topic Re: How do I tell which dependent variable is the most predictive in Proc Logistic? in Statistical Procedures

How do I tell which dependent variable is the most predictive in Proc Logistic?

LHK — Fri, 17 Nov 2017 19:35:36 GMT

Hello Experts:

I run into a difficulty to explain which independent variable is more predictive when running a Proc Logistic.

In my proc Logistic procedure, I modeled the dependent variable (Event = '1'), see code below:

proc logistic data=data_test;

model p_flag (event = '1') = var1 var2 var3 / selection=stepwise;

run;

Then I get this results for the stepwise selection:

Summary of Stepwise Selection
Step	Effect	DF	Number In	Score Chi-Square	Wald Chi-Square	Pr > ChiSq	Variable Label
Entered	Removed
1	var3		1	1	2006.955		<.0001
2	var2		1	2	150.3232		<.0001
3	var1		1	3	43.6837		<.0001

and the likelihood:

Analysis of Maximum Likelihood Estimates
Parameter	DF	Estimate	Standard Error	Wald Chi-Square	Pr > ChiSq
Intercept	1	-4.1428	0.1142	1316.869	<.0001
var1	1	1.083	0.1556	48.4503	<.0001
var2	1	-0.1256	0.00706	316.7223	<.0001
var3	1	0.2754	0.00463	3541.0916	<.0001

And the odd ratios:

Odds Ratio Estimates
Effect	Point Estimate	95% Wald Confidence Limits
var1	2.954	2.177	4.007
var2	0.882	0.87	0.894
var3	1.317	1.305	1.329

So my question is:

Which one of the 3 variables is the most predictable for p_flag =1? In the likelihood test, which output field I need to look into: estimate vs. Wald Chi-square?

Thanks a lot!

LHK

Re: How do I tell which dependent variable is the most predictive in Proc Logistic?

PaigeMiller — Fri, 17 Nov 2017 22:51:23 GMT

I don't think there is an answer to your question. I don't really even understand what you want when you say "most predictable". It doesn't make sense to me to just use one predictor given these results.

Re: How do I tell which dependent variable is the most predictive in Proc Logistic?

Ksharp — Sat, 18 Nov 2017 10:08:51 GMT

I will pick up Wald Chi-Square, since it is statistic .

And var3 has the biggest Chi-Square ,so its Pr > ChiSq should be smallest. therefore, var3 is most significant variable.

Re: How do I tell which dependent variable is the most predictive in Proc Logistic?

PaigeMiller — Sat, 18 Nov 2017 12:52:16 GMT

@Ksharp wrote:

I will pick up Wald Chi-Square, since it is statistic .

And var3 has the biggest Chi-Square ,so its Pr > ChiSq should be smallest. therefore, var3 is most significant variable.

True, but the question was not "most significant variable", it was "most predictable", which seems to me is a meaningless question, and different than "most significant".

Anyway, the idea of selecting one of these three variables is the "best" seems like a very poor idea.