In quality reports regarding school results etc it is common to use a 10 point numeric scale and then trichotomizing the results into three categories from the original ten categories, when presenting the results. In this research the questionnaire has statements and the numeric scale consists of statements where the numeric scale is used to rate to which extent the participants agree. In the trichotomization the results are divided into "problematic" (rating 8-10), "sometimes problematic" (rating 7-4) and "never a problem" (rating 1-3), from the original ratings. This have been done to enable the participants to have several answering options, while simplifying the results when presenting them.
There is much discussion about the problems with dichotomizing (e.g., MacCullum et al, 2002, Psych Methods; Cohen, 1983, Appl Psych Measurement, and I've linked to a paper I did where this is one of the topics). Similar issues exist for splitting into three categories.
However, sometimes it is still done. So, in school evaluations sometimes grades A-F or other categorical labels are used. For students, labels (A-F, "on-track") are sometimes used. Similar in medical diagnosis and clinical psychology labels are used. Sometimes this categorizing is in government policy. The main question is how much information is lost and is this amount justified by improving the ease of communication and/or adhering to policy.
If you can give some more details about your specific needs and the distributions of your scores this might help. Also, for an entire scale sometimes taxometric analyses will suggest categories are an appropriate way of looking at the data.
Article Making friends with your data: Improving how statistics are ...
Kristina, I was thinking about your problem but could not find anything to help you. Nevertheless, extreme group assignment may be an alternative for you, at least for understanding which values may be reasonable for you :
quantpsy.org/pubs/preacher_2015b.pdf
Of course the same problems already mentioned by Daniel with regard to dichotomization exist, but if you need to trichotomize your data to subdevide your data into subgroups you may think about tertiles or quartiles of the score distribution.
Hope this helps a bit, Karin
Trichotomization can be useful for communicating results to a general audience, but should not be used for statistical analysis, see attached link for extended discussion.
http://tinyurl.com/o88cs9c
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire