Statistical difference between percentages - Is this possible/which test would I use?

JOD2

New Member
#1
Hello,

I have a survey relating to staffing issues which was sent to staff across England.

An example question is - "I was provided with appropriate support and supervision" with answers on a Likert scale from Strongly agree to Strongly disagree and I've calculated the percentage of each response for England as a whole.

The data can also be broken down by region so I want to compare the percentage for each region to the percentage for England. I can calculate the percentage difference but is there a way to tell if there is a statistical difference?

Furthermore, I can break down the data even further, to look at setting specific data, so would ideally like to find whether the difference is statistically different overall for each of those too. This is a result table, with the number in brackets the difference between that region's data and the England data. The England data has 19 - 20,000 responses for each question but the numbers for each region are fairly small so will that also have an impact?

Thank you so much for any help!

regional data analysis.jpg
 

Karabiner

TS Contributor
#2
with answers on a Likert scale from Strongly agree to Strongly disagree
To be exact, this is NOT a Likert scale. This is just a Likert-type item, an item with a Likert response format.
A Likert scale consists of several of such Likert-type items.

I've calculated the percentage of each response for England as a whole.
Likert-type items measure responses on an ordinal scale, so you can use the median score for England as a whole,
and for each region, respectively, to describe grouped responses.

I want to compare the percentage for each region to the percentage for England.
It would not make sense to perform a statistical test here, IMHO. It would mean that you compare
(at least in part) each region with itself.
Why not a global test here, comparing all regions at once (Kruskal-Wallis H test)? And if this turns out
statistically significant, then pairwise comparisons between regions (U-tests). Or, one test for each
region, comparing it with the rest of England (also U tests).

With kind regards

Karabiner