Error message - Missing standard error because of stratum with single sampling unit

[Ana Vasconcelos]

Hello,

I´m using a longitudinal dataset (example below) and I´m trying to run the following command: svy: reg i_scghqa i_aidhh and it returns the following error message: "Missing standard error because of stratum with single sampling unit".
Can you help me to solve this issue? Thank you in advance.

* Example generated by -dataex-. For more info, type help dataex
clear
input long(pidp pid i_hidp) byte(i_pno i_childpno) long i_psu int i_strata byte(i_aidhh i_scghqa)
   76165  10689869 141657616 1 -8   19    6 2 2
  732365  15752658 619371216 1 -8  157   43 1 4
 1587125  17870879 618269616 1 -8  215   65 1 2
 4849085 176725733 347554816 1 -8  560  148 2 4
68010887        -8  68054416 1 -8 2012 2006 2 2
68029931        -8  68136016 2 -8 2060 2030 2 2
68035367        -8  68156416 1 -8 2060 2030 2 2
68041487        -8  68176816 1 -8 2084 2042 2 3
68041491        -8  68176816 2 -8 2084 2042 2 1
68045567        -8  68210816 1 -8 2084 2042 2 1
68051007        -8  68251616 1 -8 2108 2054 2 1
68051011        -8  68251616 2 -8 2108 2054 2 3
68058487        -8  68272016 1 -8 2108 2054 2 2
68058491        -8  68272016 2 -8 2108 2054 2 3
68060531        -8  68285616 2 -8 2108 2054 2 2
68060533 160066204  73059216 2 -8   18    6 1 3
68060537 160066239  73059216 3 -8   18    6 2 2
68063247        -8  68292416 1 -8 2132 2066 2 2
68063927        -8  68299216 1 -8 2132 2066 2 2
68063931        -8  68299216 2 -8 2132 2066 2 2
68064605  10653872  73066016 1 -8   18    6 2 2
68064609  10653902  73066016 2 -8   18    6 2 2
68068007        -8  68326416 1 -8 2132 2066 2 2
68068082        -8  68054416 2 -8 2012 2006 2 2
68125127        -8  68516816 1 -8 2252 2126 2 2
68125131        -8  68516816 2 -8 2252 2126 2 2
68125135        -8  68516816 3 -8 2252 2126 2 2
68133289  11218282  73093216 1 -8   34   11 2 3
68136009  11234989  73106816 1 -8   34   11 2 2
68138045  11242787  73120416 1 -8   34   11 2 2
68138049  11242817  73120416 2 -8   34   11 1 2
68138051        -8  68550816 2 -8 2276 2138 2 3
68144847        -8  68584816 1 -8 2276 2138 2 2
68144851        -8  68584816 2 -8 2276 2138 2 2
68148247        -8  68591616 1 -8 2300 2150 2 2
68150971        -8  68598416 2 -8 2300 2150 2 2
68150975        -8  68598416 3 -8 2300 2150 2 2
68155047        -8  68618816 1 -8 2300 2150 2 2
68155051        -8  68618816 2 -8 2300 2150 2 2
68157771        -8  68666416 2 -8 2300 2150 2 4
68159131        -8  68673216 2 -8 2300 2150 1 2
68160485  11418567  73127216 1 -8   39   11 2 3
68173407        -8  68707216 1 -8 2348 2174 2 2
68180887        -8  68754816 1  4 2348 2174 2 3
68184971        -8  68761616 2 -8 2372 2186 2 2
68185647        -8  68768416 1 -8 2372 2186 2 2
68187687        -8  68775216 1 -8 2372 2186 2 2
68187691        -8  68775216 2 -8 2372 2186 2 2
68191771        -8  68809216 1 -8 2372 2186 2 2
68193127        -8  68816016 1 -8 2372 2186 2 2
68195167        -8  68822816 1 -8 2372 2186 2 2
68195171        -8  68822816 2 -8 2372 2186 2 2
68195851        -8  68829616 2 -8 2372 2186 2 2
68197887        -8  68843216 1 -8 2396 2198 2 2
68197899        -8  68843216 3 -8 2396 2198 2 2
68197903        -8  68843216 4 -8 2396 2198 2 2
68199247        -8  76506816 1 -8 2396 2198 2 2
68207407        -8  68863616 1 -8 2396 2198 2 2
68207411        -8  68863616 2 -8 2396 2198 2 2
68216247        -8  68904416 1 -8 2420 2210 2 2
68218287        -8  68911216 1 -8 2420 2210 2 2
68231223        -8  68986016 2 -8 2444 2222 2 2
68238011        -8  69006416 2 -8 2468 2234 2 3
68262487        -8  69081216 1 -8 2516 2258 2 2
68288327        -8  69162816 1 -8 2564 2282 2 3
68288331        -8  69162816 2 -8 2564 2282 2 2
68293087        -8  69176416 1 -8 2564 2282 2 2
68293091        -8  69176416 2 -8 2564 2282 2 2
68293095        -8  69176416 3 -8 2564 2282 2 2
68293099        -8  69176416 4 -8 2564 2282 2 2
68293168        -8  68278816 2 -8 2108 2054 2 2
68294447        -8  69183216 1 -8 2588 2294 2 2
68294451        -8  69183216 2 -8 2588 2294 2 2
68297845  12521361  73154416 1 -8   69   15 2 2
68297849  12521396  73154416 2 -8   69   15 2 2
68297857  12521469  73161216 1 -8   69   15 2 2
68299207        -8  69196816 1 -8 2588 2294 2 3
68302611        -8  69210416 2 -8 2588 2294 2 2
68309407        -8  69244416 1 -8 2612 2306 2 2
68321647        -8  69285216 1 -8 2636 2318 2 2
68321651        -8  69285216 2 -8 2636 2318 2 2
68322327        -8  69292016 1 -8 2636 2318 2 3
68322331        -8  69292016 2 -8 2636 2318 2 2
68329807        -8  69305616 1 -8 2636 2318 2 2
68333207        -8  69326016 1  3 2660 2330 1 3
68337971        -8  69360016 1 -8 2660 2330 2 2
68337975        -8  69360016 2 -8 2660 2330 2 2
68340094        -8  68340016 2 -8 2132 2066 2 3
68341371        -8  69394016 1 -8 2660 2330 2 1
68348169  13036416  73188416 2 -8   83   18 2 2
68367205  13159542  73195216 1 -8   87   19 2 3
68367209  13159577  73195216 2 -8   87   19 2 2
68388287        -8  69550416 1 -8 2756 2378 2 2
68395767        -8  69584416 1 -8 2780 2390 2 3
68395771        -8  69584416 2 -8 2780 2390 2 2
68412087        -8  69618416 1 -8 2804 2402 2 2
68412091        -8  69618416 2 -8 2804 2402 2 2
68412095        -8  69618416 3 -8 2804 2402 2 2
68420247        -8  69632016 1 -8 2828 2414 2 2
68427047        -8  69638816 1 -8 2828 2414 1 2
end
label values pid pid
label def pid -8 "inapplicable", modify
label values i_childpno i_childpno
label def i_childpno -8 "inapplicable", modify
label values i_psu i_psu
label values i_strata i_strata
label values i_aidhh i_aidhh
label def i_aidhh 1 "Yes", modify
label def i_aidhh 2 "No", modify
label values i_scghqa i_scghqa
label def i_scghqa 1 "Better than usual", modify
label def i_scghqa 2 "Same as usual", modify
label def i_scghqa 3 "Less than usual", modify
label def i_scghqa 4 "Much less than usual", modify

Best regards

[Clyde Schechter]

This is a well known limitation of svy: based estimation. The message means exactly what it says: there is some stratum (maybe there is more than one) in your data that only contributes a single observation to the regression. In that situation, you cannot compute survey design-based standard errors. So first you have to find the offending strata.

// AFTER THE REGRESSION, RUN:
by i_strata, egen stratum_count = total(e(sample))
levelsof i_stratum if stratum_count == 1

That will tell you which stratum (or strata) are causing the problem. You have two ways now of resolving the problem. You can simply re-do the analysis omitting that stratum (those strata) by applying a suitable -if- condition to your regression. If there are only a handful of such strata and your total sample is large, then this can be acceptable. Alternatively, you can combine each offending stratum with some other stratum. For example, suppose stratum 5 turns out to have only one observation in the regression. Then, you can combine it with some other stratum, say, stratum 11 by running -replace i_stratum = 11 if i_stratum == 5-. Then re-run the regression. The choice of the stratum to merge with can be difficult. Ideally you should choose another stratum whose characteristics relevant to your problem are as close as possible to those of stratum 5. That is, you want to combine like with like to the extent possible.

[Ana Vasconcelos]

I treat strata with singleton PSUs as certain units. I use the command: svyset i_psu [pweight=i_indscui_lw], singleunit(certainty) strata (i_strata)
Can I solve the problem this way?
Thank you in advance.

[Clyde Schechter]

Well, it will enable you to get standard errors. Whether this is the correct solution, I don't know. I think this question needs to be answered by somebody with a deeper knowledge of survey analysis than I have, and also requires a more detailed explanation of the sampling design. Hopefully, somebody who can provide a better answer will join in.

[Ana Vasconcelos]

Thank you very much for your message.