Was wondering if anyone could help me, I am in my third year and I'm in dissertation hell.
I have used 5 point Likert Scales in my questionnaire (Strongly agree, agree, so on).
I'm now looking at differences in groups via SPSS. I have been advised to used Mann-whitney U.
I got 560 responses around equal amounts in the groups I want to compare.
However, I've read in a few articles that with a big sample size T-test might be better.
Please help :)
Leanne
Hi Leanne,
You are three years in - and you are not sure what to do? What have you been reading in that time - and how well are your supervisors advising you? Have they not advised you which tests to use?
Dean,
Yeah they have advised me to use a Mann-Whitney U as I stated. I am just checking.
There is lots of contradicting ideas out there.
One issue is that there isn't tremendous consensus about how to handle Likert data, so you may not find any definitive advice, whether from advisors or from literature. I'm assuming that since you are working on your "dissertation" (in the U.K.) that you are at the undergraduate level. If so, it is probably appropriate to follow the advice of your advisors. One solace is that you aren't likely to get tremendously different answers whichever test you use.
A few additional comments:
1) Are you analyzing a single Likert item or a scale made up of several Likert items? The analysis may be different depending on the answer.
2) With 500 observations, you are likely to find a significant p-value even with small differences between groups. Don't rely too much on the p-value. Look at measure of the size of the effect (difference in medians, or proportions of each answer, say), and determine if the effect is meaningful.
3) t-test and Mann-Whitney test very different hypotheses. Be sure to understand these hypotheses so that you can report the results correctly. Also note that the Mann-Whitney doesn't necessarily test for a difference in medians. Distrust any source that tells you it does.
If you use a Mann-Whitney test, you will need to analyze each item separately. This is quite different from combining the individual items into a scale with interval-level properties.
Because there have been so many questions here about how to analyze Likert-scored items, I have compiled a set of resources on this topic:
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
Hi Leanne,
If you're looking at differences, you might consider some type of ANOVA. If you're looking at relationships, then you might consider some type of regression.
Good luck!
--Adrian
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size