Hello!
I am conducting research on a 5-point Likert scale (strongly disagree, disagree, neutral, agree, strongly agree). The sample size is n=240. There are a few issues I have, that I can´t solve.
1.)Since all survey questions are ordinal, should I choose Spearman Correlation or would Linear Regression also be an option?
2.) Control variables such as "age" is ratio and will also be used in the analysis. What test would be a good choice for ratio and ordinal data?
I hope I provided enough information.
Thank you!
Hello Rick!
Before we step any further, consider this - maybe we don't need to reinvent the wheel! Look up online (preferably on websites like PubMed and Google Scholar) for other researchers that used questionnaires for place attachment - ideally the same questionnaire you used. Carefully examine how they used the answers and proceed with your analyses in the same way. This way you're using tried-and-tested methods of analysis. Maybe you'll find that there are pre-specified ways of adding together the results to end up with attachment scores.
Now, on to a point-by-point answer:
1) As I wrote above, if the questionnaire you're using is not entirely your own creation, it's very likely that its developers devised a way to derive an "attachment" score that may not be the mean OR the median. Summing all answer scores is not a bad idea, but you must first make sure that all statements to which interviewees must answer "tend" in the same direction e.g. that if someone answers Strongly Agree to all of them he's at 100% attachment and vice versa. If they're mixed, then the sum is not enough - you'll have to figure out which statements are in line with attachment and which are opposite from attachment. However you do it, after you come up with one singular score of attachment, things get easier. Most of the time, data analysts run a test to see if their variable is normally distributed. You can use the Shapiro-Wilk test in SPSS (it's a little tricky to find but it's there). If P>0.05, the attachment score is normally distributed - you can use the mean±standard deviation to report it. Otherwise it's best to use the median and inter-quartile range.
2) You could try either - linear regression should be preferable because you can then uncover which of the non-attachment factors are independent contributors/predictors of the patients' attachment score.
3) A good way to test the correlation between patients' age and their attachment score would indeed be a test like Spearman's correlation coefficient.
Good luck with your analyses!
If you decide to use regression wilh a likert DV. Then ordinal regression is what you want to use. See the attached screenshot reference for details and examples. This source is available in the z-library. You can Google examples in your favorite package as well. You should also check out likert scale development. Google would again be helpful to you. Best wishes David Booth
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS