topic Re: How does SAS calculate the value of the linear predictor in a weighted logistic regression? in New SAS User

How does SAS calculate the value of the linear predictor in a weighted logistic regression?

adrfinance — Mon, 10 Feb 2020 14:50:45 GMT

I am using a weighted logistic regression of the form:

rsubmit;
proc logistic data=dataset_1 outest=stepwise_ot;
model dependent_variable(event="1") = var1 var2 var3 var4 var5 var6 var7 / selection = stepwise
details expb stb lackfit rsquare;
weight correct_weight;
output out=Output_1 pred=PD Xbeta=logodds p=est_response;
store output_1;
run;
endrsubmit;

for the first row SAS gives me a value of the linear predictor equal to -1.118603402 while when I do the calculations in Excel through a linear relationship of the form intercept + b1*x1 + b2*x2 + ..... with the estimated coefficients I get -1,118603402... Could this be due to a rounding error? How does SAS calculate the value of the Linear Predictor?

Re: How does SAS calculate the value of the linear predictor in a weighted logistic regression?

PaigeMiller — Mon, 10 Feb 2020 14:59:53 GMT

@adrfinance wrote:

for the first row SAS gives me a value of the linear predictor equal to -1.118603402 while when I do the calculations in Excel through a linear relationship of the form intercept + b1*x1 + b2*x2 + ..... with the estimated coefficients I get -1,118603402... Could this be due to a rounding error? How does SAS calculate the value of the Linear Predictor?

I don't see a difference between the two numbers you show (except for the typographical error that one has a comma)

Re: How does SAS calculate the value of the linear predictor in a weighted logistic regression?

mkeintz — Mon, 10 Feb 2020 15:23:13 GMT

You have trailing periods in your posted excel result, suggesting you didn't bother to post more digits.

Just because SAS printed the number you posted (-1.118603402) doesn't mean that is the exact value SAS generated. It only means that the printed coefficient was subject to a specific format. If your proc logistic also saved the estimate in a data set, read the dataset and print it with a longer format.

It's likely rounding, yes, but not a rounding ERROR. It's just a display/printing convention that you can investigate.