Logistic Regression Potential Independent Predictors

[csa]

Hi All,

I am using proc logistic procedure to consider to get Independent Predictors "Estimate, StdErr and ProbChiSq". Could you please guide me from procedure output which value we consider as Estimate value and other for statistical analysis.

/****************************************/

Proc logistic data=source;
class mref(ref='0')/param=glm;
model cancer(Event='1')= mref /firth;
ods output ParameterEstimates=parm;
run;

From the below table observation 1 or 2 which needs to considered  to get "Estimate, StdErr and ProbChiSq".

ods output PARM dataset values:

OBS	VARIABLE	CLASSVAL0	DF	ESTIMATE	STDERR	WALDCHISQ	PROBCHISQ	_ESTTYPE_
1	INTERCEPT		1	1.9513	0.3426	25.5141	<.0001	PMLE
2	MREF	1	1	-1.1261	1.1296	0.96	0.3112	PMLE
3	MREF	0	0	0		.	.	PMLE

Thanks

[Ksharp · Accepted solution]

I don't understand your question.
From your output MREF 's PValue is 0.3112 >0.05 ,which means it is not significant . You could drop this variable ,just fit a intercept model :
model cancer(Event='1')= /firth;

[sbxkoenk]

Hello,

"Independent predictors"?

All candidate predictors are independent variables.
I have never heard about "independent predictors", that's a tautology.

Concerning your question:

mref is a class variable to which glm-coding is applied.

class mref(ref='0')/param=glm;

You have to learn about the coding schemes for class-effects : glm , effect coding (=deviation-from-the-mean coding), reference cell coding , ...

See here:

• SAS/STAT 15.3 User's Guide
  The LOGISTIC Procedure
  Example 79.2 Logistic Modeling with Categorical Predictors
  https://go.documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/statug/statug_logistic_examples02.htm
• UCLA
  CODING SYSTEMS FOR CATEGORICAL VARIABLES IN REGRESSION ANALYSIS
  https://stats.oarc.ucla.edu/spss/faq/coding-systems-for-categorical-variables-in-regression-analysis...
  (that's for SPSS, but that doesn't matter much because largely the same in all stat software)

When interpreting coefficients from a class effect, you must know how it was coded (mandatory for proper interpretation)!

If you only have 0 and 1 values for your mref and mref is a variable having the numeric data type, then you might consider to include it as a "continuous" covariate in your linear predictor. Just get rid of the class statement.

Cheers,

Koen