# Non-parametric statistical test to compare a continuous and dichotomous variable in the same group

I want to compare between the level of Th17 cells in the blood of a group patients with Sarcoidosis, and whether or not it is correlated with them having fibrosis. So, a continuous variable and a dichotomous variable, and one group, not two. Mann Whitney can't be applied as far I understand. So, what test to use if the data are not normal? Also, what test to use of one of the variables is not dichotomous, but can have multiple values??

• So you are comparing the distribution of some continuous variable between two groups? Can you tell us some more, like sample size, and show us some plots? But maybe try Wilcoxon test Commented Feb 25 at 19:32
• Fibrosis or not seem to define two groups (maybe in your case subgroups) for the purpose of the wilcox test. If you had more observations, maybe stats.stackexchange.com/questions/190156/… Commented Feb 25 at 19:51
• What question do you have for your data to (hopefully) answer, if the Th17 level is high/lower in the fibrosis group than in the control or healthy group?
– Dave
Commented Feb 25 at 20:15
• That sounds like a two-group comparison, exactly what a t-test or Wilcoxon test addresses. What do you see differently? (Do you, for instance, have “before-and-after” measurements for each group?)
– Dave
Commented Feb 25 at 20:25
• Often, and here in particular, “Wilcoxon test” refers to the Wilcoxon-Mann-Whitney U test, which does not require equal group sizes. The signed-rank test bearing Wilcoxon’s name is analogous to the paired t-test and, thus, requires equal group sizes but is a different test. Both the paired t-test and signed-rank test also require a natural pairing of the points, such as before-and-after. $//$ One of your hang-ups seems to be a lack of normality. If you had normality, would a two-sample t-test make sense?
– Dave
Commented Feb 25 at 21:52