[Limdep Nlogit List] Tobit model with selectivity: inflated standard errors from probit
andrea.mannberg at econ.umu.se
andrea.mannberg at econ.umu.se
Fri Jan 15 06:26:37 EST 2010
Dear all,
Sorry for bothering you with this probably trivial question. The question
is related to the output in Nlogit4.0. I don't get an error message, but
something fishy is clearly going on, I just do not understand exactly what
it is. If any of you could give me a hint, I would appreciate it
immensely.
I am estimating a Tobit model with selectivity (as described in a previous
thread "tobit selectivity model with an endogenous regressor"). The model
is estimated without error messages and the coefficients in the first
probit and the tobit looks ok. However, when the program prints the output
for the probit together with the tobit, the standard errors have become
substantially inflated and all the p-values are equal to 1.000. Has anyone
else encountered this problem?
Let me describe the model and the output:
The code, as I have written it is (this is assuming I do not have
endogeneity issues as I have not solved that problem yet);
PROBIT
;Lhs=hadsex;Rhs=ONE,BLACK,COLOURED,FEMALE,EDUC,INFHIV,HIVKNOW,M_INC,MLIFEEXP,
VLIFEEXP, AGE_10, AGE_2, CHILDEXP ; Hold $
SELECT ; Lhs = RISK ; Rhs = ONE, BLACK, COLOURED, FEMALE, EDUC, INFHIV,
HIVKNOW, M_INC, MLIFEEXP, VLIFEEXP, AGE_10; Tobit ; MLE $
"RISK" is a continuous measure of sexual risk taking with a large pile of
zeroes for the individuals who are sexually active but have used condoms
constently. The "BLACK", "COLOURED", "FEMALE", "EDUC" and "HIVKNOW"
variables are dummy variables while the rest of the variables are
continuous. I have scaled the variables in order to facilitate convergence
(therefore the somewhat weird names, age is divided by ten etc.).
The output for the probit is (I have no idea what this will look like in
the list, so if it is a mess I apologize):
--> PROBIT ;
Lhs=hadsex;Rhs=one,black,coloured,Female,EDUC,INFHIV,HIVKNOW,m_i...
************************************************************************
* NOTE: Deleted 329 observations with missing data. N is now 2366 *
************************************************************************
Normal exit from iterations. Exit status=0.
+---------------------------------------------+
| Binomial Probit Model |
| Maximum Likelihood Estimates |
| Model estimated: Jan 14, 2010 at 08:14:40PM.|
| Dependent variable HADSEX |
| Weighting variable None |
| Number of observations 2366 |
| Iterations completed 5 |
| Log likelihood function -1093.097 |
| Number of parameters 13 |
| Info. Criterion: AIC = .93499 |
| Finite Sample: AIC = .93506 |
| Info. Criterion: BIC = .96669 |
| Info. Criterion:HQIC = .94653 |
| Restricted log likelihood -1465.224 |
| McFadden Pseudo R-squared .2539727 |
| Chi squared 744.2537 |
| Degrees of freedom 12 |
| Prob[ChiSqd > value] = .0000000 |
| Results retained for SELECTION model. |
| Hosmer-Lemeshow chi-squared = 6.49811 |
| P-value= .59162 with deg.fr. = 8 |
+---------------------------------------------+
+--------+--------------+----------------+--------+--------+----------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| Mean of X|
+--------+--------------+----------------+--------+--------+----------+
---------+Index function for probability
Constant| -14.0301092 1.90411328 -7.368 .0000
BLACK | .50649042 .16317627 3.104 .0019 .40490279
COLOURED| -.26603903 .13943244 -1.908 .0564 .50338123
FEMALE | -.21833665 .06273571 -3.480 .0005 .53677092
EDUC | .08856995 .09005374 .984 .3253 .85460693
INFHIV | -.09238782 .03513228 -2.630 .0085 2.54353339
HIVKNOW | .35172637 .09197509 3.824 .0001 .28021978
M_INC | -.14368614 .02708634 -5.305 .0000 1.41604635
MLIFEEXP| .01620880 .00566006 2.864 .0042 63.7612925
VLIFEEXP| .00021531 .00032177 .669 .5034 311.227860
AGE_10 | 13.5059902 2.15787740 6.259 .0000 1.73453085
AGE_2 | -3.11259868 .61435353 -5.066 .0000 3.06560862
CHILDEXP| .45129232 .14623046 3.086 .0020 .07227388
THE OUTPUT AFTER CALLING THE TOBIT MODEL IS:
Normal exit from iterations. Exit status=0.
+---------------------------------------------+
| ML Estimates of Selection Model |
| Maximum Likelihood Estimates |
| Model estimated: Jan 14, 2010 at 08:14:46PM.|
| Dependent variable INDEX_C |
| Weighting variable None |
| Number of observations 1284 |
| Iterations completed 69 |
| Log likelihood function -1895.199 |
| Number of parameters 26 |
| Info. Criterion: AIC = 2.99252 |
| Finite Sample: AIC = 2.99339 |
| Info. Criterion: BIC = 3.09696 |
| Info. Criterion:HQIC = 3.03173 |
| LHS is CENSORED. Tobit Model fit by MLE. |
| FIRST 13 estimates are probit equation. |
+---------------------------------------------+
+--------+--------------+----------------+--------+--------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
---------+Selection (probit) equation for HADSEX
Constant| -10.6595622 .125521D+11 .000 1.0000
BLACK | .64782696 .965882D+10 .000 1.0000
COLOURED| -.25308439 .966332D+10 .000 1.0000
FEMALE | -1.05106446 .238309D+09 .000 1.0000
EDUC | -1.23588145 .632344D+10 .000 1.0000
INFHIV | .02660619 .287335D+08 .000 1.0000
HIVKNOW | 2.70265320 .544509D+10 .000 1.0000
M_INC | .18813209 .765177D+08 .000 1.0000
MLIFEEXP| .05416825 .116988D+08 .000 1.0000
VLIFEEXP| .00222329 370561.366 .000 1.0000
AGE_10 | 16.0126340 .564273D+10 .000 1.0000
AGE_2 | -4.13664271 .151389D+10 .000 1.0000
CHILDEXP| 1.91080902 .582914D+10 .000 1.0000
---------+Corrected regression, Regime 1
Constant| -.42101297 1.78642248 -.236 .8137
BLACK | -.68912766 .54414152 -1.266 .2054
COLOURED| -.18156593 .49736417 -.365 .7151
FEMALE | 1.00650432 .18460798 5.452 .0000
EDUC | -.64845970 .28541129 -2.272 .0231
INFHIV | -.13643627 .09034565 -1.510 .1310
HIVKNOW | .31304299 .25107535 1.247 .2125
M_INC | -.21630205 .08331704 -2.596 .0094
MLIFEEXP| -.02287139 .02039172 -1.122 .2620
VLIFEEXP| .00126792 .00122946 1.031 .3024
AGE_10 | 1.65168214 .35400211 4.666 .0000
SIGMA(1)| 2.25722999 .13925410 16.209 .0000
RHO(1,2)| .98451832 .404146D+10 .000 1.0000
Now, I have clearly made som blunt misstake somewhere, but I am too new to
this to understand what it is I am missing. I am not even sure I am
providing the info needed for you to see the problem. If not, I would be
happy to do it if you just tell me what kind of information you need. As I
said above, I would really appreciate any help that you can provide.
Best regards,
Andrea Mannberg
More information about the Limdep
mailing list