 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: output of dhurdle different from combination: Probit + truncreg
Dear Statalist Members
I am currently estimating a Cragg type two-hurdle model (Cragg 1971,
some statistical models for limited dependent variables with application
to the demand for durable goods).
It is well documented in the literature that this type of model can be
estimated in Stata using "probit" and "truncreg".
But there also exists a command "dhurdle" written by J. Fennema
(downloadable: http://www.sml.hw.ac.uk/somjaf/stata.htm
When I estimate both procedures, I get fairly different results
(truncation by 0):
dhurdle anteil_a ln_gpd_ ln_exp_ global_f corrupti plqi_ind
anteil_1 christ__ ln_pop_ communis, sel( probitest =ln_gpd_ ln_exp_
global_f corrupti plqi_ind anteil_1 christ__ ln_pop_ communis) indep
(where probitest is a dummyvariable indicating that it is truncated if =0)
Double hurdle model Number of
obs = 1706
(model with selection and censoring) Censored obs
= 432
Uncensored obs = 1274
Log likelihood = -3043.047 Independent errors
------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------+----------------------------------------------------------------
anteil_a |
ln_gpd_ | -.3560236 .0971659 -3.66 0.000 -.5464652
-.165582
ln_exp_ | .0068026 .0045881 1.48 0.138 -.00219
.0157951
global_f | .1737621 .1752133 0.99 0.321 -.1696496
.5171738
corrupti | .4872212 .2079483 2.34 0.019 .07965
.8947924
plqi_ind | -.0177344 .0045215 -3.92 0.000 -.0265963
-.0088725
anteil_1 | 24.90654 2.733824 9.11 0.000 19.54834
30.26473
christ__ | .0700423 .1389945 0.50 0.614 -.202382
.3424665
ln_pop_ | .5031898 .0521789 9.64 0.000 .400921
.6054586
communis | -1.49317 .596529 -2.50 0.012 -2.662345
-.3239948
_cons | -4.221061 .9567632 -4.41 0.000 -6.096282
-2.34584
-------------+----------------------------------------------------------------
probitest |
ln_gpd_ | -.5145746 .2017658 -2.55 0.011 -.9100283
-.1191209
ln_exp_ | .0033303 .0084804 0.39 0.695 -.013291
.0199515
global_f | .2962727 .3330498 0.89 0.374 -.3564929
.9490384
corrupti | -.0864238 .3263287 -0.26 0.791 -.7260163
.5531686
plqi_ind | .0170198 .0104676 1.63 0.104 -.0034962
.0375359
anteil_1 | 132368.3 199918.2 0.66 0.508 -259464.2
524200.8
christ__ | .1458046 .3011484 0.48 0.628 -.4444354
.7360447
ln_pop_ | .1061033 .0941986 1.13 0.260 -.0785225
.290729
communis | -.7897535 1.178506 -0.67 0.503 -3.099583
1.520076
_cons | 1.037665 1.693086 0.61 0.540 -2.280721
4.356052
-------------+----------------------------------------------------------------
/lnsigma | .8095791 .0200602 40.36 0.000 .7702618
.8488964
-------------+----------------------------------------------------------------
sigma | 2.246962 .0450746 2.160332
2.337066
------------------------------------------------------------------------------
other procedure:
probit anteil_a ln_gpd_ ln_exp_ global_f corrupti plqi_ind
anteil_1 christ__ ln_pop_ communis
Probit estimates Number of obs
= 1706
LR chi2(9)
= 785.92
Prob > chi2
= 0.0000
Log likelihood = -572.37985 Pseudo R2 = 0.4071
------------------------------------------------------------------------------
anteil_a | Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------+----------------------------------------------------------------
ln_gpd_ | -.4133993 .0608054 -6.80 0.000 -.5325758
-.2942228
ln_exp_ | .0103626 .0025291 4.10 0.000 .0054057
.0153195
global_f | .0721878 .1061359 0.68 0.496 -.1358347
.2802104
corrupti | .1809791 .1145471 1.58 0.114 -.043529
.4054872
plqi_ind | -.0030565 .0027014 -1.13 0.258 -.0083511
.0022381
anteil_1 | 288.4433 27.17308 10.62 0.000 235.1851
341.7016
christ__ | .075396 .0932029 0.81 0.419 -.1072782
.2580703
ln_pop_ | .2684596 .0299844 8.95 0.000 .2096913
.3272279
communis | -1.448146 .4195454 -3.45 0.001 -2.27044
-.6258523
_cons | -.8459712 .5491671 -1.54 0.123 -1.922319
.2303766
------------------------------------------------------------------------------
truncreg anteil_a ln_gpd_ ln_exp_ global_f corrupti plqi_ind
anteil_1 christ__ ln_pop_ communis, ll(0)
Truncated regression
Limit: lower = 0 Number of
obs = 1274
upper = +inf Wald
chi2(9) = 20.63
Log likelihood = -1362.0893 Prob > chi2 =
0.0144
------------------------------------------------------------------------------
anteil_a | Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------+----------------------------------------------------------------
eq1 |
ln_gpd_ | -32.7136 10.71833 -3.05 0.002 -53.72114
-11.70606
ln_exp_ | .513018 .2324739 2.21 0.027 .0573775
.9686585
global_f | -18.6152 11.84124 -1.57 0.116 -41.82361
4.593197
corrupti | -67.65709 22.36747 -3.02 0.002 -111.4965
-23.81765
plqi_ind | -1.82605 .5016721 -3.64 0.000 -2.80931
-.842791
anteil_1 | 373.6971 99.63596 3.75 0.000 178.4142
568.98
christ__ | -.9076562 9.842575 -0.09 0.927 -20.19875
18.38344
ln_pop_ | 45.06314 11.03242 4.08 0.000 23.44
66.68628
communis | -90.83652 35.48364 -2.56 0.010 -160.3832
-21.28987
_cons | -616.4984 160.5081 -3.84 0.000 -931.0884
-301.9083
-------------+----------------------------------------------------------------
sigma |
_cons | 14.19904 1.747253 8.13 0.000 10.77448
17.62359
------------------------------------------------------------------------------
For dhurdle there is no official help file, but one is written by a user:
http://www.datasets.org/statalist/archive/2008-01/msg00117.html
The results of dhurdle look more reliable on me (Coefficitents of
regression), although I cant imagine why the other procedure is wrong.....
As I need Tests for Normality (and correction of it) and
Heteroscedasticity, I would prefer using dhurdle (it has options for ihs
transformation and a het-option, see:
http://fmwww.bc.edu/repec/dsug2007/Fennema.pdf )
Do you have any suggestions where this different results come from an
what I can do?
Many thanks
Anita
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/