I'm running a linear regression in SPSS to test for effects of a binary variable (X) on cost of hospital admission. The variable is correlated with a cost increase of around $3000W
When the model has variable X and a few other correlated categorical variables, I get a B value of $2400 for X.
As soon as I add length of stay, I get a B value for X of -$2400.
What might be causing this? I already tested for multicollinearity, all of my VIF values are less than 1.1. I haven't had any success switching to a generalized linear model.
I have tried transforming every variable with log, inverse, square root, etc. Length of stay and cost have a fairly linear relationship (Rsqu = ~.42).
I am convinced at this point that the issue is with how the variable is coded or how I am imputing it. However, recoding the variable X from 0 and 1 to 1 and 2 or 10 and 20 did not help. Has anyone had this problem before?
