Assessing change using McNemar's test

rsm

New Member
#1
I have Pre and post-intervention test scores of a number of skills rated from 0-3 (0 = Poor, 1 = Fair, 2 = Average, 3 = Good) for 5 participants. I am trying to use McNemar's test for each skill to test for change and would greatly appreciate some help. I am unclear about whether -

1) I can use McNemar's with only 5 scores ? I know this violates assumptions for using Pearson's chi square test and I'd hae to use Fsher exact instead.

2) If I use McNemar's, can my table look like Pre-test (0,1,2,3) x Post-test (0,1,2,3) ?
Someone suggested I count the number of participants whose scores increased (c), decreased (a) or did not change (b) and use the formula (c-a)^2 / (c+a) ? However, this seems like I am directly counting the change rather than calculating it from scores.

Any help would be greatly appreciated! Thanks!
 

rsm

New Member
#3
Thanks Karabiner, I see what you mean. I have run the Wilcoxon signed rank test but was worried about the very small sample size.Would that test be appropriate?
Alternatively, would you recommend McNemar's if I analyzed it as Pre-test (score less than or equal to 1, greater than 1) x Post-test (score less than or equal to 1, greater than 1) ?
 
Last edited:

Karabiner

TS Contributor
#5
I have run the Wilcoxon signed rank test but was worried about the very small sample size.Would that test be appropriate?
I recommended the sign test instead of the signed rank test,
since I do not know whether the signed rank test is appropriate
for ordinal data. In any case, with n=5 the power of any
test is extremely poor.
Alternatively, would you recommend McNemar's if I analyzed it as Pre-test (score less than or equal to 1, greater than 1) x Post-test (score less than or equal to 1, greater than 1) ?
No.


With kind regards

K.