2
$\begingroup$

We asked a group of subjects to tell us their preferences for a given procedure, of which there were 4 choices. We then provided them with educational material and asked them again of their preference.

I would like to compare the two groups and see if there is a statistically significant difference between the two groups. As far as I know, I am making the assumption that the distribution of choices is non parametric since I assume that these subjects had a pence to begin with. Additionally I am polling the same group so the test I would need to use would have to be a paired test.

Which test should I use?

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ What are these "nonparametric data" you speak of? DNE. $\endgroup$
    – Alexis
    Commented Aug 7, 2014 at 19:51
  • $\begingroup$ Related threads: stats.stackexchange.com/q/10719/930, stats.stackexchange.com/q/11717/930. $\endgroup$
    – chl
    Commented Aug 7, 2014 at 19:56
  • 2
    $\begingroup$ Data is neither parametric nor nonparametric, its just data. What do you actually mean when you call data 'nonparametric'? $\endgroup$
    – Glen_b
    Commented Aug 7, 2014 at 22:57
  • 2
    $\begingroup$ You have asked 15 questions here w/o ever accepting an answer or even upvoting one. Have none of the answers you've received been helpful to you? If any have, please consider upvoting them by clicking the upwards normal distribution to the answers left; if any have resolved your question, please consider accepting it by clicking the check mark below the vote total. $\endgroup$
    – gung - Reinstate Monica
    Commented Apr 2, 2015 at 16:35

1 Answer 1

Reset to default
2
$\begingroup$

The standard test to use for this type of study is McNemar's test. The usual McNemar's test is for only 2 outcomes, but you can extend it for more, the R function mcnemar.test does this for 2 or more outcomes. This basically just tests the null that the matrix of choices is symmetric, but does not give detail on where any differences occur.

Another approach would be a proportional odds logistic regression where you use the "before" results as the predictor and the "after" results as the response variable.

Improve this answer
$\endgroup$
2
  • $\begingroup$ The response variable can have > 2 factors ? $\endgroup$
    – oort
    Commented Aug 7, 2014 at 19:25
  • 1
    $\begingroup$ @oort, I am unsure what you are asking. McNemar's test (or extensions to it) work on paired data (so 2 variables, in your case before and after). The original test only dealt with 2 levels in each variable (yes/no, high/low, etc.) but the extensions allow for more than 2 levels (low/med/high/veryHigh, etc.). Proportional odds logistic regression also uses a response variable that is an ordered categorical variable with 2 or more levels (again it can work with low/med/high/veryHigh). $\endgroup$
    – Greg Snow
    Commented Aug 7, 2014 at 20:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.