Is there an alternative to paired t-test two multilevel ordinal variables? (similar to McNemar test) Suppose I have two variables that measure some form of severity of illness on identical ordinal scale; one measured before treatment and the other measured after the treatment.
I want to test if there is an improvement between treatments.
Non-parametric correlations wouldn't be suitable; they would tell me if there is a relationship between the condition before and after the treatment. 
 A: Two points:

*

*As ttnphns pointed out, the Wilcoxon sign-rank test is indeed the nonparametric analog to the paired t test (just as McNemar's test is the binary outcome analog to the paired t test).


*You write that nonparametric correlations (presumably Spearman's $r_{\text{S}}$) "wouldn't be suitable; they would tell me if there is a relationship between the condition before and after the treatment."
But I would suggest that $r_{\text{S}}$ tells you if there is an association between group and severity of illness, and further, that (assuming $d_{i}=\text{rank}_{\text{pre}}-\text{rank}_{\text{post}}$) the sign of $r_{\text{S}}$ tells you whether that association is a monotonic decrease ($r_{\text{S}} > 0$ implies that the post group had lower severity), or a monotonic increase ($r_{\text{S}} < 0$ implies that the post group had higher severity).
A: If you are familiar with ordinal logistic regression models and mixed-effects models with lmer in the lme4 package, you may want to check out the clmm function in the ordinal package. This function, built on lmer, may prove especially useful if you have covariates to control for.
