Hello Kazusa,
the reported behavior is a bug. The number of variables generated from a multiple-choice question should be the smaller number of the:
nvars=Min(200, nchoices, nmax)
where nchoices is a number of categories in the multiselect question, and nmax is the restriction on maximum number of selections.
Apparently in this case the number of variables is 200 regardless of the other values, which is incorrect.
My immediate reaction was to specify the max number of selections to be equal to number of choices (11 in your case), but that still didn’t help.
This is a bug and we will fix it. The number of columns will be smaller (11 in the case of the example shown).
Note that in this case (combobox presentation) the data is not a set of dummies, but a set of selections, so var__0 will have the code of the first item selected, var__1 will have the code of the second item selected, and so on. Other variables will have a missing code.
After the fix the number of variables will be smaller (variables after var__10 will always be missing, since there are max 11 choices, so all choices will be accommodated by the first 11 vars making the remaining ones obsolete/nonsense).
If you are using a script for a statistical package, it should scan the observation var__* from left to right and stop on encountering a missing value. When written generically, such script would equally tolerate the buggy (current) and correct (after fix) versions of exported data.
Or you could manually patch it right now (if you know all your multiselect variables in combobox mode), e.g. in Stata:
capture drop varname__11-varname__199
The command capture will ensure that your script remains in working condition even after the fix will remove these variables.
Best, Sergiy Radyakin
