I am preparing a complex survey data for analysis. The survey involved a two-stage stratified cluster sampling. In the first stage, I have 7 strata and 14 sampling units and in the second stage I have 13 strata and 29 sampling units. I have made sure that there are at least 2 sampling units per sampling stratum. However, after declaring the survey design in Stata, the svydescribe command yields strata that are duplicates of already existing strata but with single units. The following table depicts the Stata output after submitting the command svydescribe, final.

enter image description here

As can be seen from the table, in the second stage of sampling, unit/cluster 7 belongs to stratum 4. However, one observation which belongs to the same stratum and unit/cluster has been considered as belonging to a different stratum and unit/cluster. The same holds with stratum 7 and several other strata. What causes such problems and how are they handled?

Update: after serially issuing the commands

egen _svy_tag = tag( PSEUDOPSU PSUSTRATA SSU PSEUDOSSUST FPC1 FPC2), missing
replace _svy_tag = _svy_tag | missing(NWEIGHT)

as suggested by StasK in his comment below, I obtained the following output. enter image description here enter image description here

  • $\begingroup$ I'm not sure this is on-topic, but anyway, you should include the exact commands you issued and exact output. A complete, reproducible example, is even better. $\endgroup$ Aug 20 '14 at 16:41
  • $\begingroup$ This is the command I issued: svyset PSEUDOPSU [pweight=NWEIGHT], strata( PSUSTRATA) fpc(FPC1) || SSU, strata( PSEUDOSSUST) fpc(FPC2). Subsequently, I issued the command 'svydescribe, final'. [I cannot include the output due to limitation of number of characters for comments on this site.] In the output I see several duplicate strata with single units where as in the actual dataset there are no strata with single units. $\endgroup$
    – Ayalew A.
    Aug 21 '14 at 6:18
  • 1
    $\begingroup$ Do not use the comment section to give additional information that is not suited for this section. Use the edit button to edit your original post. $\endgroup$ Aug 21 '14 at 11:53
  • 1
    $\begingroup$ egen _svy_tag = tag( PSEUDOPSU PSUSTRATA SSU PSEUDOSSUST FPC1 FPC2), missing [ENTER] replace _svy_tag = _svy_tag | missing(NWEIGHT) [ENTER] list PSEUDOPSU PSUSTRATA SSU PSEUDOSSUST FPC1 FPC2 NWEIGHT if _svy_tag [ENTER] and add the output to your original post. (+1 to Roberto: this is a live website where you can edit a lot of stuff -- you can certainly edit your own posts). $\endgroup$
    – StasK
    Aug 21 '14 at 14:45
  • 1
    $\begingroup$ The problems are obviously in your code that creates the survey variables from whatever you obtained from the field. You'd have to review and resolve the offending units that must have been mislabeled. $\endgroup$
    – StasK
    Aug 21 '14 at 14:48

i believe you might be looking for the stata option singleunit(centered) ?

the cdc has a thorough description of this issue and how to deal with it programmatically in different statistical languages


the R survey package also has a good succinct page on the subject


  • $\begingroup$ Actually I don't have strata with single units. For strata with single units, I have created pseudostrata by merging them with adjacent strata. Now no stratum contains a single unit in my dataset. However, after declaring the survey design using 'svyset' the 'svydescribe' command returns some strata that contain single units. The strata with single units are just duplicates of already existing strata except that they contain single units as I indicated in my initial post. $\endgroup$
    – Ayalew A.
    Aug 21 '14 at 6:29
  • $\begingroup$ @user52898 are you sure everything's been recoded properly? i'd be surprised if svydescribe prints duplicates as you've shown. check tab [stratum] and make sure you don't have a stratum "17" and another " 17" or something like that $\endgroup$ Aug 21 '14 at 7:13
  • $\begingroup$ @ Anthony Damico, I have checked tab [stratum] and tab [clusters]. I also have checked a cross-tabulation of stratum vs cluster. Everything is fine there. Unfortunately due to space limitations, I couldn't include the tab outputs here. $\endgroup$
    – Ayalew A.
    Aug 21 '14 at 12:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.