Testing Significance of coefficients Usually, first examination of model Does the model including the independent variable provide significantly more information than a model without the variable?
Compare two models Model containing x Null (Intercept only model)
The null (intercept only) model proc freq data=s5238.chdage; tables chd; run; proc logistic data=s5238.chdage descending; model chd=; run;
Simple Linear Regression title "Height and FEV in 12 boys, years old"; data fev1ht; input height fev label fev="Forced Expiratory Volume (liters)" height="Height (cm)"; datalines; ; proc sql; select * from fev1ht order by height ; quit; title;
Simple linear regression, proc reg proc reg data=fev1ht plots=none; model fev=height; quit;
Two Equivalent Tests
Two tests (no longer equivalent -- asymptotically equivalent) Wald Test Likelihood Ratio Test
Wald Test (two equivalent versions)
ods select parameterestimates; proc logistic data=s5238.chdage descending; model chd=age; run;
The Likelihood Ratio Statistic Software usually report either log(likelihood) or -2log(Likelihood) SAS reports -2log(Likelihood) (deviance)
ods select globaltests parameterestimates; proc logistic data=s5238.chdage ; model chd(event="1")=age; run;