Data for my dependent variable will be obtained from 50 respondents who will indicate their agreement or otherwise to 10 questions using a Likert scale (1-4 pts). For each respondent, the scores will then be aggregated or summed up so that each will have an aggregated score of between 10 and 40 i.e., the dependent data. I will like to inquire from experts if multiple regression analysis can be used (instead of ordinal logistic regression) for such aggregated Likert-score data. If so, are there scholarly sources to support the use of multiple regression in such cases. I appreciate your contributions.
Thanks
Reginald.
Ajit kumar Roy
Is correct. For A nice introduction to ordinal logistic regression see the attached notes. Best, David Booth PS remember that ordinal data is always ordinal data. you can't make it something else.
You may want to read these in relation to simply adding up scores
Article Thinking twice about sum scores
Article Understanding the population structure of the GHQ-12: Method...
If you have a total score in dependent without any categorization then multiple regression analysis is fine. Mind it the assumptions first to fit for the analysis.
Use the proportional odds model for Likert data, as recommended by Frank Harrell (mentioned in first post). I went into the gory details on the problems with using metric models on ordinal data over at Frank Harrell's Data Methods forum.
I concluded that there is no reason to take for granted that these models communicate any information (in a statistical sense), which would make any attempt at meta-analysis or information synthesis based upon primary reports using these techniques, futile.
Preprint: Using Metric Models on Ordinal Data
https://tinyurl.com/55d4uty9
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire