Good morning,
I could not find literature about this so I am asking.
I am organizing data for a questionnaire (N=100) that foresees some personal questions like level of education. For the aim I asked the respondents to select from a list of possible choices their level of educational. The "problem" is that some categories were indicated by only two or one people.
Personally, I would exclude those categories for the analysis because they are maybe not representative, but I have never had such a situation and so I dont know what really do. In case I have to include them, a non-parametric test, say Kruskal-Wallis, can be used?
With only one or two people, the variable has close to zero variance, so your statistical power is likewise somewhere near the floor.
I would either pool education categories so that you have categories broad enough to have a reasonable number of people in them without being so broad as to be meaningless. If you can't do that – if, for example, you have someone who was educated in a Tibetan monastery – then you may simply have to exclude them from analysis.
With only one or two people, the variable has close to zero variance, so your statistical power is likewise somewhere near the floor.
I would either pool education categories so that you have categories broad enough to have a reasonable number of people in them without being so broad as to be meaningless. If you can't do that – if, for example, you have someone who was educated in a Tibetan monastery – then you may simply have to exclude them from analysis.
@Vito I completely agree with @Ronan on the point of pooling education categories into lesser number and broader scope as greater number of categories with minimal number of responses would lead to no valid results.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics