0

I am running a logistic regression and I both standardise and normalise my input to ensure that all the independent variables are homogenised.

My friend is telling me that I need to activate the standardised beta coefficients in my regression code (STB in proc logistic, as shown below). Is there really a difference when I have already ensured that my independent variables are on common scale?

I will be re-calibrating my beta coefficients to [0,1] scale where they add up to 1 to act as a score card but I am curious about standardised beta coefficients.

Example code

 DATA baseball;
      SET sashelp.baseball;
      IF logsalary > 6.5 THEN flag = 1;
      ELSE flag = 0;
 RUN;

 PROC STDIZE
      DATA = BASEBALL
      METHOD = RANGE
      OUT = BASEBALL_S
      ;
      VAR CR:;
 RUN;

 PROC LOGISTIC
 DATA = baseball_s  OUTEST=result;
       MODEL  FLAG (event='1')= CR: /expb stb 
 SELECTION=none rsquare OUTROC=roc ;
 RUN;
78282219
  • 21
  • 2
  • 1
    Did you try activating the standardized beta coefficients to see whether there was a difference? – The Laconic Feb 22 '19 at 18:03
  • You don't need to standardize predictors in logistic regression. See https://stats.stackexchange.com/questions/29781/when-conducting-multiple-regression-when-should-you-center-your-predictor-varia and https://stats.stackexchange.com/questions/201909/when-to-normalize-data-in-regression/202002#202002 – kjetil b halvorsen Feb 23 '19 at 11:31
  • @TheLaconic - There is a difference – 78282219 Feb 25 '19 at 08:20
  • @kjetilbhalvorsen, thank you, I'll check these links – 78282219 Feb 25 '19 at 08:20

1 Answers1

0

The standardized beta coefficients are equivalent to the coefficients obtained from the regression with standardized dependent and independent variables. Therefore, standardized coefficients encode the effects of dependent over independent variables in the units of their respective standard deviations. You can check the wikipedia entry.

gsanroma
  • 108
  • 1
  • 9