# Regression modelling with mixed data set: categorical and numerical predictor variables

There are thirteen predictor variables which are a combination of 8 continuous, 4 binary and 1 categorical variables. The dependent variable is again categorical. I understand that I need to use dummy coding for binary and categorical variables.

How should I model the dependent variable with respect to predictor variables? Which regression model can be used?

• The standard regression model in these cases is called : multinomial logistic regression. To my knowledge implementations of it can be found in almost all major packages (eg. R, MATLAB, SAS, etc.) – usεr11852 Feb 5 '16 at 5:42
• The fact that you have both categorical and numerical predictors is of no consequence, with a suitable response variable, multiple regression could deal with both. However, the fact that the response is categorical is crucial in determining what kinds of analyses may be suitable. – Glen_b Feb 5 '16 at 7:41
• I tried looking for multinominal logistic regression in R. Initially, I modified my dataset using, mlm <- mlogit.data(diffdata, varying = NULL, choice = "D", shape = "wide") , model.1 <- mlogit (D ~ T5 + T6 + KC + (1|T1) + (1|T2) + (1|T3) + (1|T4) + (1|S2) + (1|S3) + (1|S4) + (1|S5), data = mlm, reflevel = "easy"), I am getting following error, Error in solve.default(H, g[!fixed]) : Lapack routine dgesv: system is exactly singular: U[3,3] = 0 – Shilpi Feb 5 '16 at 8:42
• are your columns independent? see stats.stackexchange.com/questions/70899/… – mandata Feb 5 '16 at 14:20