6

I want to compare mean inhibitory concentration (continous variable) in patients with outcome A versus outcome B. I used Mann-Whitney U test. However I would like to control those results for the parasite density, which is also a continous variable. Van Elteren test only allows to correct for strata. Is there any similar test that allows to correct for a continous variable, or does this not make any sense?

Kamala
  • 61
  • 2
  • If you're comparing a continuous response variable between two groups, have you ruled out just transforming the variable to normality and fitting an ANCOVA? – atrichornis May 16 '13 at 15:28

1 Answers1

8

A generalization of the Wilcoxon-Mann-Whitney test is the proportional odds ordinal logistic model, which accepts covariates in addition to the group variable you are mainly testing. Note that the prop. odds model does not need more than one observations at each unique value of $Y$ in order to work well.

Frank Harrell
  • 91,879
  • 6
  • 178
  • 397
  • Thanks for this. Sounds a bit above my stats... Is there a STATA command for it? – Kamala May 16 '13 at 14:09
  • 1
    The command in Stata [not STATA] is ologit. – Nick Cox May 16 '13 at 15:16
  • 1
    I am sure @NickCox is right about Stata, but if this is "a bit above" your stats, it might be wise to hire a consultant (or do something simpler). – Peter Flom May 16 '13 at 18:20
  • 1
    I always look forward to your questions and answers, Frank. Even when you discuss topics I feel I already understand, there's usually some interesting idea or fresh perspective. – Glen_b May 17 '13 at 00:06