation_dummy Time_Trend, re theta

SAS STATA
Model FE RE FE (Using STATA) RE (Using STATA)
Coeff. Coeff. Coeff. Coeff.

Ln F -.108 -0.09892 -.108 -0.06321
Ln VKM .114 0.135992 .115 0.082707
Ln Income -.560 -0.52503 -.566 -0.297
Ln Qdt-1 .695 0.747298 .692 0.924061
Der. DV -.046 -0.04975 -.047 -0.05055
TT .011 0.010877 .011 0.00887

Mets .196 0.198
Scot .153 0.154
Wales -.023 -0.023
constant 5.999 5.908

F 3624.282 3618.020
R2 (Adj.) .997 0.973 0.9969 0.9967
Durbin-Watson 1.703

(Incremental) F 5.57 (0.0015) 5.57 (0.0015)
Breusch Pagan Test 0.00 (0.9781) 0.00 (0.9779)
Hausman Test 2.34 (0.8859) 20.16 (0.0026)

[Dale]

What do you get if you use the ORTHOREG procedure instead of PROC REG to fit the fixed effect model in SAS? If you get coefficients which differ from those produced by the REG procedure, then your data are probably ill conditioned, making it very difficult to get exact results.

The ORTHOREG procedure is designed to produce estimates for a fixed efffect model with more accuracy than the estimates produced by the REG procedure. There is no equivalent SAS procedure which produces highly accurate results for a mixed model. I can't tell you what option to use in STATA to produce highly accurate results.