I have a bit of an issue with a data set I’m working through in SPSS. The data comes from a study wherein the independent variable is dichotomous (treatment vs. control) and the dependent variable is continuous (hours in the hospital). There are three demographic factors, both dichotomous and continuous, which I am trying to statistically examine as possible confounding factors. However, I cannot seem to find a statistical method to accomplish this task. I am also open to simply examining one potential confounding variable at a time if need be. Here are the methods I have considered:
ANCOVA. This should work for all possible confounders at once, but the dependent variable has 11 outliers and is not normal according to the Shapiro-Wilk test. These issues could invalidate the result of an ANCOVA.
Cochrane-Mantel-Haenszel statistic after stratification (examine one dichotomous potential confounding variable at a time). Stratification does not seem possible as the dependent variable is continuous and there is no logical cut-off to use to transform it into a dichotomous variable.
Multinomial logistic regression. I do not believe this approach works because the dependent variable must be nominal.
Multiple linear regression. This approach will not work as the independent variable is not continuous.
I am beginning to think that there is no statistical way to find what I desire. Is that correct or am I missing something?