You have a collinearity problem. In the first instance, the constant is
"dropped" while in the second, yr04 is "dropped". I would have expected Stata
to have dropped one variable before proceeding to estimation as it typically
does, but it apparently does not here. Remove one year dummy should solve
your problem.
HTH.
Partha
Hewan Belay wrote:
Dear Listers,
I am facing continued challenges with Stata's HausmanTaylor estimation command. Is it possible to get different coefficient values when running a rather basic oneline xthtaylor command twice??
This is what I am experiencing. More specifically, when I run the same oneline xthtaylor command repeatedly a couple times after each other (with no other command interspersed between these repeats), there are two (and only two, not 3 or more) different sets of commands that show up in any given run. E.g. when I run the command 6 times. Two times it may be result set A, and 4 times result set B. Another 6 runs and I may get 3 times set A and 3 times set B. etc. The order and frequency with which stata presents these two different results for a given number of repeat runs seems pretty arbitrary. One of these two result sets appears to be degenerate (intercept is zero, chisquare is through the roof).
What is going on? What is causing what appears to be some kind of binary stochasticity in generating the regression output for xthtaylor? This doesn't happen with all of my xthtaylor commands. It appears to happen in the specifications in which I have year dummies. I can understand that certain specifications may create degenerate outputs, or even error messages, but I don't understand why one would get different outputs on different runs of the identically same xthtaylor command. The variables in the model are fixed and are not generated through any random process, so that's not what's going on. In any case, I get different results in rerunning the same command twice (or multiple times) without running any other command in between these repeats.
To make it more concrete, here are the two different results I get (notice that between the two sets, the only coefficients that change values are the time dummies and the constant):
.. xthtaylor rev_IGF popurb_share popdens pop p0 rain_av road_no literate rel_christ *akan *ewe L.(rev_EXT exp_pers_act exp_NPR exp_cap_act) yr95yr04, endog( L.rev_EXT) varying(L.(rev_EXT exp_pers_act exp_NPR exp_cap_act) yr95yr04)
HausmanTaylor estimation Number of obs = 1028
Group variable: code Number of groups = 110
Obs per group: min = 6
avg = 9.3
max = 10
Random effects u_i ~ i.i.d. Wald chi2(24) = 85788.98
Prob > chi2 = 0.0000

rev_IGF  Coef. Std. Err. z P>z [95% Conf. Interval]
+
TVexogenous 
exp_pers_act 
L1.  .0458131 .0290744 1.58 0.115 .0111717 .1027979
exp_NPR 
L1.  .455071 .0324329 14.03 0.000 .3915037 .5186382
exp_cap_act 
L1.  .0251445 .0149172 1.69 0.092 .0040928 .0543817
yr95  4.727803 .4757539 9.94 0.000 3.795342 5.660263
yr96  4.847803 .4815382 10.07 0.000 3.904006 5.791601
yr97  4.702486 .4848417 9.70 0.000 3.752214 5.652758
yr98  4.774614 .4842344 9.86 0.000 3.825532 5.723696
yr99  4.717167 .4894038 9.64 0.000 3.757953 5.676381
yr00  4.710529 .4904061 9.61 0.000 3.74935 5.671707
yr01  4.732772 .4912148 9.63 0.000 3.770008 5.695535
yr02  5.031636 .4890057 10.29 0.000 4.073203 5.99007
yr03  4.932353 .4966293 9.93 0.000 3.958977 5.905728
yr04  5.003962 .5006898 9.99 0.000 4.022628 5.985296
TVendogenous 
rev_EXT 
L1.  .0333631 .0123719 2.70 0.007 .0576115 .0091146
TIexogenous 
popurb_share  .0010771 .001679 0.64 0.521 .0022136 .0043679
popdens  .0001782 .0001626 1.10 0.273 .0004969 .0001405
pop  .0003678 .0003443 1.07 0.285 .000307 .0010426
p0  1.336879 .3018935 4.43 0.000 1.928579 .7451785
rain_av  .000174 .0001693 1.03 0.304 .0005058 .0001578
road_no  .2678196 .1942266 1.38 0.168 .6484967 .1128574
literate  .0035209 .0047665 0.74 0.460 .0058213 .0128632
rel_christ  .0012864 .0029608 0.43 0.664 .0045167 .0070894
ethn_akan  .0025563 .0014801 1.73 0.084 .0054572 .0003446
ethn_ewe  .0024125 .0017515 1.38 0.168 .0058455 .0010204

_cons  0 0 . . 0 0
+
sigma_u  .24884157
sigma_e  .35746651
rho  .32641371 (fraction of variance due to u_i)

Note: TV refers to time varying; TI refers to time invariant.
..
end of dofile
.. xthtaylor rev_IGF popurb_share popdens pop p0 rain_av road_no literate rel_christ *akan *ewe L.(rev_EXT exp_pers_act exp_NPR exp_cap_act) yr95yr04, endog( L.rev_EXT) varying(L.(rev_EXT exp_pers_act exp_NPR exp_cap_act) yr95yr04)
HausmanTaylor estimation Number of obs = 1028
Group variable: code Number of groups = 110
Obs per group: min = 6
avg = 9.3
max = 10
Random effects u_i ~ i.i.d. Wald chi2(23) = 728.95
Prob > chi2 = 0.0000

rev_IGF  Coef. Std. Err. z P>z [95% Conf. Interval]
+
TVexogenous 
exp_pers_act 
L1.  .0458131 .0290744 1.58 0.115 .0111717 .1027979
exp_NPR 
L1.  .455071 .0324329 14.03 0.000 .3915037 .5186382
exp_cap_act 
L1.  .0251445 .0149172 1.69 0.092 .0040928 .0543817
yr95  .2761593 .0720174 3.83 0.000 .4173107 .1350078
yr96  .1561589 .0617779 2.53 0.011 .2772413 .0350765
yr97  .3014758 .0579642 5.20 0.000 .4150835 .1878681
yr98  .2293477 .0558028 4.11 0.000 .3387192 .1199761
yr99  .2867948 .0544548 5.27 0.000 .3935242 .1800653
yr00  .2934332 .0534874 5.49 0.000 .3982667 .1885998
yr01  .2711902 .0514675 5.27 0.000 .3720647 .1703157
yr02  .0276744 .0543115 0.51 0.610 .0787742 .134123
yr03  .0716094 .0503505 1.42 0.155 .1702947 .0270758
yr04  0 0 . . 0 0
TVendogenous 
rev_EXT 
L1.  .0333631 .0123719 2.70 0.007 .0576115 .0091146
TIexogenous 
popurb_share  .0010771 .001679 0.64 0.521 .0022136 .0043679
popdens  .0001782 .0001626 1.10 0.273 .0004969 .0001405
pop  .0003678 .0003443 1.07 0.285 .000307 .0010426
p0  1.336879 .3018935 4.43 0.000 1.928579 .7451785
rain_av  .000174 .0001693 1.03 0.304 .0005058 .0001578
road_no  .2678196 .1942266 1.38 0.168 .6484967 .1128574
literate  .0035209 .0047665 0.74 0.460 .0058213 .0128632
rel_christ  .0012864 .0029608 0.43 0.664 .0045167 .0070894
ethn_akan  .0025563 .0014801 1.73 0.084 .0054572 .0003446
ethn_ewe  .0024125 .0017515 1.38 0.168 .0058455 .0010204

_cons  5.003962 .5006898 9.99 0.000 4.022628 5.985296
+
sigma_u  .24884157
sigma_e  .35746651
rho  .32641371 (fraction of variance due to u_i)

Note: TV refers to time varying; TI refers to time invariant.
..
end of dofile
..
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/

Partha Deb
Department of Economics
Hunter College
ph: (212) 7725435
fax: (212) 7725398
http://urban.hunter.cuny.edu/~deb/
Emancipate yourselves from mental slavery
None but ourselves can free our minds.
 Bob Marley
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/