missing value problem with gsvy and gologit2?

[Jocelyn Li]

Hello, my first post here although I have been reading the forum for a long time.

I tried to use gologit2 for my svy data but the error message is puzzling. I am positive that I have no missing values in the data because I dropped those.
Actually, I am not even sure missing data is the problem here. I am using the most recent version of gologit2.

. nmissing  bday_3 age rengrp educ_4 marst_3 if sample1==1 & female==0
. svyset psu [pweight=newperwt], strata(strata)
. gsvy, subpop(if sample1==1 & female==0): gologit2 bday_3 age i.rengrp i.educ_4 i.marst_3, auto nolabel
matrix has missing values
an error occurred when svy executed gologit2

Any insight is appreciated! Thanks!

[Jocelyn Li]

I tried to run gologit2 without gsvy and now this data has missing values for the dependent variable. However, this time gologit2 worked. So I guess the problem is the svy part?

. gologit2 bday_3 age i.rengrp i.educ_4 i.marst_3 if sample1==1 & female==0, auto nolabel

------------------------------------------------------------------------------
Testing parallel lines assumption using the .05 level of significance...

Step  1:  Constraints for parallel lines imposed for 7.rengrp (P Value = 0.5841)
Step  2:  Constraints for parallel lines imposed for 8.rengrp (P Value = 0.0742)
Step  3:  Constraints for parallel lines are not imposed for 
          age (P Value = 0.00000)
          3.rengrp (P Value = 0.00000)
          1.educ_4 (P Value = 0.00000)
          2.educ_4 (P Value = 0.00000)
          3.educ_4 (P Value = 0.00000)
          2.marst_3 (P Value = 0.00000)
          3.marst_3 (P Value = 0.00000)

Wald test of parallel lines assumption for the final model:

 ( 1)  [eq1]7.rengrp - [eq2]7.rengrp = 0
 ( 2)  [eq1]8.rengrp - [eq2]8.rengrp = 0

           chi2(  2) =    3.49
         Prob > chi2 =    0.1746

An insignificant test statistic indicates that the final model
does not violate the proportional odds/ parallel lines assumption

If you re-estimate this exact same model with gologit2, instead 
of autofit you can save time by using the parameter

pl(1b.rengrp 2o.rengrp 4o.rengrp 5o.rengrp 6o.rengrp 7.rengrp 8.rengrp 0b.educ_4 4o.educ_4 1b.marst_3)

------------------------------------------------------------------------------

Generalized Ordered Logit Estimates             Number of obs     =     88,262
                                                LR chi2(16)       =    3481.83
                                                Prob > chi2       =     0.0000
Log likelihood = -69434.378                     Pseudo R2         =     0.0245

 ( 1)  [eq1]7.rengrp - [eq2]7.rengrp = 0
 ( 2)  [eq1]8.rengrp - [eq2]8.rengrp = 0
------------------------------------------------------------------------------------------------
                        bday_3 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------------------------+----------------------------------------------------------------
eq1                            |
                           age |  -.0041027   .0006267    -6.55   0.000    -.0053311   -.0028744
                               |
                        rengrp |
        non-Hispanic NB black  |  -.1836288   .0222085    -8.27   0.000    -.2271567   -.1401008
         native-born Hispanic  |  -.0358244   .0262893    -1.36   0.173    -.0873505    .0157017
        foreign-born Hispanic  |  -.8225086   .0274713   -29.94   0.000    -.8763513   -.7686659
                               |
                        educ_4 |
                          <hs  |  -.0974978   .0257083    -3.79   0.000    -.1478851   -.0471104
                           hs  |  -.1447337   .0196244    -7.38   0.000    -.1831969   -.1062704
some college/associate degree  |   .0032983   .0189259     0.17   0.862    -.0337958    .0403925
                 BA and above  |          0  (omitted)
                               |
                       marst_3 |
   separated/divorced/widowed  |   .1463106   .0200851     7.28   0.000     .1069446    .1856766
                never married  |     .01379   .0180391     0.76   0.445    -.0215659    .0491459
                               |
                         _cons |  -.3780634   .0323696   -11.68   0.000    -.4415066   -.3146201
-------------------------------+----------------------------------------------------------------
eq2                            |
                           age |   .0265874   .0011619    22.88   0.000     .0243101    .0288647
                               |
                        rengrp |
        non-Hispanic NB black  |   .0952442   .0354787     2.68   0.007     .0257073    .1647811
         native-born Hispanic  |  -.0358244   .0262893    -1.36   0.173    -.0873505    .0157017
        foreign-born Hispanic  |  -.8225086   .0274713   -29.94   0.000    -.8763513   -.7686659
                               |
                        educ_4 |
                          <hs  |   .9329333   .0457874    20.38   0.000     .8431916    1.022675
                           hs  |    .678657   .0404978    16.76   0.000     .5992828    .7580313
some college/associate degree  |   .6199999   .0402559    15.40   0.000     .5410998       .6989
                 BA and above  |          0  (omitted)
                               |
                       marst_3 |
   separated/divorced/widowed  |   .4139872   .0325959    12.70   0.000     .3501003     .477874
                never married  |   .2074178   .0342287     6.06   0.000     .1403308    .2745047
                               |
                         _cons |  -4.415686   .0651466   -67.78   0.000    -4.543371   -4.288001
------------------------------------------------------------------------------------------------

. nmissing bday_3 age rengrp educ_4 marst_3 if sample1==1 & female==0

bday_3            586

[Jocelyn Li]

I did more exploration and now I am really puzzled. I used the example data and code that includes gsvy from the gologit2 help file. However, it didn't work either. I got the same error message! What could be going wrong here?

The first two lines is to show that I have the most current version of gologit2. Is it possible that even this most current version of gologit2 is not compatible with my Stata 15?

. ssc install gologit2, replace
checking gologit2 consistency and verifying not already installed...
all files already exist and are up to date.

. update all
(contacting http://www.stata.com)

Update status
    Last check for updates:  07 Dec 2017
    New update available:    none         (as of 07 Dec 2017)
    Current update level:    21 Nov 2017  (what's new)

Possible actions

    Do nothing; all files are up to date.

. webuse nhanes2f, clear

. gsvy, subpop(female): gologit2 health i.black age c.age#c.age, autofit
matrix has missing values
an error occurred when svy executed gologit2
r(504);

[Richard Williams]

I agree that this is weird.

I just ran

webuse nhanes2f, clear
gsvy, subpop(female): gologit2 health i.black age c.age#c.age, autofit

in Stata 14.2, and it works fine. So something about Stata 15 is zapping gologit2. Further, version control is not helping.

In the short run, if you have an older version of Stata, try using it. I will try to see what is wrong with 15.

[Richard Williams]

I think I have found a temporary fix and have emailed it to Jocelyn. If anybody else desperately needs it before there is a permanent fix, you can email me.

FYI, the problem seems to be with lincom.ado. The 15.1 version differs a little from the 14.2 version, and those changes zap gologit2. Version control does not help. I don't know if any other program in the world is affected, but mine is. gologit2's calls to lincom are usually not necessary so the temporary fix disables them.

I have written Stata about it. Hopefully they will tweak lincom.ado. But if not I think I can work out a fix so that lincom.ado never needs to be called.

[Jocelyn Li]

Thank you so much, Richard!

I used the fix and reran some of the programs. The code works for the nhanes2f dataset, but this is what I got for my data.

. gsvy, subpop(if sample1==1 & female==0): gologit2 bday_3 age i.rengrp i.educ_4 i.marst_3, auto nolabel

------------------------------------------------------------------------------
Testing parallel lines assumption using the .05 level of significance...

Step  1:  Constraints for parallel lines imposed for 7.rengrp (P Value = 0.7697)
Step  2:  Constraints for parallel lines imposed for 8.rengrp (P Value = 0.3610)
Step  3:  Constraints for parallel lines are not imposed for
          age (P Value = 0.00000)
          3.rengrp (P Value = 0.00000)
          1.educ_4 (P Value = 0.00000)
          2.educ_4 (P Value = 0.00000)
          3.educ_4 (P Value = 0.00000)
          2.marst_3 (P Value = 0.00000)
          3.marst_3 (P Value = 0.00000)

Wald test of parallel lines assumption for the final model:

Adjusted Wald test

 ( 1)  [eq1]7.rengrp - [eq2]7.rengrp = 0
 ( 2)  [eq1]8.rengrp - [eq2]8.rengrp = 0

       F(  2,   299) =    0.45
            Prob > F =    0.6409

An insignificant test statistic indicates that the final model
does not violate the proportional odds/ parallel lines assumption

If you re-estimate this exact same model with gologit2, instead
of autofit you can save time by using the parameter

pl(1b.rengrp 2o.rengrp 4o.rengrp 5o.rengrp 6o.rengrp 7.rengrp 8.rengrp 0b.educ_4 4o.educ_4 1b.marst_3)

------------------------------------------------------------------------------

Survey: Generalized Ordered Logit Estimates

Number of strata   =       300                Number of obs     =      933,920
Number of PSUs     =       600                Population size   =  304,728,355
                                              Subpop. no. obs   =       88,262
                                              Subpop. size      = 31,553,033.6
                                              Design df         =          300
                                              F(  16,    285)   =       148.80
                                              Prob > F          =       0.0000

 ( 1)  [eq1]7.rengrp - [eq2]7.rengrp = 0
 ( 2)  [eq1]8.rengrp - [eq2]8.rengrp = 0
------------------------------------------------------------------------------------------------
                               |             Linearized
                        bday_3 |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------------------------+----------------------------------------------------------------
eq1                            |
                           age |  -.0050333   .0007596    -6.63   0.000     -.006528   -.0035385
                               |
                        rengrp |
        non-Hispanic NB black  |  -.2088175   .0266952    -7.82   0.000     -.261351    -.156284
         native-born Hispanic  |  -.0386773   .0317493    -1.22   0.224    -.1011569    .0238022
        foreign-born Hispanic  |  -.7992292   .0333405   -23.97   0.000      -.86484   -.7336184
                               |
                        educ_4 |
                          <hs  |  -.0671374   .0321279    -2.09   0.037    -.1303619   -.0039128
                           hs  |  -.1561431   .0247395    -6.31   0.000    -.2048281   -.1074582
some college/associate degree  |   .0156297   .0238495     0.66   0.513    -.0313038    .0625631
                 BA and above  |          0  (omitted)
                               |
                       marst_3 |
   separated/divorced/widowed  |   .1254352   .0227145     5.52   0.000     .0807352    .1701351
                never married  |   .0022532   .0228905     0.10   0.922    -.0427932    .0472995
                               |
                         _cons |  -.3463112    .040932    -8.46   0.000    -.4268615   -.2657609
-------------------------------+----------------------------------------------------------------
eq2                            |
                           age |   .0259341   .0013227    19.61   0.000     .0233311     .028537
                               |
                        rengrp |
        non-Hispanic NB black  |   .0625554   .0388302     1.61   0.108    -.0138586    .1389694
         native-born Hispanic  |  -.0386773   .0317493    -1.22   0.224    -.1011569    .0238022
        foreign-born Hispanic  |  -.7992292   .0333405   -23.97   0.000      -.86484   -.7336184
                               |
                        educ_4 |
                          <hs  |   .9907265   .0567714    17.45   0.000     .8790058    1.102447
                           hs  |   .6540072   .0476726    13.72   0.000     .5601922    .7478223
some college/associate degree  |   .6319154    .045443    13.91   0.000      .542488    .7213429
                 BA and above  |          0  (omitted)
                               |
                       marst_3 |
   separated/divorced/widowed  |   .4076962   .0368969    11.05   0.000     .3350867    .4803058
                never married  |   .2129917   .0413782     5.15   0.000     .1315634    .2944199
                               |
                         _cons |  -4.399057   .0778894   -56.48   0.000    -4.552336   -4.245779
------------------------------------------------------------------------------------------------
if not found
r(111);

end of do-file

r(111);

I guess the program also worked for my data because I got results and they seem to make sense. However, the error message a the bottom concerns me. Is it a problem or I can safely ignore it?

[Richard Williams]

It doesn't like the way you are specifying subpop. Instead of using if, try something like

gen mysample = sample1==1 & female==0

gsvy, subpop(mysample): gologit2

Make sure that the mysample var is computed correctly. I'll have to see what is causing this. But at least there is an easy workaround (I think).

[Jocelyn Li]

Awesome! Problem solved! Thanks so much Richard!

[Jocelyn Li]

OK, more questions.

I re-installed Stata 14 on my computer and used it to run my code again. The good news that the results are identical with what I got from using 15. However, this time I would like to have the gamma option, but it doesn't work, even with 14. It seems that gamma doesn't do anything at all when gsvy is used. Is this something that you already knew, Richard?

. gsvy, subpop(msample): gologit2 bday_3 $dems, auto nolabel gamma

------------------------------------------------------------------------------
Testing parallel lines assumption using the .05 level of significance...

Step  1:  Constraints for parallel lines imposed for 7.rengrp (P Value = 0.7697)
Step  2:  Constraints for parallel lines imposed for 8.rengrp (P Value = 0.3610)
Step  3:  Constraints for parallel lines are not imposed for 
          age (P Value = 0.00000)
          3.rengrp (P Value = 0.00000)
          1.educ_4 (P Value = 0.00000)
          2.educ_4 (P Value = 0.00000)
          3.educ_4 (P Value = 0.00000)
          2.marst_3 (P Value = 0.00000)
          3.marst_3 (P Value = 0.00000)

Wald test of parallel lines assumption for the final model:

Adjusted Wald test

 ( 1)  [eq1]7.rengrp - [eq2]7.rengrp = 0
 ( 2)  [eq1]8.rengrp - [eq2]8.rengrp = 0

       F(  2,   299) =    0.45
            Prob > F =    0.6409

An insignificant test statistic indicates that the final model
does not violate the proportional odds/ parallel lines assumption

If you re-estimate this exact same model with gologit2, instead 
of autofit you can save time by using the parameter

pl(1b.rengrp 2o.rengrp 4o.rengrp 5o.rengrp 6o.rengrp 7.rengrp 8.rengrp 0b.educ_4 4o.educ_4 1b.marst_3)

------------------------------------------------------------------------------

Survey: Generalized Ordered Logit Estimates

Number of strata   =       300                Number of obs     =      933,920
Number of PSUs     =       600                Population size   =  304,728,355
                                              Subpop. no. obs   =       88,262
                                              Subpop. size      = 31,553,033.6
                                              Design df         =          300
                                              F(  16,    285)   =       148.80
                                              Prob > F          =       0.0000

 ( 1)  [eq1]7.rengrp - [eq2]7.rengrp = 0
 ( 2)  [eq1]8.rengrp - [eq2]8.rengrp = 0
------------------------------------------------------------------------------------------------
                               |             Linearized
                        bday_3 |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------------------------+----------------------------------------------------------------
eq1                            |
                           age |  -.0050333   .0007596    -6.63   0.000     -.006528   -.0035385
                               |
                        rengrp |
        non-Hispanic NB white  |          0  (base)
        non-Hispanic NB black  |  -.2088175   .0266952    -7.82   0.000     -.261351    -.156284
         native-born Hispanic  |  -.0386773   .0317493    -1.22   0.224    -.1011569    .0238022
        foreign-born Hispanic  |  -.7992292   .0333405   -23.97   0.000      -.86484   -.7336184
                               |
                        educ_4 |
                          <hs  |  -.0671374   .0321279    -2.09   0.037    -.1303619   -.0039128
                           hs  |  -.1561431   .0247395    -6.31   0.000    -.2048281   -.1074582
some college/associate degree  |   .0156297   .0238495     0.66   0.513    -.0313038    .0625631
                 BA and above  |          0  (omitted)
                               |
                       marst_3 |
  married/living with partner  |          0  (base)
   separated/divorced/widowed  |   .1254352   .0227145     5.52   0.000     .0807352    .1701351
                never married  |   .0022532   .0228905     0.10   0.922    -.0427932    .0472995
                               |
                         _cons |  -.3463112    .040932    -8.46   0.000    -.4268615   -.2657609
-------------------------------+----------------------------------------------------------------
eq2                            |
                           age |   .0259341   .0013227    19.61   0.000     .0233311     .028537
                               |
                        rengrp |
        non-Hispanic NB white  |          0  (base)
        non-Hispanic NB black  |   .0625554   .0388302     1.61   0.108    -.0138586    .1389694
         native-born Hispanic  |  -.0386773   .0317493    -1.22   0.224    -.1011569    .0238022
        foreign-born Hispanic  |  -.7992292   .0333405   -23.97   0.000      -.86484   -.7336184
                               |
                        educ_4 |
                          <hs  |   .9907265   .0567714    17.45   0.000     .8790058    1.102447
                           hs  |   .6540072   .0476726    13.72   0.000     .5601922    .7478223
some college/associate degree  |   .6319154    .045443    13.91   0.000      .542488    .7213429
                 BA and above  |          0  (omitted)
                               |
                       marst_3 |
  married/living with partner  |          0  (base)
   separated/divorced/widowed  |   .4076962   .0368969    11.05   0.000     .3350867    .4803058
                never married  |   .2129917   .0413782     5.15   0.000     .1315634    .2944199
                               |
                         _cons |  -4.399057   .0778894   -56.48   0.000    -4.552336   -4.245779
------------------------------------------------------------------------------------------------

The following code doesn't work either with Stata 14.

. webuse nhanes2f, clear

. gsvy, subpop(female): gologit2 health i.black age c.age#c.age, autofit gamma

[Richard Williams]

You don't show the error. If you use the official gologit2 with Stata 14.2 it works fine. If you use the tweaked temporary fix I sent you get an error, so I guess my temporary fix won't be the permanent one.

So, best advice for now: use 14,2 until I get a better fix. If somebody can only use gologit2 with Stata 15, and is getting the types of errors you originally got, I can email them a temporary fix that works so long as they don't use the gamma option. The gamma option uses lincom and lincom in Stata 15 causes gologit2 to have problems, so the temporary fix disables gamma, which most people don't use anyway.

Since you re-installed 14, run -update all- to make sure it is up to date.

[Richard Williams]

Thanks to Kit Baum, a new version of gologit2 is now on SSC. I think it fixes most of the problems noted in this thread.

[Tiarnan Kearney]

Hi Richard,

I appear to be having similar issues with the above. I am using stata 15.1 and version 3.1.2 16dec20167 of gologit2.

which gologit2
c:\ado\plus\g\gologit2.ado
*! version 3.1.2 16dec20167 Richard Williams, [email protected]

I have set my survey data with a pweight only.

svyset [pweight=WS]

      pweight: WS
          VCE: linearized
  Single unit: missing
     Strata 1: <one>
         SU 1: <observations>
        FPC 1: <zero>

I then run the model

svy:gologit2 ocastydr age

The output is:

(running gologit2 on estimation sample)
- invalid name
an error occurred while attempting to compute scores
r(322);

The model ran fine without the prefix, even if the weights are included. I have attempted both the gsvy and svy prefix. The model runs for the same data using ologit.

I have dropped the missing data with the same error. The example data set ran without a problem.

Any help would be greatly appreciated.

Kind regards,

Tiarnán

[Richard Williams]

Try adding the -nolabel- option. It sometimes fixes weird problems.

If problems persist, if you are free to share the data and code with me, I will take a look.

[Tiarnan Kearney]

Thank you very much, that seems to of worked.

Is there a problem in my label so, a character that is incompatible, even though it works without the prefix.

Many thanks again

[Richard Williams]

You don't show the labels so I don't know what the problem is. The gologit2 help says

nolabel causes the equations to be named eq1, eq2, etc. The default is to use the first 32 characters of the value labels and/or the values of Y as the equation labels. Note that some characters cannot be used in equation names, e.g. the space ( ), the period (.), the dollar sign ($), and the colon(, and will be replaced with the underscore (_) character. Square brackets ([]) and parentheses will be replaced with curly brackets ({}). The default behavior works well when the value labels are short and descriptive. It may not work well when value labels are very long and/or include characters that have to be changed. If the printout looks unattractive and/or you are getting strange errors, try changing the value labels of Y or else use the nolabel option.