ivtobit and mfx command

[Guest]

Hi, I am trying to figure out why my the result from computing marginal effects using mfx after ivtobit (eqn 2) is the same as the OLS (see eqn 1)?
Basically, after running ivtobit, I computed the marginal effects (3) using the command below.

mfx,predict(ystar(0,.)) varlist(ln_ttotex tariff new_fsize hgc1 hgc2 hgc3 bldg_type1 tenure1 urb aircon_qty pc_qty ref_qty tv_qty cellphone_qty)

I am particularly interested with the reversal of the sign of the variable new_fsize. Is it because of heteroskedasticity. Am I doing it correctly?

Many thanks!

--------------------------------------------------------------------
(eqn 1)
. regress ln_q_electricity ln_ttotex tariff new_fsize hhtype1 hgc1 hgc2 hgc3 bldg_type1 ///
> tenure1 urb aircon_qty pc_qty ref_qty tv_qty cellphone_qty,vce(r)

Linear regression Number of obs = 10,785
F(15, 10769) = 1951.33
Prob > F = 0.0000
R-squared = 0.7479
Root MSE = .59288

-------------------------------------------------------------------------------
| Robust
ln_q_electr~y | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------+----------------------------------------------------------------
ln_ttotex | .8441134 .0160047 52.74 0.000 .8127411 .8754856
tariff | -.055246 .0036667 -15.07 0.000 -.0624335 -.0480585
new_fsize | -.0294666 .003484 -8.46 0.000 -.0362959 -.0226374
hhtype1 | -.1160412 .0130697 -8.88 0.000 -.1416602 -.0904223
hgc1 | -.1706358 .0221079 -7.72 0.000 -.2139714 -.1273003
hgc2 | -.0868492 .0202549 -4.29 0.000 -.1265526 -.0471458
hgc3 | -.0094742 .0152363 -0.62 0.534 -.0393402 .0203917
bldg_type1 | -.1099637 .0230043 -4.78 0.000 -.1550563 -.064871
tenure1 | .1017127 .0143515 7.09 0.000 .0735811 .1298443
urb | .2626222 .0124795 21.04 0.000 .23816 .2870844
aircon_qty | .0596394 .0126529 4.71 0.000 .0348375 .0844414
pc_qty | -.0146301 .0104353 -1.40 0.161 -.0350852 .0058251
ref_qty | .4376226 .0155271 28.18 0.000 .4071865 .4680586
tv_qty | .1280694 .0114754 11.16 0.000 .1055756 .1505633
cellphone_qty | .0272034 .0046982 5.79 0.000 .0179942 .0364127
_cons | -3.423873 .1763979 -19.41 0.000 -3.769645 -3.0781
-------------------------------------------------------------------------------

(eqn 2)
. ivtobit ln_q_electricity tariff new_fsize hgc1 hgc2 hgc3 bldg_type1 tenure1 ///
> urb aircon_qty pc_qty ref_qty tv_qty cellphone_qty (ln_ttotex=wages employed_prof), ll(0) ul

Fitting exogenous tobit model

Fitting full model

Iteration 0: log likelihood = -10921.662
Iteration 1: log likelihood = -10899.397
Iteration 2: log likelihood = -10890.205
Iteration 3: log likelihood = -10890.07
Iteration 4: log likelihood = -10890.07

Tobit model with endogenous regressors Number of obs = 7,960
Wald chi2(14) = 16609.82
Log likelihood = -10890.07 Prob > chi2 = 0.0000

----------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-----------------+----------------------------------------------------------------
ln_ttotex | .2820867 .0764749 3.69 0.000 .1321986 .4319748
tariff | -.0298058 .0062657 -4.76 0.000 -.0420864 -.0175251
new_fsize | .0170321 .0066903 2.55 0.011 .0039194 .0301447
hgc1 | -.3553831 .0371551 -9.56 0.000 -.4282057 -.2825606
hgc2 | -.2101462 .0327534 -6.42 0.000 -.2743417 -.1459506
hgc3 | -.1041262 .0244564 -4.26 0.000 -.15206 -.0561925
bldg_type1 | -.1408628 .0343255 -4.10 0.000 -.2081394 -.0735861
tenure1 | .1565058 .0188411 8.31 0.000 .119578 .1934336
urb | .3517264 .0193347 18.19 0.000 .313831 .3896218
aircon_qty | .1499583 .0215558 6.96 0.000 .1077096 .1922069
pc_qty | .0944405 .0197229 4.79 0.000 .0557843 .1330967
ref_qty | .6047249 .0255706 23.65 0.000 .5546075 .6548423
tv_qty | .2074243 .0160965 12.89 0.000 .1758757 .2389729
cellphone_qty | .0853855 .0095698 8.92 0.000 .066629 .1041419
_cons | 2.496896 .8211542 3.04 0.002 .8874629 4.106328
-----------------+----------------------------------------------------------------
/alpha | .5897581 .0785896 7.50 0.000 .4357254 .7437907
/lns | -.4982022 .0079265 -62.85 0.000 -.5137378 -.4826666
/lnv | -.9715705 .0079259 -122.58 0.000 -.987105 -.9560361
-----------------+----------------------------------------------------------------
s | .6076221 .0048163 .5982552 .6171356
v | .3784881 .0029999 .372654 .3844137
----------------------------------------------------------------------------------
Instrumented: ln_ttotex
Instruments: tariff new_fsize hgc1 hgc2 hgc3 bldg_type1 tenure1 urb aircon_qty
pc_qty ref_qty tv_qty cellphone_qty wages employed_prof
----------------------------------------------------------------------------------
Wald test of exogeneity (/alpha = 0): chi2(1) = 56.31 Prob > chi2 = 0.0000

0 left-censored observations
7,959 uncensored observations
1 right-censored observation at ln_q_elect~y >= 10.296153

-------------------------------------------------------
(3)
Marginal effects after tobit
y = E(ln_q_electricity*|ln_q_electricity>0) (predict, ystar(0,.))
= 6.5589295
------------------------------------------------------------------------------
variable | dy/dx Std. Err. z P>|z| [ 95% C.I. ] X
---------+--------------------------------------------------------------------
tariff | -.0555393 .00357 -15.58 0.000 -.062527 -.048551 8.84239
new_fs~e | -.0212078 .00317 -6.70 0.000 -.027416 -.015 4.75086
hgc1*| -.1509938 .02124 -7.11 0.000 -.19263 -.109358 .174038
hgc2*| -.0714311 .02038 -3.50 0.000 -.111377 -.031485 .172369
hgc3*| -.0061862 .01611 -0.38 0.701 -.037767 .025395 .375522
bldg_t~1*| -.1049611 .02523 -4.16 0.000 -.154407 -.055515 .941029
tenure1*| .103883 .0139 7.47 0.000 .076644 .131122 .751507
urb*| .2603556 .01257 20.71 0.000 .23572 .284991 .482151
aircon~y | .0584996 .01297 4.51 0.000 .03308 .083919 .304404
pc_qty | -.0147926 .01158 -1.28 0.201 -.037485 .0079 .398887
ref_qty | .4482114 .0135 33.20 0.000 .421749 .474674 .541029
tv_qty | .1318323 .01085 12.15 0.000 .110567 .153098 1.11145
cellph~y | .0297863 .00489 6.09 0.000 .020194 .039379 2.0994
ln_tto~x | .8451629 .01457 58.01 0.000 .81661 .873716 12.0688
------------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1



.

[Richard Williams]

First off, your post would be much easier to read if you used code tags. See pt 12 of the FAQ.

Second, why are you using mfx? The margins command is much superior. If you aren't familiar with it, a brief overview is at

http://www3.nd.edu/~rwilliam/stats3/Margins01.pdf

Third, mfx is probably defaulting to xb, which is why the coefficients and marginal effects are the same. You want to specify some other option. If you have Stata 14, type

help ivtobit_postestimation

and then click on margins to see what your options are.

[Guest]

Thanks Richard for your suggestions. Yup, I'm using Stata 14.2. Reading it now.