What's the best way to analyze a survey question that is a matrix with one answer per row, where the rows are categories (a, b, c, d) and the columns are an evaluation (not important to very important)? Does the question need to be split up (i.e. into each category) for analysis or is there a way to capture individuals' evaluation of categories in relation to each other?
Seems possible, but could you please explain a bit more about your data arrangement? Does the data set have only four rows? How have you arranged the response from all participants/individuals?
If I understand correctly, you have a question with four subquestions. Assigning ordinal responses to your index of not important (0) to very important (4 or 5?), you can first separately score each of the four responses, and then combine them for a type of (possibly weighted) total representing all 4 rows. Depending on what software you are using there are different ways to compress/combine rows of data per responder into a single row.
Is this helpful?
Hi Prasath,
Thanks for replying. The question asks participants to score four individual items on a scale of importance, with each item being a factor that possibly influenced their decision to do something, we are investigating (i.e. Please rate the following options in terms of their importance (on a scale...) in your decision to.... There are four options and the scale goes from 1-4. I know how to seperate it out and analyse it in SPSS, but I wondered if there was anyway to capture the spread and relationships between all four catergories for the group (i.e. those who ranked a) as very important most often also ranked b), as very important, or those ranked a) as very important most often also ranked b) as not important, etc.). I guess you could do basic relationships like this once the columns were separated out in SPSS, but I just wondered if there was a method I was unaware of that could capture the situation across all four categories for individuals in order to then compare the different variations within the group - does that make sense?
necessarily depends on the objective or purpose of the research. usually when nothing is clear, data mining is an alternative. When the study is exploratory, the search for association (correlation) is another possibility. But like a good statistician, I can not say which is the best, without knowing the purpose of research or survey.
Thanks for the additional information Emma. If the data is arranged as one row per respondent, try "cross tabs" on SPSS and see if you get what you wanted.
I am not sure if you would get same kind of table with a "split file" function.
You could also do a statistical test of significance if required. However as mentioned above, the test may not be appropriate without knowing your actual question/purpose.
Thanks Prasath, Nora and Carlos for the input, much appreciated. It's really exploratory at this stage. I can definitely do the crosstabs though to start off with. I think I am interested in finding out if there are correlations between scoring of categories, but it may be best to do the more simple analysis first.
Some indications are given in the following post, that maybe you can see the usefulness
https://www.researchgate.net/post/Frequency_versus_percentages_which_one_is_better_to_report
For this type of data, you can for example use a multiple correspondence analysis (MCA). It's alike principal component analysis but more appropriate for categorical data.
If you are interested whether subjects rating a particular criterion as very important rate also another one as very important as well, polychoric correlations may be helpful.
Hi Kremena - thank you for the reply. Yes, that's kind of what I'm trying to do, so I'll look this up.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis