topic Re: Interpretation of Logistic Regression Coefficients in Statistical Procedures

Interpretation of Logistic Regression Coefficients

AlainX — Mon, 18 Jul 2022 18:54:36 GMT

Dear All,

I am new to SAS Studio on Demand (and SAS in general), thus I apologise in advance for my silly question.

I run a logistic regression with 100 obs, and 2 independent variables, one categorical ("Card") and one continuous ("Spending"). See the screenshots in the attached PDF.

In Excel/Minitab, I obtain the following coefficients: b0=-2.146, b2=1.099, b3=0.342

In SAS Studio, I obtain the following coefficients: b0=-1.0476, b2=-1.0987, b3=0.3416

May you clarify for me what happens with b0 and b1??? I am really going crazy. My questions:

a) Why the estimates for b0 are different, between Excel/Minitab and SAS?

b) Is it just luck that b0+b1 for SAS is equal to b0 for Excel/Minitab?

b) Why there are opposite signs for b2 (between Excel/Minitab and SAS)?

Once again I deeply apologise for this silly question.

Best,

Alain

Re: Interpretation of Logistic Regression Coefficients

SAS_Rob — Mon, 18 Jul 2022 19:04:53 GMT

It looks like the Excel/Minitab is using a different reference level for Card. SAS is setting the reference level to '1' while it appears the other package sets it to '0'.

Regardless of the values however, the interpretation would be the same.

You can control the reference level on the CLASS statement.

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

Re: Interpretation of Logistic Regression Coefficients

AlainX — Mon, 18 Jul 2022 19:23:18 GMT

Millions, millions, millions of thanks.. I am so obliged. Now I realise the power of SAS communities, and the underlying learning curve is totally worths the efforts. Huge thanks.