topic Re: Proc Logistic Output in SAS Programming

Proc Logistic Output

manonlyn — Thu, 10 Oct 2019 10:23:01 GMT

I am running Proc Logistic in SAS with Stepwise selection. My results show the Intercept with an estimate of -3290.3 while the other parameters vary with values such as 0.5337, 3.11E-6, 0.000088 etc.

This is the first time I've had such differing values, would anyone be able to explain in layman's terms what this means? And possibly what fixes I could do? Or maybe more things to look into?

Unfortunately, I am the only analyst within the business so I have no support from others who work from me as they don't understand the methodology.

Thanks in advance!

Re: Proc Logistic Output

PaigeMiller — Thu, 10 Oct 2019 10:46:54 GMT

Intercept is where the regression line crosses the y-axis when all the x-variables are set to zero. The other numbers are slopes of the individual variables.

I don't know what there is to fix here, it's not clear that anything needs to be fixed, it's not clear that anything is wrong.

Have you checked for outliers in the x-variables? .

Re: Proc Logistic Output

manonlyn — Thu, 10 Oct 2019 10:52:21 GMT

Sorry it's just in the past my Intercept value has been more similar to the parameter estimates, this is the first time they've been extremely different and I was wondering if I needed to fix something.

Are the outliers outputted as part of Proc Logistic?

It might be worth noting that I'm just building quick scorecard for Segmentation Analysis before I start the actual scorecard builds.

Re: Proc Logistic Output

PaigeMiller — Thu, 10 Oct 2019 11:02:31 GMT

@manonlyn wrote:

Sorry it's just in the past my Intercept value has been more similar to the parameter estimates, this is the first time they've been extremely different and I was wondering if I needed to fix something.

Are the outliers outputted as part of Proc Logistic?

It might be worth noting that I'm just building quick scorecard for Segmentation Analysis before I start the actual scorecard builds.

You can check outliers in the x-variables via a histogram or PROC UNIVARIATE, or just about any other plotting method. There are also a number of "goodness of fit" statistics computed by PROC LOGISTIC, such as the C statistic (area under the curve), percent concordance and Somer's D.

Re: Proc Logistic Output

manonlyn — Thu, 10 Oct 2019 11:04:13 GMT

Thank you I'll start with those, this has been really helpful thanks again.

Re: Proc Logistic Output

Reeza — Thu, 10 Oct 2019 13:46:33 GMT

Sorry it's just in the past my Intercept value has been more similar to the parameter estimates, this is the first time they've been extremely different and I was wondering if I needed to fix something.

That may have been coincidental. I'm not aware of any rules or logic that would indicate that should happen.

Re: Proc Logistic Output

Ksharp — Fri, 11 Oct 2019 12:27:50 GMT

I didn't see any problem in your output, so what is your problem .

For those variables have very small coefficient (3.11E-6, 0.000088 etc) , you can ignore/remove them due to not significant .

Re: Proc Logistic Output

PaigeMiller — Fri, 11 Oct 2019 12:29:34 GMT

@Ksharp wrote:

I didn't see any problem in your output, so what is your problem .

For those variables have very small coefficient (3.11E-6, 0.000088 etc) , you can ignore/remove them due to not significant .

But they could be statistically significant, the statistical significance of these coefficients wasn't mentioned.

Re: Proc Logistic Output

Ksharp — Fri, 11 Oct 2019 12:37:49 GMT

I guess these P values must be >0.05 . And hope OP post these P values and confirm my guess .

Re: Proc Logistic Output

manonlyn — Fri, 11 Oct 2019 12:41:38 GMT

The P value for the variable with the estimate of 3.11E-6 is <.0001

Re: Proc Logistic Output

Ksharp — Sat, 12 Oct 2019 11:16:44 GMT

Can you post the whole parameter estimator table ?

Re: Proc Logistic Output

Reeza — Sat, 12 Oct 2019 19:10:19 GMT

It's often considered good practice to standardize your variables in some manner otherwise larger variables can seem more relevant just by the size. Did you standardize your variables ahead of time? Otherwise parameter estimates will just reflect the estimates.

Re: Proc Logistic Output

manonlyn — Mon, 14 Oct 2019 07:53:46 GMT

At the moment I'm building quick models to analyse if it's worth segmenting. I'm not sure how I would standardize variables? What does this usually include? Thank you.

Re: Proc Logistic Output

manonlyn — Mon, 14 Oct 2019 07:56:55 GMT

Analysis of Maximum Likelihood Estimates
Parameter		DF	Estimate	Standard Error	Wald Chi-Square	Pr > ChiSq
Intercept		1	-3290.3	703.3	21.8867	<.0001
Var1		1	-0.00163	0.000226	52.2701	<.0001
Var2		1	-0.00526	0.000466	127.441	<.0001
Var3		1	0.000958	0.00027	12.608	0.0004
Var4		1	-0.0001	0.000038	7.1978	0.0073
Var5		1	-0.3936	0.0369	113.85	<.0001
Var6		1	0.000088	0.000016	31.0853	<.0001
Var7		1	-0.002	0.000298	45.2192	<.0001
Var8		1	0.0126	0.00317	15.871	<.0001
Var9		1	-0.00046	0.000152	9.2876	0.0023
Var10		1	-0.00307	0.000654	22.0835	<.0001
Var11	Y	1	-0.4135	0.0961	18.5191	<.0001
Var12		1	3.11E-06	7.86E-07	15.6228	<.0001
Var13		1	0.000163	0.000035	21.8936	<.0001
Var14	HI	1	-0.276	0.0988	7.8036	0.0052
Var14	LO	1	0.0772	0.0714	1.1689	0.2796
Var14	VH	1	-0.4861	0.0891	29.767	<.0001
Var14	VL	1	0.5337	0.1432	13.8923	0.0002
Var15		1	0.0193	0.00414	21.7022	<.0001
Var16		1	0.00149	0.000291	26.3576	<.0001
Var17		1	-0.2849	0.0557	26.1488	<.0001
Var18		1	0.000088	0.000018	23.3451	<.0001
Var19		1	0.00291	0.000628	21.5079	<.0001

Re: Proc Logistic Output

PaigeMiller — Mon, 14 Oct 2019 12:55:59 GMT

@manonlyn wrote:

At the moment I'm building quick models to analyse if it's worth segmenting. I'm not sure how I would standardize variables? What does this usually include? Thank you.

Even though @Reeza says it is good practice to standardize the variables, I don't think standardizing is required in any way, and your models are fine without standardizing.

But the answer about should you standardize really depends on what "building quick models to analyse if it's worth segmenting" means, and I don't know what that means, and I don't know how you intend to do the analysis to determine if it's worth segmenting. Please explain further.