Issues with Time-Invariant Variables when using Stack Command in panel regression

[Ashley Roccapriore]

Hi everyone -

I'm trying to use the stack command to compare the effects of the same independent variable on two different dependent variables. We're using a Hausman Taylor (HT) regression using xthtaylor, as we need to account for random effects and potential unobserved variables creating endogeneity issues.

This becomes an issue, however, when using the stack command because we can either: 1) have all the time-invariant variables show the same in both stacks, or 2) have the time-invariant variables take on a 0 depending on it's corresponding stack. When we do option 1, our coefficients are not the same as when we run the HT regression since it's not accounting for separate time-invariant variables, and 2) it gives us an error for HT saying there are no time-invariant variables in the model.

We did attempt to cluster the id variable in the HT regression with the stack (created a concatatend variable of id and stack number), which helped get us closer, but still not the same coefficients.

Our main concern is we aren't sure why the results are different than when we run the models alone (i.e., just using HT regression without stack), when compared to the version with the stack command since they should be the same.

Anyone have a solution for time-invariant variables in panel regression when using stack?

Thanks in advance for your help!

[Andrew Musau]

I think I got it to work by reassigning new ids to the observations corresponding to the second variable. It's annoying that the command does not support factor variables.

webuse psidextract, clear
set seed 09232021
gen depvar2=rnormal()
xthtaylor lwage wks exp exp2 fem blk ed, endog(exp exp2 ed) constant(fem blk ed)
xthtaylor depvar2 wks exp exp2 fem blk ed, endog(exp exp2 ed) constant(fem blk ed)
rename (lwage depvar2) (var1 var2)
reshape long var, i(id t) j(which) 
qui sum id
replace id= id + `r(max)' if which==2
xtset id t
gen which1=1.which
gen which2=2.which
gen cons=1
foreach var in wks exp exp2 fem blk ed cons{
    gen which1`var'=which1*`var'
    gen which2`var'=which2*`var'
}
xthtaylor var which1wks which2wks which1exp which2exp which1exp2 which2exp2 which1fem which2fem ///
which1blk which2blk which1ed which2ed which1cons which2cons, nocons endog(which1exp which2exp ///
which1exp2 which2exp2 which1ed which2ed) constant(which1fem which2fem which1blk which2blk ///
which1ed which2ed which1cons which2cons) vce(robust)

Res.:

. xthtaylor lwage wks exp exp2 fem blk ed, endog(exp exp2 ed) constant(fem blk ed)

Hausman-Taylor estimation                       Number of obs     =      4,165
Group variable: id                              Number of groups  =        595

                                                Obs per group:
                                                              min =          7
                                                              avg =          7
                                                              max =          7

Random effects u_i ~ i.i.d.                     Wald chi2(6)      =    6815.88
                                                Prob > chi2       =     0.0000

------------------------------------------------------------------------------
       lwage |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
TVexogenous  |
         wks |   .0008359   .0005999     1.39   0.164      -.00034    .0020117
TVendogenous |
         exp |   .1137879   .0024699    46.07   0.000     .1089469    .1186288
        exp2 |  -.0004244   .0000547    -7.76   0.000    -.0005315   -.0003172
TIexogenous  |
         fem |   .3669631   1.677234     0.22   0.827    -2.920355    3.654281
         blk |  -3.249538   9.037829    -0.36   0.719    -20.96336    14.46428
TIendogenous |
          ed |  -2.065993   6.688445    -0.31   0.757     -15.1751    11.04312
             |
       _cons |   31.32838   86.40063     0.36   0.717    -138.0137    200.6705
-------------+----------------------------------------------------------------
     sigma_u |   6.185633
     sigma_e |   .1521392
         rho |  .99939542   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

. xthtaylor depvar2 wks exp exp2 fem blk ed, endog(exp exp2 ed) constant(fem blk ed)

Hausman-Taylor estimation                       Number of obs     =      4,165
Group variable: id                              Number of groups  =        595

                                                Obs per group:
                                                              min =          7
                                                              avg =          7
                                                              max =          7

Random effects u_i ~ i.i.d.                     Wald chi2(6)      =       4.15
                                                Prob > chi2       =     0.6562

------------------------------------------------------------------------------
     depvar2 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
TVexogenous  |
         wks |    -.00764   .0038844    -1.97   0.049    -.0152533   -.0000267
TVendogenous |
         exp |  -.0025355   .0159923    -0.16   0.874    -.0338797    .0288087
        exp2 |   .0000138   .0003539     0.04   0.969    -.0006798    .0007074
TIexogenous  |
         fem |   .0227675   .4804313     0.05   0.962    -.9188605    .9643956
         blk |  -.8402209   2.590867    -0.32   0.746    -5.918227    4.237786
TIendogenous |
          ed |  -.6169704    1.91641    -0.32   0.747    -4.373066    3.139125
             |
       _cons |    8.35608   24.77947     0.34   0.736    -40.21079    56.92295
-------------+----------------------------------------------------------------
     sigma_u |  1.7002202
     sigma_e |  .98507038
         rho |  .74868288   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.


. xthtaylor var which1wks which2wks which1exp which2exp which1exp2 which2exp2 which1fem which2fem ///
> which1blk which2blk which1ed which2ed which1cons which2cons, nocons endog(which1exp which2exp ///
> which1exp2 which2exp2 which1ed which2ed) constant(which1fem which2fem which1blk which2blk ///
> which1ed which2ed which1cons which2cons) vce(robust)

Hausman-Taylor estimation                       Number of obs     =      8,330
Group variable: id                              Number of groups  =      1,190

                                                Obs per group:
                                                              min =          7
                                                              avg =          7
                                                              max =          7

Random effects u_i ~ i.i.d.                     Wald chi2(14)     =    3580.96
                                                Prob > chi2       =     0.0000

                                 (Std. Err. adjusted for 1,190 clusters in id)
------------------------------------------------------------------------------
             |               Robust
         var |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
TVexogenous  |
   which1wks |   .0008359   .0008697     0.96   0.336    -.0008687    .0025404
   which2wks |    -.00764   .0042352    -1.80   0.071    -.0159408    .0006608
TVendogenous |
   which1exp |   .1137879   .0040289    28.24   0.000     .1058914    .1216843
   which2exp |  -.0025355   .0161228    -0.16   0.875    -.0341356    .0290646
  which1exp2 |  -.0004244   .0000822    -5.16   0.000    -.0005855   -.0002632
  which2exp2 |   .0000138   .0003547     0.04   0.969    -.0006814     .000709
TIexogenous  |
   which1fem |   .3669631   1.761769     0.21   0.835     -3.08604    3.819967
   which2fem |   .0227675    .505234     0.05   0.964    -.9674728    1.013008
   which1blk |  -3.249538   10.13487    -0.32   0.748    -23.11351    16.61444
   which2blk |  -.8402209   2.896666    -0.29   0.772    -6.517583    4.837141
  which1cons |   31.32838   97.25625     0.32   0.747    -159.2904    221.9471
  which2cons |    8.35608   27.78633     0.30   0.764    -46.10412    62.81628
TIendogenous |
    which1ed |  -2.065993   7.531867    -0.27   0.784    -16.82818     12.6962
    which2ed |  -.6169704   2.148765    -0.29   0.774    -4.828472    3.594531
-------------+----------------------------------------------------------------
     sigma_u |   4.536122
     sigma_e |  .70480848
         rho |  .97642712   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

[Ashley Roccapriore]

Thanks Andrew Musau! I apologize for the delay in replying; I've been trying to troubleshoot before replying, as I tried your same code with our data and we are having the same exact problem. I think, and could be completely wrong here, but the issue had something to do with our intercept or constant.

For example, in the results I have below, using the code you so kindly provided, the variable "which2Cred~y" has a p-value of 0.614. But in the broken out HT regression, you can see that the same variable (labeled as " CredBinary" in that one), has a p-value of 0.085, so marginally significant.

The only thing I can come up with is something with the constants. As you can see from the code right below this, I'm including the "which" variable as a constant to attempt to account for this. However, it's still off.

Any advice or thoughts as to why this may be happening? I appreciate any help you (or anyone else) may be able to provide!

---------------------------------------------------*****************---------------------------------------------------
This is the code I'm using, which gets me closest to our unstacked results, but still not there:

rename (z_LNdv1 z_dv2) (var1 var2)
reshape long var, i(id month) j(which)
qui sum id
replace id=id + `r(max)' if which==2
xtset id month
gen which1=1.which
gen which2=2.which
gen cons=1
foreach var in postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage cons{
gen which1`var'=which1*`var'
gen which2`var'=which2*`var'
}

xthtaylor var which1postwc which1commentwc which1attract which1fit which1gend which1ethn which1postsmo which1commentsmo which1numreplies which1vidpost which1postsdel which1profage which1CredBinary which1warmth which1compimage which1warmimage which2postwc which2commentwc which2attract which2fit which2gend which2ethn which2postsmo which2commentsmo which2numreplies which2vidpost which2postsdel which2profage which2CredBinary which2warmth which2compimage which2warmimage which, vce(robust) endog(which1postwc which1commentwc which1postsmo which1commentsmo which1numreplies which1profage which1warmth which2postwc which2commentwc which2postsmo which2commentsmo which2numreplies which2profage which2warmth) constant(which1attract which1fit which1gend which1ethn which2attract which2fit which2gend which2ethn which)

Results:

xthtaylor var which1postwc which1commentwc which1attract which1fit which1gend which1ethn which1postsmo which1commentsmo which1numreplies whic
> h1vidpost which1postsdel which1profage which1CredBinary which1warmth which1compimage which1warmimage which2postwc which2commentwc which2attra
> ct which2fit which2gend which2ethn which2postsmo which2commentsmo which2numreplies which2vidpost which2postsdel which2profage which2CredBinar
> y which2warmth which2compimage which2warmimage which,  vce(robust) endog(which1postwc which1commentwc which1postsmo which1commentsmo which1nu
> mreplies which1profage which1warmth which2postwc which2commentwc which2postsmo which2commentsmo which2numreplies which2profage which2warmth)
> constant(which1attract which1fit which1gend which1ethn which2attract which2fit which2gend which2ethn which)

Hausman–Taylor estimation                       Number of obs     =      5,832
Group variable: id                              Number of groups  =        972

                                                Obs per group:
                                                              min =          6
                                                              avg =          6
                                                              max =          6

Random effects u_i ~ i.i.d.                     Wald chi2(33)     =    3075.09
                                                Prob > chi2       =     0.0000

                                   (Std. err. adjusted for 972 clusters in id)
------------------------------------------------------------------------------
             |               Robust
         var | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
TVexogenous  |
which1vidp~t |   .0006066   .0006306     0.96   0.336    -.0006293    .0018426
which2vidp~t |  -.0023396   .0015432    -1.52   0.129    -.0053641    .0006849
which1post~l |   .0016425   .0027351     0.60   0.548    -.0037183    .0070032
which2post~l |  -.0060408   .0034218    -1.77   0.077    -.0127474    .0006658
which1Cred~y |   .0025215   .0142716     0.18   0.860    -.0254504    .0304933
which2Cred~y |  -.0050212   .0099669    -0.50   0.614    -.0245561    .0145136
which1comp~e |   .0233614    .005695     4.10   0.000     .0121993    .0345234
which2comp~e |  -.0064016   .0167763    -0.38   0.703    -.0392826    .0264793
which1warm~e |     .00497   .0026448     1.88   0.060    -.0002137    .0101537
which2warm~e |   .0185382   .0085721     2.16   0.031     .0017372    .0353393

TVendogenous |
which1postwc |   2.59e-06   3.27e-06     0.79   0.429    -3.83e-06    9.00e-06
which2postwc |   -.000033   .0000126    -2.62   0.009    -.0000577   -8.28e-06
which1comm~c |   8.13e-08   8.24e-08     0.99   0.324    -8.03e-08    2.43e-07
which2comm~c |    .000023   1.88e-06    12.28   0.000     .0000193    .0000267
which1post~o |  -.0002648    .000626    -0.42   0.672    -.0014918    .0009621
which2post~o |   .0031033   .0012865     2.41   0.016     .0005818    .0056249
which1comm~o |  -1.10e-07   1.77e-07    -0.62   0.534    -4.57e-07    2.37e-07
which2comm~o |  -.0000149   4.00e-06    -3.74   0.000    -.0000228   -7.11e-06
which1numr~s |   8.95e-06   7.06e-06     1.27   0.205    -4.89e-06    .0000228
which2numr~s |    .000198   .0000453     4.37   0.000     .0001092    .0002869
which1prof~e |   .1079754   .0144306     7.48   0.000     .0796919    .1362589
which2prof~e |   .0252472   .0276266     0.91   0.361    -.0288999    .0793944
which1warmth |   -.012735   .0071575    -1.78   0.075    -.0267635    .0012934
which2warmth |   .1199108   .0421445     2.85   0.004     .0373091    .2025125

TIexogenous  |
which1attr~t |   .4590263   .0896882     5.12   0.000     .2832407    .6348119
which2attr~t |   .0758922    .022501     3.37   0.001     .0317911    .1199933
   which1fit |   .3674291    .110017     3.34   0.001     .1517998    .5830584
   which2fit |   .0256794   .0383368     0.67   0.503    -.0494594    .1008182
  which1gend |   .2004464   .0877548     2.28   0.022     .0284501    .3724426
  which2gend |   .1357634   .0309924     4.38   0.000     .0750194    .1965073
  which1ethn |   .0522698   .0976012     0.54   0.592     -.139025    .2435647
 which2ethn |   .0531484   .0463822     1.15   0.252    -.0377591    .1440559
       which |   .6777835   .1807456     3.75   0.000     .3235286    1.032038
             |
       _cons |  -1.990053   .2963166    -6.72   0.000    -2.570823   -1.409283
-------------+----------------------------------------------------------------
     sigma_u |  .66821243
     sigma_e |  .14033951
         rho |  .95775408   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

---------------------------------------------------*****************---------------------------------------------------
Now when I run the Hausman Taylor without Stacking, this is what I get:

First DV:

xthtaylor z_LNdv1 postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth com
> pimage warmimage, vce(robust) endog(postwc commentwc postsmo commentsmo numreplies profage warmth)        

Hausman–Taylor estimation                       Number of obs     =      2,916
Group variable: id                              Number of groups  =        486

                                                Obs per group:
                                                              min =          6
                                                              avg =          6
                                                              max =          6

Random effects u_i ~ i.i.d.                     Wald chi2(16)     =     150.21
                                                Prob > chi2       =     0.0000

                                   (Std. err. adjusted for 486 clusters in id)
------------------------------------------------------------------------------
             |               Robust
     z_LNdv1 | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
TVexogenous  |
     vidpost |   .0002047   .0005317     0.39   0.700    -.0008373    .0012468
    postsdel |    .002598   .0027049     0.96   0.337    -.0027035    .0078995
  CredBinary |   .0155909   .0109847     1.42   0.156    -.0059386    .0371205
   compimage |   .0141383    .005277     2.68   0.007     .0037955     .024481
   warmimage |   .0036695   .0024819     1.48   0.139     -.001195     .008534

TVendogenous |
      postwc |   2.65e-06   3.14e-06     0.84   0.398    -3.50e-06    8.81e-06
   commentwc |   3.34e-08   7.37e-08     0.45   0.651    -1.11e-07    1.78e-07
     postsmo |  -.0005505   .0006189    -0.89   0.374    -.0017634    .0006625
  commentsmo |  -2.44e-08   1.56e-07    -0.16   0.876    -3.31e-07    2.82e-07
  numreplies |   8.22e-06   6.35e-06     1.29   0.196    -4.23e-06    .0000207
     profage |   .0791629   .0136328     5.81   0.000     .0524432    .1058826
      warmth |   -.013647   .0067644    -2.02   0.044     -.026905   -.0003891

TIexogenous  |
     attract |   .4691929   .0911456     5.15   0.000     .2905508    .6478349
         fit |   .4052177   .1109377     3.65   0.000     .1877839    .6226516
        gend |   .2135305    .089342     2.39   0.017     .0384233    .3886376
        ethn |   .0501947   .1003869     0.50   0.617      -.14656    .2469494
             |
       _cons |  -1.203777    .135699    -8.87   0.000    -1.469742   -.9378121
-------------+----------------------------------------------------------------
     sigma_u |  .89746199
     sigma_e |  .07564185
         rho |  .99294628   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

Second DV:

xthtaylor z_dv2 postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage, vce(robust) endog(postwc commentwc postsmo commentsmo numreplies profage warmth)  

Hausman–Taylor estimation                       Number of obs     =      2,916
Group variable: id                              Number of groups  =        486

                                                Obs per group:
                                                              min =          6
                                                              avg =          6
                                                              max =          6

Random effects u_i ~ i.i.d.                     Wald chi2(16)     =    2998.24
                                                Prob > chi2       =     0.0000

                                   (Std. err. adjusted for 486 clusters in id)
------------------------------------------------------------------------------
             |               Robust
       z_dv2 | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
TVexogenous  |
     vidpost |  -.0015115   .0015131    -1.00   0.318    -.0044772    .0014541
    postsdel |  -.0055198   .0034133    -1.62   0.106    -.0122099    .0011702
  CredBinary |  -.0255325   .0148094    -1.72   0.085    -.0545584    .0034935
   compimage |  -.0004394   .0169282    -0.03   0.979    -.0336181    .0327393
   warmimage |   .0193616   .0088239     2.19   0.028      .002067    .0366561

TVendogenous |
      postwc |  -.0000341   .0000127    -2.69   0.007     -.000059   -9.24e-06
   commentwc |   .0000231   1.87e-06    12.31   0.000     .0000194    .0000267
     postsmo |   .0034201   .0013424     2.55   0.011     .0007892    .0060511
  commentsmo |   -.000015   3.99e-06    -3.76   0.000    -.0000228   -7.19e-06
  numreplies |    .000199   .0000457     4.36   0.000     .0001095    .0002885
     profage |   .0377401   .0231116     1.63   0.102    -.0075578     .083038
      warmth |   .1212602   .0422509     2.87   0.004       .03845    .2040703

TIexogenous  |
     attract |   .0720937    .022387     3.22   0.001     .0282161    .1159714
         fit |   .0060661   .0350571     0.17   0.863    -.0626446    .0747767
        gend |   .1311991   .0290432     4.52   0.000     .0742754    .1881228
        ethn |   .0513238   .0454866     1.13   0.259    -.0378282    .1404758
             |
       _cons |  -.6804503   .1065218    -6.39   0.000    -.8892293   -.4716713
-------------+----------------------------------------------------------------
     sigma_u |  .29593527
     sigma_e |  .18349023
         rho |  .72231193   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

[Andrew Musau]

If you look at my code closely, I specify the -nocons- option and include my own constants (which1cons and which2cons). It is apparent that you haven't done this. You should start by replicating what I did so that you can understand the procedure. Otherwise, post a data example using dataex for any specific code suggestions (see FAQ Advice #12 for details).

[Ashley Roccapriore]

Hi Andrew Musau - apologies, I should have worded my response differently. I used the -noncons- and included the constants, and the results were even further off. The code I originally included in my response was just me saying "this is what I used to get as close as I could get." I tried to do it the way you mentioned in your code exactly (see below) and the results are much different, so I included the code that got me closest to my unstacked/uncombined results.

For example:
In the separate one, with the CredBinary variable to the first DV the coefficient is .0155909 and p-value of 0.156. For the second DV the coefficient is -.0255325 and p-value of 0.085.

In the combined one, the CredBinary variable to the first DV the coefficient is -.0194046 and p-value of 0.210. For the second DV the coefficient is -.0160924 and p-value of 0.116.

So you can see that not only are the coefficients different (even flipped for the first DV), but the p-values are too (as is the case specifically for the second DV).

To illustrate, here's me replicating your exact code:

xtset  id month
xthtaylor z_LNdv1 postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage, vce(robust) endog(postwc commentwc postsmo commentsmo numreplies profage warmth) constant(attract fit gend ethn)
xthtaylor z_dv2 postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage, vce(robust) endog(postwc commentwc postsmo commentsmo numreplies profage warmth) constant(attract fit gend ethn)
rename (z_LNdv1 z_dv2) (var1 var2)
reshape long var, i(id month) j(which)
qui sum id
replace id= id + `r(max)' if which==2
xtset id month
gen which1=1.which
gen which2=2.which
gen cons=1
foreach var in postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage cons{
    gen which1`var'=which1*`var'
    gen which2`var'=which2*`var'
}
xthtaylor var which1attract which2attract which1fit which2fit which1gend which2gend which1ethn which2ethn which1postwc which2postwc which1commentwc which2commentwc which1postsmo which2postsmo which1commentsmo which2commentsmo which1numreplies which2numreplies which1vidpost which2vidpost which1postsdel  which2postsdel which1profage which2profage which1CredBinary which2CredBinary which1compimage which2compimage which1warmth  which2warmth which1warmimage which2warmimage, nocons endog(which1postwc which2postwc which1commentwc which2commentwc which1postsmo which2postsmo which1commentsmo which2commentsmo which1numreplies which2numreplies which1warmth which2warmth) constant(which1attract which2attract which1fit which2fit which1gend which2gend which1ethn which2ethn) vce(robust)

And the results I get from it:

*............... DV 1 - Separate

. xthtaylor z_LNdv1 postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage, vce(robust) endo
> g(postwc commentwc postsmo commentsmo numreplies profage warmth) constant(attract fit gend ethn)

Hausman–Taylor estimation                       Number of obs     =      2,916
Group variable: id                              Number of groups  =        486

                                                Obs per group:
                                                              min =          6
                                                              avg =          6
                                                              max =          6

Random effects u_i ~ i.i.d.                     Wald chi2(16)     =     150.21
                                                Prob > chi2       =     0.0000

                                   (Std. err. adjusted for 486 clusters in id)
------------------------------------------------------------------------------
             |               Robust
     z_LNdv1 | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
TVexogenous  |
     vidpost |   .0002047   .0005317     0.39   0.700    -.0008373    .0012468
    postsdel |    .002598   .0027049     0.96   0.337    -.0027035    .0078995
  CredBinary |   .0155909   .0109847     1.42   0.156    -.0059386    .0371205
   compimage |   .0141383    .005277     2.68   0.007     .0037955     .024481
   warmimage |   .0036695   .0024819     1.48   0.139     -.001195     .008534
TVendogenous |
      postwc |   2.65e-06   3.14e-06     0.84   0.398    -3.50e-06    8.81e-06
   commentwc |   3.34e-08   7.37e-08     0.45   0.651    -1.11e-07    1.78e-07
     postsmo |  -.0005505   .0006189    -0.89   0.374    -.0017634    .0006625
  commentsmo |  -2.44e-08   1.56e-07    -0.16   0.876    -3.31e-07    2.82e-07
  numreplies |   8.22e-06   6.35e-06     1.29   0.196    -4.23e-06    .0000207
     profage |   .0791629   .0136328     5.81   0.000     .0524432    .1058826
      warmth |   -.013647   .0067644    -2.02   0.044     -.026905   -.0003891
TIexogenous  |
     attract |   .4691929   .0911456     5.15   0.000     .2905508    .6478349
         fit |   .4052177   .1109377     3.65   0.000     .1877839    .6226516
        gend |   .2135305    .089342     2.39   0.017     .0384233    .3886376
        ethn |   .0501947   .1003869     0.50   0.617      -.14656    .2469494
             |
       _cons |  -1.203777    .135699    -8.87   0.000    -1.469742   -.9378121
-------------+----------------------------------------------------------------
     sigma_u |  .89746199
     sigma_e |  .07564185
         rho |  .99294628   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.


*............... DV 2 - Separate
. xthtaylor z_dv2 postwc commentwc attract fit gend ethn postsmo commentsmo numreplies vidpost postsdel profage CredBinary warmth compimage warmimage, vce(robust) endog(
> postwc commentwc postsmo commentsmo numreplies profage warmth) constant(attract fit gend ethn)

Hausman–Taylor estimation                       Number of obs     =      2,916
Group variable: id                              Number of groups  =        486

                                                Obs per group:
                                                              min =          6
                                                              avg =          6
                                                              max =          6

Random effects u_i ~ i.i.d.                     Wald chi2(16)     =    2998.24
                                                Prob > chi2       =     0.0000

                                   (Std. err. adjusted for 486 clusters in id)
------------------------------------------------------------------------------
             |               Robust
       z_dv2 | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
TVexogenous  |
     vidpost |  -.0015115   .0015131    -1.00   0.318    -.0044772    .0014541
    postsdel |  -.0055198   .0034133    -1.62   0.106    -.0122099    .0011702
  CredBinary |  -.0255325   .0148094    -1.72   0.085    -.0545584    .0034935
   compimage |  -.0004394   .0169282    -0.03   0.979    -.0336181    .0327393
   warmimage |   .0193616   .0088239     2.19   0.028      .002067    .0366561
TVendogenous |
      postwc |  -.0000341   .0000127    -2.69   0.007     -.000059   -9.24e-06
   commentwc |   .0000231   1.87e-06    12.31   0.000     .0000194    .0000267
     postsmo |   .0034201   .0013424     2.55   0.011     .0007892    .0060511
  commentsmo |   -.000015   3.99e-06    -3.76   0.000    -.0000228   -7.19e-06
  numreplies |    .000199   .0000457     4.36   0.000     .0001095    .0002885
     profage |   .0377401   .0231116     1.63   0.102    -.0075578     .083038
      warmth |   .1212602   .0422509     2.87   0.004       .03845    .2040703
TIexogenous  |
     attract |   .0720937    .022387     3.22   0.001     .0282161    .1159714
         fit |   .0060661   .0350571     0.17   0.863    -.0626446    .0747767
        gend |   .1311991   .0290432     4.52   0.000     .0742754    .1881228
        ethn |   .0513238   .0454866     1.13   0.259    -.0378282    .1404758
             |
       _cons |  -.6804503   .1065218    -6.39   0.000    -.8892293   -.4716713
-------------+----------------------------------------------------------------
     sigma_u |  .29593527
     sigma_e |  .18349023
         rho |  .72231193   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

*............... Combined
. xthtaylor var which1attract which2attract which1fit which2fit which1gend which2gend which1ethn which2ethn which1postwc which2postwc which1commentwc which2commentwc whi
> ch1postsmo which2postsmo which1commentsmo which2commentsmo which1numreplies which2numreplies which1vidpost which2vidpost which1postsdel  which2postsdel which1profage w
> hich2profage which1CredBinary which2CredBinary which1compimage which2compimage which1warmth  which2warmth which1warmimage which2warmimage, nocons endog(which1postwc wh
> ich2postwc which1commentwc which2commentwc which1postsmo which2postsmo which1commentsmo which2commentsmo which1numreplies which2numreplies which1warmth which2warmth) c
> onstant(which1attract which2attract which1fit which2fit which1gend which2gend which1ethn which2ethn) vce(robust)

Hausman–Taylor estimation                       Number of obs     =      5,832
Group variable: id                              Number of groups  =        972

                                                Obs per group:
                                                              min =          6
                                                              avg =          6
                                                              max =          6

Random effects u_i ~ i.i.d.                     Wald chi2(32)     =    2093.76
                                                Prob > chi2       =     0.0000

                                   (Std. err. adjusted for 972 clusters in id)
------------------------------------------------------------------------------
             |               Robust
         var | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
TVexogenous  |
which1vidp~t |   .0001241   .0006218     0.20   0.842    -.0010946    .0013427
which2vidp~t |  -.0025269   .0015408    -1.64   0.101    -.0055467     .000493
which1post~l |    .001455   .0027985     0.52   0.603      -.00403    .0069399
which2post~l |  -.0061856   .0033919    -1.82   0.068    -.0128336    .0004623
which1prof~e |   .0541315   .0138564     3.91   0.000     .0269735    .0812896
which2prof~e |  -.0143468   .0098867    -1.45   0.147    -.0337244    .0050308
which1Cred~y |  -.0194046   .0154851    -1.25   0.210    -.0497548    .0109457
which2Cred~y |  -.0160924   .0102245    -1.57   0.116    -.0361319    .0039472
which1comp~e |   .0274278   .0057944     4.73   0.000      .016071    .0387847
which2comp~e |  -.0052879   .0168673    -0.31   0.754    -.0383472    .0277713
which1warm~e |   .0079132   .0028613     2.77   0.006     .0023052    .0135212
which2warm~e |   .0199645   .0085566     2.33   0.020     .0031938    .0367352
TVendogenous |
which1postwc |  -2.87e-06   3.46e-06    -0.83   0.406    -9.66e-06    3.91e-06
which2postwc |  -.0000355   .0000127    -2.80   0.005    -.0000604   -.0000107
which1comm~c |   6.69e-08   7.91e-08     0.85   0.397    -8.81e-08    2.22e-07
which2comm~c |    .000023   1.87e-06    12.31   0.000     .0000193    .0000267
which1post~o |  -.0006132   .0006489    -0.95   0.345     -.001885    .0006585
which2post~o |   .0029157   .0012705     2.29   0.022     .0004255     .005406
which1comm~o |  -9.43e-08   1.70e-07    -0.56   0.579    -4.27e-07    2.38e-07
which2comm~o |  -.0000149   3.99e-06    -3.75   0.000    -.0000227   -7.12e-06
which1numr~s |  -5.25e-07   7.30e-06    -0.07   0.943    -.0000148    .0000138
which2numr~s |   .0001939   .0000451     4.30   0.000     .0001056    .0002823
which1warmth |   .0029384    .007762     0.38   0.705    -.0122748    .0181515
which2warmth |   .1267008   .0424077     2.99   0.003     .0435832    .2098183
TIexogenous  |
which1attr~t |   .3238171   .0963861     3.36   0.001     .1349039    .5127303
which2attr~t |   .0225062   .0247671     0.91   0.364    -.0260364    .0710488
   which1fit |  -.2703932   .0979016    -2.76   0.006    -.4622769   -.0785096
   which2fit |  -.2344285   .0420243    -5.58   0.000    -.3167946   -.1520625
  which1gend |  -.1535059   .0910009    -1.69   0.092    -.3318643    .0248526
  which2gend |  -.0107818   .0308066    -0.35   0.726    -.0711615     .049598
  which1ethn |  -.1768501   .1094623    -1.62   0.106    -.3913922    .0376921
  which2ethn |  -.0492448   .0476972    -1.03   0.302    -.1427296      .04424
-------------+----------------------------------------------------------------
     sigma_u |  .67442888
     sigma_e |  .14033951
         rho |  .95849711   (fraction of variance due to u_i)
------------------------------------------------------------------------------
Note: TV refers to time varying; TI refers to time invariant.

.
end of do-file

.

[Andrew Musau]

It's difficult to tell why the coefficients are not the same. It may be an issue with your data. If you can post your data or send it to me by email (in case you cannot make it public), I'd be happy to have a look.