Dear Experts,
I've to fit one logistic regression model based on one of our survey data where we 've a "segmented variable(coded as 1 to 5).Now I've to predict this variable by using census data.As I've already selected some variables from census by using univariate analysis/ANOVA and merged these variables with that segmented variables that's why I got 16 independent variables and (after splitting that dependent variable as 5 variables"just to make them binary variables" ) 5 dependent variables.Now my problem is when I'm running the logistic regression by using SAS-Eminer it doesn't give me a good predicted probability but if I took that segmented variable(coded as 1-5) as an independent variable including with other 16 variables then it gives me a better result.
I mean if I take one binary variable as an dependent variable and 17 variables(16 var from census + 1 segmented 'coded as 1-5' variable) then it gives me a good predicted probability for the particular binary variable ;so now totally I've got 5 predicted probability.After ranking this 5 predicted probability by using percentile; I've decided that the customer number which has the maximum probability should be fallen in that segment.Now I'm wondering- is it a right decision to include that segment variable as an independent variable.Is it worthless?
Please give me some good suggestion. Please.I've to fit that model based on 1700 sample size.and I've been struggling for 2 weeks to fit that model properly but out of the blue when I've included that variable a good result came-up.Am I doing right/Something seriously wrong?
Regards,
Salam.
====================To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
