0

I have a data set of two groups of people, both are n=30 so n=60 total.

I have administered a scale with 11 items which produces a score of either 0, 1, or 2.

Each of the items measures something different, so I want to know if people in group 1 score differently on specific items compared to people in group 2.

I am not sure what test is appropriate to use here, can I perform an ANOVA with just two groups and compare them on all eleven variables? Or do I need to use a chi-square test of independence to assess the relationship between each item and group?

Alternatively, if I converted the scale from 0,1,2 to a dichotomous 0-1 scale, would that enable any different types of tests? It makes sense theoretically to do this, but I am unsure if it is wise to compare means of items between groups on a 0-1 scale.

Riverarodrigoa
  • 121
  • 7
KyleKatarn
  • 1
  • See also @FrankHarrell 's [answer here](https://stats.stackexchange.com/questions/137060/3-values-ordinal-variable-mann-whitney-or-chi-squared/137074#137074) – Sal Mangiafico Jul 17 '18 at 13:44

1 Answers1

1

As a general consensus, Likert-type item responses are usually treated as ordinal data ‡.

Probably your best tool when dealing with an ordinal dependent variable is ordinal regression. Depending on what software you use, this can be fairly straightforward §.

Using a Wilcoxon-Mann-Whitney test is also appropriate.

The Cochran–Armitage test is also appropriate, although I believe this test needs to make an assumption about the spacing of the categories. That is, for example, that the space between 0 and 1 is the same between 1 and 2.

I don't recommend converting your responses to 0/1. That just discards information.

I don't recommend using t-test or anova for this kind of data. This kind of data likely doesn't meet the assumptions of this test.

_____

‡ This can be distinct from Likert scale data, which is often treated as interval data.

§ Two-sample example in R.

Sal Mangiafico
  • 7,128
  • 2
  • 10
  • 24