There seems to be a bug in using time series variables with xtprobit/xtlogit and robust SEs. It has been reported and discussed before here:
http://www.statalist.org/forums/foru...bit-or-xtlogit
http://www.statalist.org/forums/foru...atory-variable
http://www.statalist.org/forums/foru...s-with-xtlogit
It was reported fixed in the first thread, but apparently it's not. I have the latest stata 14 SE (march 30 2016). Here's an output sample to show the issue:
As you can see, i'm fitting a model with a lead on Y as the dependent variable.
The model with the regular vce estimates without an issue. trying robust SEs results in:
Calculating robust standard errors:
calculation of robust standard errors failed
r(198);
Yet generating the variable "by hand" as f_Y2 and requiesting robust SEs estimates without an issue. you can see that the number of observations and panel units is the same for all three regression commands, so this is most probably indeed a bug.
http://www.statalist.org/forums/foru...bit-or-xtlogit
http://www.statalist.org/forums/foru...atory-variable
http://www.statalist.org/forums/foru...s-with-xtlogit
It was reported fixed in the first thread, but apparently it's not. I have the latest stata 14 SE (march 30 2016). Here's an output sample to show the issue:
Code:
xtprobit f.Y x1 x2 x3 x4 x5 x6
Random-effects probit regression Number of obs = 898,717
Group variable: i Number of groups = 152,358
Random effects u_i ~ Gaussian Obs per group:
min = 1
avg = 5.9
max = 6
Integration method: mvaghermite Integration pts. = 12
Wald chi2(6) = 23273.05
Log likelihood = -549726.75 Prob > chi2 = 0.0000
-----------------------------------------------------------------------------------
F.Y | Coef. Std. Err. z P>|z| [95% Conf. Interval]
------------------+----------------------------------------------------------------
x1 | 4.20e-07 5.66e-09 74.16 0.000 4.09e-07 4.31e-07
x2 | .0144303 .0001676 86.08 0.000 .0141017 .0147588
x3 | .0711422 .0012444 57.17 0.000 .0687032 .0735812
x4 | -.1926475 .0061751 -31.20 0.000 -.2047504 -.1805445
x5 | -.3055726 .0232097 -13.17 0.000 -.3510627 -.2600824
x6 | .0802345 .0062972 12.74 0.000 .0678922 .0925768
_cons | -1.352513 .0094188 -143.60 0.000 -1.370974 -1.334052
------------------+----------------------------------------------------------------
/lnsig2u | -.6555548 .0072311 -.6697276 -.6413821
------------------+----------------------------------------------------------------
sigma_u | .7205234 .0026051 .7154355 .7256474
rho | .3417389 .0016267 .3385578 .3449342
-----------------------------------------------------------------------------------
LR test of rho=0: chibar2(01) = 7.2e+04 Prob >= chibar2 = 0.000
. xtprobit f.Y x1 c.x2##c.x2 x3 x4 x5 x6, vce(robust)
Fitting comparison model:
Calculating robust standard errors:
calculation of robust standard errors failed
r(198);
. gen f_Y2 = f.Y
. xtprobit f_Y2 x1 c.x2##c.x2 x3 x4 x5 x6, vce(robust)
Calculating robust standard errors:
Random-effects probit regression Number of obs = 898,717
Group variable: i Number of groups = 152,358
Random effects u_i ~ Gaussian Obs per group:
min = 1
avg = 5.9
max = 6
Integration method: mvaghermite Integration pts. = 12
Wald chi2(7) = 26456.46
Log pseudolikelihood = -548620.13 Prob > chi2 = 0.0000
(Std. Err. adjusted for 152,358 clusters in i)
-----------------------------------------------------------------------------------
| Robust
f_Y2 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
------------------+----------------------------------------------------------------
x1 | 3.65e-07 8.70e-09 41.90 0.000 3.48e-07 3.82e-07
x2 | .0601118 .001042 57.69 0.000 .0580695 .062154
|
c.x2#c.x2 | -.0004426 .0000103 -42.99 0.000 -.0004628 -.0004224
|
x3 | .0513391 .0014949 34.34 0.000 .0484092 .054269
x4 | -.1632547 .0069597 -23.46 0.000 -.1768954 -.1496139
x5 | -.328031 .0195098 -16.81 0.000 -.3662695 -.2897925
x6 | .0809134 .0061836 13.09 0.000 .0687937 .0930331
_cons | -2.369385 .0233229 -101.59 0.000 -2.415097 -2.323673
------------------+----------------------------------------------------------------
/lnsig2u | -.6275059 .0070927 -.6414073 -.6136045
------------------+----------------------------------------------------------------
sigma_u | .7306995 .0025913 .7256383 .7357961
rho | .3480763 .0016095 .3449285 .3512374
-----------------------------------------------------------------------------------
As you can see, i'm fitting a model with a lead on Y as the dependent variable.
The model with the regular vce estimates without an issue. trying robust SEs results in:
Calculating robust standard errors:
calculation of robust standard errors failed
r(198);
Yet generating the variable "by hand" as f_Y2 and requiesting robust SEs estimates without an issue. you can see that the number of observations and panel units is the same for all three regression commands, so this is most probably indeed a bug.