|
|
Dear list,
I'd like to ask about one issue - more statistical in general than SPSS-specific.
(sorry for posting again, first time I was having technical difficulties with registration)
I have three variables based on open question: "name up to three your favourite movies". Following some criteria these movie titles were recoded to 5 categories + "nothing mentioned" option (aka missing - respondents don't have to use up all three "slots"). I would like to try some clever scaling methods to construct syntethic variable (continouos or categorical) based on these three variables. But of course, there is a problem:
1) these three variables are not independent by design - "missings" are correlated - e.g. if someone write in one movie, "missing" was entered in second and third variable (so second variable "missing" implies third variable "missing")
2) order of entered movies shouldn't matter, but in this structure it does
I tried to restructure data to consist of 5+1 variables "number of movies in category 'x'" with range zero to three. It will deal with problem (2), but the independence (1) would be even more significant. E.g. it's clear, that if someone with value 3 on one of these new variable has to have value 0 on all others...
I wonder, if it could be solved for example by LCA with constraints? Or any other technique? I wouldn't mind if it would be MCA, LCA, or something else.
Thanks for any suggestions
Ondrej
|
Locked
