I am using Stata 17 MP (duly updated).
The models and the data
I am fitting negative binomial (NB2) 3-levels growth curve models of weekly counts, using meglm.
The nesting structure is : 30 occasions (weeks) < 1710 settlements < 96 districts. An extract (1 settlement 30 weeks) of the data is provided below.
A) Time-varying predictors :
at level-1
-weeknum (the ordinal number of a week in the series of 30)
-weektp (week type: four categories entered as 3 dummies)
-tclass (eight ordered categories entered as 7 dummies)
at level-3
-L1weekstock
-L1weekmob
B) Time-invariant predictors: all other.
Code:
* Example generated by -dataex-. For more info, type help dataex clear input double weeknumber float distrid long settlid float(countinf std2_weeknum std2_weeknumsqr weektp tclass std2_L1weekstock std2_L1weekmob std2_pcrural std2_pc65yover std2_pcforeign std2_medinc milp17) 1 74 1 0 -.8376152 .7015992 1 2 -.3326102 -.8563091 -.39327615 -.6413893 .4492839 .4750123 27.122 2 74 1 0 -.7798486 .6081639 1 1 -.3356403 -.1969749 -.39327615 -.6413893 .4492839 .4750123 27.122 3 74 1 0 -.7220821 .52140254 1 1 -.3348828 -.335782 -.39327615 -.6413893 .4492839 .4750123 27.122 4 74 1 0 -.6643155 .4413151 1 1 -.3375342 .05461322 -.39327615 -.6413893 .4492839 .4750123 27.122 5 74 1 0 -.606549 .36790165 1 1 -.3356403 .14136773 -.39327615 -.6413893 .4492839 .4750123 27.122 6 74 1 0 -.54878235 .3011621 1 2 -.3322314 .1587186 -.39327615 -.6413893 .4492839 .4750123 27.122 7 74 1 0 -.4910158 .24109654 2 3 -.3363979 .21944684 -.39327615 -.6413893 .4492839 .4750123 27.122 8 74 1 0 -.4332492 .1877049 2 2 -.334504 .05461322 -.39327615 -.6413893 .4492839 .4750123 27.122 9 74 1 0 -.3754827 .14098725 2 4 -.3299588 -.4138612 -.39327615 -.6413893 .4492839 .4750123 27.122 10 74 1 3 -.3177161 .10094354 2 4 -.29624802 .002560559 -.39327615 -.6413893 .4492839 .4750123 27.122 11 74 1 0 -.25994954 .06757376 2 3 -.2568557 -.24035217 -.39327615 -.6413893 .4492839 .4750123 27.119 12 74 1 6 -.202183 .04087796 2 3 -.26594624 -.6307475 -.39327615 -.6413893 .4492839 .4750123 27.119 13 74 1 6 -.1444164 .0208561 2 4 -.2716278 -.8042565 -.39327615 -.6413893 .4492839 .4750123 27.113 14 74 1 14 -.08664985 .007508196 2 5 -.24928026 -.604721 -.39327615 -.6413893 .4492839 .4750123 27.107 15 74 1 3 -.02888328 .000834244 3 4 -.2326143 -.17094846 -.39327615 -.6413893 .4492839 .4750123 27.093 16 74 1 6 .02888328 .000834244 3 4 -.2299629 .08931507 -.39327615 -.6413893 .4492839 .4750123 27.09 17 74 1 6 .08664985 .007508196 3 5 -.1913281 -.006114925 -.39327615 -.6413893 .4492839 .4750123 27.084 18 74 1 14 .1444164 .0208561 3 4 -.19890356 .3669295 -.39327615 -.6413893 .4492839 .4750123 27.078 19 74 1 14 .202183 .04087796 3 5 -.1966309 .4450085 -.39327615 -.6413893 .4492839 .4750123 27.064 20 74 1 14 .25994954 .06757376 3 5 -.17579845 .59249115 -.39327615 -.6413893 .4492839 .4750123 27.05 21 74 1 41 .3177161 .10094354 3 6 -.02088059 .6185175 -.39327615 -.6413893 .4492839 .4750123 27.036 22 74 1 136 .3754827 .14098725 3 8 .4128136 .59249115 -.39327615 -.6413893 .4492839 .4750123 26.995 23 74 1 272 .4332492 .1877049 3 8 1.637763 -.032141294 -.39327615 -.6413893 .4492839 .4750123 26.859 24 74 1 272 .4910158 .24109654 4 8 3.026342 -.1969749 -.39327615 -.6413893 .4492839 .4750123 26.587 25 74 1 136 .54878235 .3011621 4 8 3.214592 -.01479041 -.39327615 -.6413893 .4492839 .4750123 26.315 26 74 1 136 .606549 .36790165 4 8 1.8313158 -.3878348 -.39327615 -.6413893 .4492839 .4750123 26.179 27 74 1 68 .6643155 .4413151 4 6 1.193842 .01123596 -.39327615 -.6413893 .4492839 .4750123 26.043 28 74 1 68 .7220821 .52140254 4 6 .54235375 .09799048 -.39327615 -.6413893 .4492839 .4750123 25.975 29 74 1 68 .7798486 .6081639 4 6 .3870571 .2367977 -.39327615 -.6413893 .4492839 .4750123 25.907 30 74 1 41 .8376152 .7015992 4 8 .26585 .28017497 -.39327615 -.6413893 .4492839 .4750123 25.839 end
(Gelman A 2008 “Scaling regression inputs by dividing by two standard deviations” Statistics in Medicine).
The problem:
-A) when I use the # operator to create quadratic terms (weeknum##weeknum or weeknum#weeknum), the final iteration returns the error message shown below.
Code:
meglm countinf c.std2_weeknum##c.std2_weeknum ib2.weektp ib2.weektp#c.std2_weeknum ib2.weektp#c.std2_weeknum#c.std2_weeknum /// ib1.tclass std2_pcrural ib1.tclas s#c.std2_pcrural std2_L1weekstock std2_L1weekmob std2_pc65yover std2_pcforeign std2_medinc /// , exposure(milp17) || distrid : || settlid: c.std2_weeknum##c.std2_weeknum /// , difficult family(nbinomial mean) link(log) cov(unstructured) vce(robust)
Code:
meglm countinf std2_weeknum c.std2_weeknum#c.std2_weeknum ib2.weektp ib2.weektp#c.std2_weeknum ib2.weektp#c.std2_weeknum#c.std2_weeknum /// ib1.tclass std2_pcrural ib1.tclass#c.std2_pcrural std2_L1weekstock std2_L1weekmob std2_pc65yover std2_pcforeign std2_medinc /// , exposure(milp17) || distrid : || settlid: std2_weeknum c.std2_weeknum#c.std2_weeknum /// , difficult family(nbinomial mean) link(log) cov(unstructured) vce(robust)
Code:
Iteration 11: log pseudolikelihood = -131293.52 (not concave)
Iteration 12: log pseudolikelihood = -131285.01 (not concave)
Iteration 13: log pseudolikelihood = -131277.98
Iteration 14: log pseudolikelihood = -131272.86
Iteration 15: log pseudolikelihood = -131271.5
Iteration 16: log pseudolikelihood = -131270.98
Iteration 17: log pseudolikelihood = -131270.98
*: 3200 conformability error
_gsem_ereturn__sd(): - function returned error
_gsem_ereturn(): - function returned error
st_gsem_ereturn(): - function returned error
<istmt>: - function returned error
r(3200);
end of do-file
r(3200);- B) when I do not use #, but instead create X^2 (weeknumsqr), the estimation procedure ends normally,
Code:
meglm countinf std2_weeknum std2_weeknumsqr ib2.weektp ib2.weektp#c.std2_weeknum ib2.weektp#c.std2_weeknumsqr /// ib1.tclass std2_pcrural ib1.tclass#c.std2_pcrural std2_L1weekstock std2_L1weekmob std2_pc65yover std2_pcforeign std2_medinc /// , exposure(milp17) || distrid : || settlid: std2_weeknum std2_weeknumsqr /// , difficult family(nbinomial mean) link(log) cov(unstructured) vce(robust)
Code:
Iteration 11: log pseudolikelihood = -131293.52 (not concave)
Iteration 12: log pseudolikelihood = -131285.01 (not concave)
Iteration 13: log pseudolikelihood = -131277.98
Iteration 14: log pseudolikelihood = -131272.86
Iteration 15: log pseudolikelihood = -131271.5
Iteration 16: log pseudolikelihood = -131270.98
Iteration 17: log pseudolikelihood = -131270.98
Mixed-effects GLM Number of obs = 51,300
Family: Negative binomial
Link: Log
Overdispersion: mean
Grouping information
-------------------------------------------------------------
| No. of Observations per group
Group variable | groups Minimum Average Maximum
----------------+--------------------------------------------
distrid | 96 90 534.4 2,670
settlid | 1,710 30 30.0 30
-------------------------------------------------------------
Integration method: mvaghermite Integration pts. = 7
Wald chi2(31) = 29843.31
Log pseudolikelihood = -131270.98 Prob > chi2 = 0.0000
(Std. err. adjusted for 96 clusters in distrid)
---------------------------------------------------------------------------------------------------
| Robust
countinf | Coefficient std. err. z P>|z| [95% conf. interval]
----------------------------------+----------------------------------------------------------------
std2_weeknum | 3.298914 .7736116 4.26 0.000 1.782663 4.815165
std2_weeknumsqr | -5.317793 1.640426 -3.24 0.001 -8.532969 -2.102617
|
weektp |
1 | -3.454767 2.946621 -1.17 0.241 -9.230038 2.320504
3 | -.3165163 .0706098 -4.48 0.000 -.454909 -.1781236
4 | 4.843987 .4872818 9.94 0.000 3.888932 5.799042
|
weektp#c.std2_weeknum |
1 | -3.15082 8.370738 -0.38 0.707 -19.55716 13.25553
3 | -2.463644 .7949061 -3.10 0.002 -4.021631 -.9056565
4 | -14.65873 1.526555 -9.60 0.000 -17.65073 -11.66674
|
weektp#c.std2_weeknumsqr |
1 | 7.216591 5.968627 1.21 0.227 -4.481703 18.91488
3 | 10.91105 2.087026 5.23 0.000 6.820551 15.00154
4 | 11.94664 1.911539 6.25 0.000 8.200094 15.69319
|
tclass |
2 | 1.579361 .1812937 8.71 0.000 1.224032 1.93469
3 | 2.36908 .1987663 11.92 0.000 1.979505 2.758655
4 | 2.763744 .1977046 13.98 0.000 2.37625 3.151238
5 | 3.076674 .2004631 15.35 0.000 2.683773 3.469574
6 | 3.453519 .2009128 17.19 0.000 3.059737 3.847301
7 | 3.750355 .2029003 18.48 0.000 3.352678 4.148032
8 | 4.095897 .2064415 19.84 0.000 3.691279 4.500515
|
std2_pcrural | .624201 .2692187 2.32 0.020 .096542 1.15186
|
tclass#c.std2_pcrural |
2 | -.7134393 .2503491 -2.85 0.004 -1.204115 -.222764
3 | -.7459787 .2489973 -3.00 0.003 -1.234005 -.2579529
4 | -.8330824 .251747 -3.31 0.001 -1.326498 -.3396672
5 | -.7421679 .2588821 -2.87 0.004 -1.249568 -.2347682
6 | -.6488989 .2634231 -2.46 0.014 -1.165199 -.1325992
7 | -.5745428 .2704703 -2.12 0.034 -1.104655 -.0444308
8 | -.5618221 .28133 -2.00 0.046 -1.113219 -.0104255
|
std2_L1weekstock | .2227779 .0439275 5.07 0.000 .1366815 .3088743
std2_L1weekmob | .1751561 .0303288 5.78 0.000 .1157127 .2345994
std2_pc65yover | -.0672125 .0224725 -2.99 0.003 -.1112578 -.0231672
std2_pcforeign | .2212634 .0261615 8.46 0.000 .1699877 .272539
std2_medinc | .0396633 .0216403 1.83 0.067 -.0027508 .0820775
_cons | -4.043469 .2109965 -19.16 0.000 -4.457014 -3.629923
ln(milp17) | 1 (exposure)
----------------------------------+----------------------------------------------------------------
/lnalpha | -1.695915 .0367845 -1.768011 -1.623818
----------------------------------+----------------------------------------------------------------
distrid |
var(_cons)| .0587462 .0118504 .0395616 .0872341
----------------------------------+----------------------------------------------------------------
distrid>settlid |
var(std2_weeknum)| 1.195491 .1552879 .9267871 1.542101
var(std2_weeknumsqr)| 2.423071 .2574262 1.967591 2.983991
var(_cons)| .1038953 .0107819 .0847737 .1273299
----------------------------------+----------------------------------------------------------------
distrid>settlid |
cov(std2_weeknum,std2_weeknumsqr)| -1.232618 .1880016 -6.56 0.000 -1.601095 -.8641421
cov(std2_weeknum,_cons)| -.1914082 .0298032 -6.42 0.000 -.2498215 -.132995
cov(std2_weeknumsqr,_cons)| -.0505272 .0422252 -1.20 0.231 -.1332871 .0322327
---------------------------------------------------------------------------------------------------Specifying simpler models or more complex ones does not help.
I would like to use the # operator and then "margins" extentively.
What mistakes do I make?
0 Response to Error (3200) in meglm, when using X##X, but no error with X and X_squared
Post a Comment