topic Re: How do I build a logistic model for each categorical value of some categorical variable of a dat in SAS Programming

How do I build a logistic model for each categorical value of some categorical variable of a dataset

junlue — Tue, 12 Apr 2016 16:05:01 GMT

proc logistic data=BTS201506 ; 
class Carrier ; 
model DepDelayInd(Descending) = CRSDepTime seqnum DepDelayLagInd DepDelayLag DepDelayLagCum ArrDelayLagInd ArrDelayLag ArrDelayLagCum DepDelayLag2 ArrDelayLag2;
where cancelled=0 ;
run ;

So basically, Carrier is my categorical variable. I want to partition the dataset based on the values of this categorical variable and then build a logistic regression for each of the values of this categorical variable, Carrier.

But the code above does not work.

Help

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

ballardw — Tue, 12 Apr 2016 16:11:08 GMT

Instead of Class Carrier use By Carrier. This will do a complete separate analysis for each level of Carrier. The data must be sorted by Carrier first though.

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

junlue — Tue, 12 Apr 2016 16:12:54 GMT

But Carrier is a character variable.

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

junlue — Tue, 12 Apr 2016 16:14:56 GMT

When I use By Carrier.

this is error message

ERROR: Variable Carrier should be either numeric or specified in the CLASS statement

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

stat_sas — Tue, 12 Apr 2016 16:23:35 GMT

First sort your data set by Carrier then use the sorted data set in logistic regression with

By Carrier;

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

Reeza — Tue, 12 Apr 2016 16:45:21 GMT

If it doesn't work, please post your code AND log.

You're likely putting something in the wrong place, using a BY statement is the correct answer.

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

junlue — Tue, 12 Apr 2016 17:34:00 GMT

Actually it worked.

My final question is,

It seems that the

"where cancelled = 0"

statement has no effect.

I'm still getting warnings like these in the ouput window:

"Note: 2898 observations were deleted due to missing values for the response or explanatory variables."

But that shouldn't be the case because all the missing values are where cancelled = 1.

By stating where cancelled = 1, i am selecting those observations without missing values.

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

ballardw — Wed, 13 Apr 2016 03:16:32 GMT

@junlue wrote:

Actually it worked.

My final question is,

It seems that the

"where cancelled = 0"

statement has no effect.

I'm still getting warnings like these in the ouput window:

"Note: 2898 observations were deleted due to missing values for the response or explanatory variables."

But that shouldn't be the case because all the missing values are where cancelled = 1.

By stating where cancelled = 1, i am selecting those observations without missing values.

Try Running this code and see what you get:

Proc freq data=BTS201506;

tables cancelled* (DepDelayInd CRSDepTime seqnum DepDelayLagInd DepDelayLag DepDelayLagCum ArrDelayLagInd ArrDelayLag ArrDelayLagCum DepDelayLag2 ArrDelayLag2) / list missing;

run;

See if you have any rows with cancelled=0 and something else missing. Or possibly you really meant to keep cancelled=1??

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

junlue — Wed, 13 Apr 2016 04:51:06 GMT

There was a typo in my previous post.

I meant "where cancelled = 0", not 1.

However, this was a non-issue and you are right. There are indeed still missing values in for rows with the value of 0 for the cancelled variable

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

ballardw — Wed, 13 Apr 2016 15:21:31 GMT

The procedure, as do most of the regressions, excludes any record with a missing value for the variables on the model statement. It may be that you have some mis-coded variables such as missing should have become 0 or similar that was intended but skipped.

Re: How do I build a logistic model for each categorical value of some categorical variable of a dat

Reeza — Wed, 13 Apr 2016 16:12:44 GMT

@junlue it sounds like you have your question answered. Please mark the appropriate solution as the correct answer.