If I have a questionnaire with items on a Likert scale (1 to 5) and a sixth option "I do not know/Does not apply", is it okay to consider a "I do not know/Does not apply" response as a missing value and use an imputation method, just like I would consider and impute the missing values in the case when respondent to not answer at all for certain items?
6
-
1$\begingroup$ Re the "does not apply" option: Are you suggesting you would impute values where the respondents say they are not even applicable? For instance, would you really try to use the female responses to a health questionnaire to impute the numbers of pregnancies that male respondents have had? $\endgroup$– whuber ♦Commented Jul 29, 2014 at 19:21
-
$\begingroup$ I am interested in the perception of the respondents regarding certain issues so if something is not applicable from their point of view, this does not neccesarily mean it is not applicable in general. $\endgroup$– LucyCommented Jul 29, 2014 at 19:24
-
$\begingroup$ Take the pregnancy analogy to heart: it seems you would impute perceptions to people who say they are not even applicable. That sounds like the same kind of mistake to me. This suggests there may be a sharp distinction between "do not know" and "does not apply" answers. $\endgroup$– whuber ♦Commented Jul 29, 2014 at 19:27
-
$\begingroup$ I very much agree with @whuber 's comments. That said, multiple imputation can be improved by incorporating different kinds of missingness into the imputation model itself (e.g. in some cases "refuse" to answer missing values could reasonably have a different imputation model than "don't recall" missing values; same for "don't know", etc.). $\endgroup$– AlexisCommented Jul 29, 2014 at 19:42
-
$\begingroup$ My research is in the field of business so the respondents give their perception regarding a phenomenon in this field. In the case of the field of medicine yes, the problems are more clear, less interpretable. But I am not sure this is the case in business. I am interested if it is okay to impute these kind of answers or not. I need to undertake factor analysis, correlations analysis and regression analysis. Can I interpreted from your answer, Alexis, that it is okay to impute them? I used only a single option, named "I do not know/Does not apply". $\endgroup$– LucyCommented Jul 29, 2014 at 20:02
|
Show 1 more comment
1 Answer
$\begingroup$
$\endgroup$
You should think about when a person would simply not respond to the questionaire and when a person would say "I don't know, etc." Are these signals for the same thing or possibly different things? Meaning, is a person who responded with "I don't know" just as likely to ignore the question?
Depending on what model you are trying to use, the distinction can prove useful.
Edit: just read all the comments. I agree. A question asking about number of pregnancies with responses "not applicable", "I don't know", and "ignore" contain different information. It would be a mistake to conflate them.
