Hello everyone,
I want to analyze several Likert Scales ranging from 1 Strongly Disagree to 7 Strongly Agree using SPSS. I have a total of 6 Likert Scales, which I put together out of 5 Likert items. In other words, I have 30 Likert item questions, which deal with 6 theoretical concepts. I tested for Cronbach's Alpha and all scales are above .70. Now I want to find the correlation between one of these Likert Scales and the rest of them. Also, when I put the Likert items together to form a scale and create a new variable, do I use the Mean or Mode? My question is what kind of analysis is suitable? One-way anova? Linear regression? Our just Pearson? Thank you in advance!
Have a look at this rather sobering paper:
Liddell TM, Kruschke JK. Analyzing ordinal data with metric models: What could possibly go wrong?. Journal of Experimental Social Psychology. 2018 Nov 1;79:328-48.
https://osf.io/9h3et/download?format=pdf
When you combine items into a scale, you are assuming that the Likert item responses are actually numeric and, in particular, equally spaced. Since you have already made that assumption, you can treat the scale responses as numeric. Whether you can use tradition parametric models (anova, pearson correlation, and so on) ---- or stick with nonparametric approaches (kruskall-wallis, spearman correlation, and so on) --- depends on whether the data meet the assumptions of the traditional parametric tests, or can be reasonably assumed to meet them.
Typically, scales are created for each respondent based on the mean or sum of her responses to items. Though I suppose there are other possible approaches.
What analyses you do depends mostly on why you did this research, which theories/hypotheses you are evaluating. Specific choices do make assumptions, like about distributions, but first you should say you are doing this research.
Have a look at this rather sobering paper:
Liddell TM, Kruschke JK. Analyzing ordinal data with metric models: What could possibly go wrong?. Journal of Experimental Social Psychology. 2018 Nov 1;79:328-48.
https://osf.io/9h3et/download?format=pdf
Ronán Michael Conroy, I just did a Ctrl-F search on in that article by Liddell & Kruschke, and found that it does not appear until the reference list. I suspect, therefore, that they are talking about analysis of single Likert-type items, not proper Likert scales (http://www.john-uebersax.com/stat/likert.htm). As I understand it, Likert never claimed interval scale properties for single items. He claimed (approximate) interval scale properties for the means (or sums) of several items intended to measure the same thing. (Although Sal will probably jump in at this point and say that summing or averaging assumes interval scale properties.) ;-)
Lukas Sommer , see also this thread that David Morgan started:
- https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
HTH.
EDITED 7-May-2019
I've now read much of the Liddell & Kruschke article, and see that they did indeed address the issue of averaging (or summing) several items. The following is from Section 7 (Average of multiple ordinal items), with emphasis added.
--- start of excerpt ---
Some authors have argued that, despite the ordinal character of individual Likert items, averaged ordinal items can have an emergent property of an interval scale and so it is appropriate to apply metric methods to the averaged values (e.g., Carifio & Perla, 2007, 2008). It is intuitively plausible that the averaging could produce data that at least look more continuous and therefore may not suffer from the problems pointed out above. Unfortunately that intuition is wrong. We show in this section that an average of ordinal items has the same problems as a single item. (p. 334)
--- end of excerpt ---
That page number is from the final published manuscript, which can be found on the sciencedirect.com website for the journal. (When I tried to paste it here, it got linked to a RG page for the article.) HTH.
Article Analyzing ordinal data with metric models: What could possib...
Lukas,
If you (or your departament) can invest USD149+taxes, I would strongly recommend you to buy winsteps (www.winsteps.com). It is a super cheap software specificaly oriented to scale analysis with Rasch models . It has a fully functional demostration version called ministeps. If you want to check it out I can give you more advice on how to perform Rasch analysis with likert items.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS