Am analyzing my data and I have this multiple response questions where respondents are required to choose all options that apply. However, my supervisors can't seem to agree on the best way to analyze them. Are the percentages based on the sample size or the total number of responses? Any clarification is welcome.
This kind of data is indeed difficult to analyze. SPSS has a function called "Multiple Response" that can handle this kind of coding, but I do not know any options for statistical analysis. You may have to limit your reporting to descriptions of the patterns in the responses.
Treat as a set of dummy variables, so if you have 10 things that they may tick, you make ten 0/1 variables (this may be how you entered the data into the computer anyway). If you have a "none of the above" option you don't need to use this.
The difficulty is that is that some respondents may treat this as "I'll tick the five most important ones if all are applicable".
SPSS has the module 'tables' to analyze multiple responses. You can tests for proportions and means. The module also performs Bonferroni corrections as many comparisons are made. The module 'tables' has to be acquired separately, so may not be available to you. In the Netherlands, it is delivered with the version that university staff and students can buy.
Daglous, percentages based on sample size makes better intuitive sense. Percentages of the total number of responses is an artefact that is based on varying number of responses per survey respondent and harder to interpret.
Thomas Puliyel , what you need might be based on what you want to test. If you have two groups, you not only want to test for percentages within a group, but also to the total responses between the two groups. SPSS (in tables) allows you to do both. Maybe the 'average' amount of responses are different as well.
The tables in contention are from simple frequencies (attached sample). Which of the two can you recommend? Why? Peter Moorer Daniel Wright Thomas Puliyel Mahadi Hasan Miraz Niyungeko Antoine David L Morgan
Daglous Gesora , In your case it might be not that different. The correlation between the percentage within cases or responses are very high (>0.99 see file). On average respondents ticked two options. If you divide the number in the cases (if I guessed correct) column, you more or less get the number in the responses column. Maybe you should combine them into one table (as in the file I attach). It only adds on column to your table. You might want to give the number of responses and cases separately.
Daglous Gesora , I gave a fairly vague answer before since you didn't say what your research questions were or how you want to use these responses. Can you say what these are? Something like an IRT/latent trait model on the items (i.e., on the dummy variables for these) might be worthwhile. Bartholomew et al. ( https://www.amazon.com/Latent-Variable-Models-Factor-Analysis/dp/0470971924 ) discuss examples with five or so items, and argue why this is more defensible than those using more.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size