I perform binominal log reg to predict the presence of a disease using 9 variables. One of the variables was log transformed because it was not normally distributed. This log transformed variable came out of the log reg equation with a negative coefficient although in the univariate analysis it significantly has a higher mean value in the group with the disease. Can this be possible? Thanks in advance for any answers.
Asked
Active
Viewed 11 times
0
-
See https://stats.stackexchange.com/search?q=regression+change+sign for more. – whuber Nov 30 '20 at 19:15
-
"One of the variables was log transformed because it was not normally distributed." This concerns me. We do not care about having normal predictors. In fact, predictors often are categorical! – Dave Nov 30 '20 at 19:18