0

My research is about how senior citizens perceive doing exercise as a way improve mental health.

0 1 2 3 4 5 prefer not to say?

0 = not important
5 = people think that exercise is important for mental health.


Total participants: 15






  Choices
  # of response






  0
  0




  1
  0




  2
  0




  3
  1




  4
  2




  5
  11




  Prefer not to say
  1

My questions are:

  1. How to calculate mean on this type of data?
  2. How do I interpret mean? How can mean prove that senior citizens think that exercise is good for mental health?
  3. What should I do with the "Prefer not to say" data? Should I do common-point imputation (so that the answer from common-point imputation will be an addition response to choice #5)?
  4. Can I just say that since choice #5 has the most votes then the study proves that senior citizens perceive that exercise is important for mental health?

I am trying to understand the role of mean on this type of study.

gung - Reinstate Monica
  • 145,122
rahstame
  • 101
  • 1
    A Likert scale provides ordinal categorical data, for which the sample median is an appropriate way to express the 'center' of the sample. You can use the sample mean if you think the Likert scale is numerical. For example, if you think the difference between responses 0 and 1 is of the same importance as the difference between 4 and 5, and if you think the a difference between 3 and 5 is twice the difference between 4 and 5.// A bar chart showing numbers of answers for each Likert value might go along with reporting the median. – BruceET Dec 15 '20 at 04:46
  • Great input. In the sample data above the choices 0 to 2 have no responses. Should I include them as part of my data set? – rahstame Dec 15 '20 at 07:18
  • 1
    I'd probably leave them in. In any case any report should mention that 0-2 were not chosen. – BruceET Dec 15 '20 at 08:03
  • 1
    @BruceET: do you want to post your comment(s) as an answer? Better to have a short answer than no answer at all. Anyone who has a better answer can post it. – Stephan Kolassa Dec 15 '20 at 08:14
  • Thank you. The mean is 2.5 because I added the 'prefer not to say' in choice #5 which makes it 12. Data set (0+0+0+1+2+12)/6. I think the value of mean in this case is not significant for a conclusion. Can you suggest for a better data interpretation? in such I way I can give a more convincing realization. As I can see that most of the choices is item #5. – rahstame Dec 15 '20 at 08:46
  • 1
    @StephanKolassa. Done. – BruceET Dec 15 '20 at 08:58
  • I'd worry that 'prefer not to say' is out of order for your ordinal variable. – BruceET Dec 15 '20 at 09:04
  • During my research learned that 'prefer not to say' can be considered a missing value. I've used imputation to address it. By using "Common-Point Imputation", I chose to transfer the vote to the value of scale #5. Scale #5 has the largest value that's why I transfered the vote to it. Sounds valid for you? – rahstame Dec 15 '20 at 09:11
  • I disagree with the last comment. If Prefer not to stay is an option given to (e.g.) me, it is unethical and even insulting to impute that I really have in mind a different answer. Same holds for everyone else. – Nick Cox Dec 15 '20 at 09:38
  • There are many threads on this topic. https://stats.stackexchange.com/questions/67551/calculate-mean-of-ordinal-variable shows a variety of views, including more support for using the mean as a pragmatic choice. Examples like 2,2,2,2,3 and 2,2,2,3,4 are almost everywhere to underline that medians are not subtle summaries. Grade-point averages, anyone? – Nick Cox Dec 15 '20 at 09:45

1 Answers1

3

A Likert scale provides ordinal categorical data, for which the sample median is an appropriate way to express the 'center' of the sample.

You can use the sample mean if you think the Likert scale is numerical. For example, if you think the difference between responses 0 and 1 is of the same importance as the difference between 4 and 5, and if you think the a difference between 3 and 5 is twice the difference between 4 and 5.

A bar chart showing numbers of answers for each Likert value might go along with reporting the median.

I would probably leave the 0-counts in the table. If you don't, you should mention them in the narrative of a report.

BruceET
  • 56,185