I know that there has been many questions regarding analysing likert scales but I still have not found the easiest way to approach my analyse.
I have done a survey with 100 participants that has 3 groups of questions based on the likert scale (1-7), with 8 questions in each group. Depending on the answers of the first group of questions, I would like to see the how it correlate with the second and the third group of questions.
My hypothesis is that if the participants answer positively to group 1 questions, that it will correlate more to positive answers in groups 2 then group 3 questions. I therefor wanted to compare two groups of likert scales with each other to see how the correlate (group 1 and 2 + groups 1 and 3).
I have started to look into conducting a principle component analysis as a data reduction technique to reduce the amount to components, but Im not sure if this is the best way to approach it.
I know there is probably several complicated ways to do this that may be more correct then others, but I'm looking for an easy way that I can easily understand. Any help would be very much appreciated!
Try Kendall rank coefficient, Cronbach alpha statistic and/or factor analysis
Just to clarify. I have done the cronbach alpha already to test reliability of the group questions. Im aware I can do bivariate analysis using spearman or kendall rank correlation to investigate the coefficient between 2 ordinal variables. But it didn't seems to be correct approach with several grouped questions?
One way people handle this problem is by summing the individual Likert items to a composite Likert scale. That is, you would sum the values for a respondent for each of the 8 questions.
Obviously, each respondent will have had to answer all questions, and all the questions will need to be in the same "direction".
This is common. But it has some theoretical problems. For example, is it true that one person answering 4 on question A and 4 on question B is “equal” to someone answering 1 on question A and 7 on question B? Probably not. On a junk food scale, someone eating chips “sometimes” and cake “sometimes” probably eats better than someone who eats chips “never” but cake “always”.
If you do this, you will have a single value for each respondent for each of the scales. You can then do the correlations as you are thinking. You might consider Kendall or Spearman correlation.
For terminology and simple scale-creation, the introduction to this Wikipedia article is fairly clear.
https://en.wikipedia.org/wiki/Likert_scale
The following is documentation for a specific function for a specific function in R, but I found it to be a helpful read. The function does produce correlations among the scales, probably in a statistically acceptable way, but it's not clear to me how it does it.
https://personality-project.org/r/html/score.items.html
Finally, there's some more in-depth discussion here.
http://stats.stackexchange.com/questions/9918/how-to-compute-correlation-between-within-groups-of-variables
You can use Cronbach alpha, Principal Component analysis then, your can further use it to create any index, you can use cluster analysis and non-linear regression analysis
- Badges
- Science topic
- Similar topics
- Agricultural Science
- Horticulture