topic Including categorical and grouped numerical variables in logistic regression in Statistical Procedures

Including categorical and grouped numerical variables in logistic regression

pdhara — Fri, 07 Dec 2018 12:48:58 GMT

I have categorical variables (Education and Gender) and numerical variables (Age , Income and No.of policies). Age and Income should be grouped as below. I want to run a logistic model on the below data. Which of the following codes is correct? So basically my question is ,should the categorical variables education and gender , binned numeric variables (age and income) be included as class variables explicitly (code 2) or can all of these variables be put together as done in code 1?

Target	Age	Income	Education	Gender	Policy_count
1	26	500000	Graduate	F	2
0	38	300000	Graduate	M	4
0	42	1000000	Post Graduate	F	3
0	68	2200000	Post Graduate	M	5
0	18	65000	12th	M	1
0	71	3500000	Post Graduate	M	2
1	40	2400000	Post Graduate	M	2
1	43	5000000	Post Graduate	M	1
1	52	7000000	Post Graduate	M	5
0	61	10000000	PHD	F	7
0	33	650000	Graduate	M	3
0	14	80000	10th	M	1
0	58	200000	Graduate	M	4

Age_Group	Income_group
<20	<100000
20-30	100000-500000
30-40	500000-1000000
40-50	1000000-2000000
50-60	>2000000
>60

/*Code 1:*/

 

proc logistic data=test descending

plots(only)=(roc(id=obs) effect)  PLOTS(MAXPOINTS=NONE)

namelen=34 outmodel=Logistic_Result;

model target=

Age_Group

Income_group

Policy_count

Gender

Education

/ selection=stepwise

  slentry=0.05

  slstay=0.05

 

outroc = ROC_Stats

lackfit rsq stb;

output out =pred p=phat ;

run;

 

 

/*Code 2:*/

 

PROC LOGISTIC DATA=test

Namelen=34  PLOTS(ONLY)=ALL;

CLASS  age_group (PARAM=EFFECT)  income_group (PARAM=EFFECT)  Gender (PARAM=EFFECT)    Education (PARAM=EFFECT)

Model target (event=’1’)=  Policy_count

/          SELECTION=STEPWISE

                                SLE=0.05

                                SLS=0.05

                                LACKFIT

                                LINK=LOGIT

                                CLPARM=WALD

                                CLODDS=WALD

                                ALPHA=0.05;

RUN;

Re: Including categorical and grouped numerical variables in logistic regression

Ksharp — Fri, 07 Dec 2018 13:26:32 GMT

Only Education and Gender need to be included in CLASS statement.

proc logistic data=test descending

plots(only)=(roc(id=obs) effect)  PLOTS(MAXPOINTS=NONE)

namelen=34 outmodel=Logistic_Result;


class age education;
model target=

Age_Group

Income_group

Policy_count

Gender

Education

/ selection=stepwise

  slentry=0.05

  slstay=0.05

 

outroc = ROC_Stats

lackfit rsq stb;

output out =pred p=phat ;

run;

Re: Including categorical and grouped numerical variables in logistic regression

sld — Sat, 08 Dec 2018 21:09:49 GMT

Neither Code 1 or Code 2 is correct. You need to include Age_group, Income_group, Education, and Gender in the CLASS statement (as in Code 2) as well as in the MODEL statement (Code 1).

The default is PARAM=EFFECT, so you do not need to include that specification unless you just want to.

I hope this helps.