Interpreting glm

["Molinari, Luciano" <Luciano.Molinari@kispi.unizh.ch>]

I am working with SP7 for Windows.

Can anybody help, give some hint, in interpreting, numericaly and statistically, the following printout of glm:

--------------------------------------

S+>zz0_glm(kop~mprae+einviuni+einvizya+kpros1,na="na.exclude",family=binomial)
S+>summary(zz0)

 

Call: glm(formula = kop ~ mprae + einviuni + einvizya + kpros1, family = binomial, na.action = ""> "na.exclude")
Deviance Residuals:
   Min       1Q Median    3Q  Max
 -2.27 0.000722  0.224 0.513 1.45

 

Coefficients:
             Value Std. Error t value      P
(Intercept)  9.416    210.017  0.0448 0.9643            #The P values for the t's
      mprae -0.591      0.336 -1.7582 0.0808            #are calculated in my local version of summary.glm
   einviuni -0.698      0.376 -1.8559 0.0655
   einvizya -0.855      0.417 -2.0482 0.0424
     kpros1  7.887    210.017  0.0376 0.9701

 

(Dispersion Parameter for Binomial family taken to be 1 )

 

    Null Deviance: 102 on 148 degrees of freedom

 

Residual Deviance: 72.1 on 144 degrees of freedom

 

Number of Fisher Scoring Iterations: 15
Warning messages:
      LM: This is my local version of summary.glm!
 (P-values based on t added)
  in: summary(zz0)
S+>anova(zz0,test="Chi")
Analysis of Deviance Table

 

Binomial model

 

Response: kop

 

Terms added sequentially (first to last)
         Df Deviance Resid. Df Resid. Dev Pr(Chi)
    NULL                   148        102       
   mprae  1      7.7       147         94  0.0055
einviuni  1      4.2       146         90  0.0414
einvizya  1      3.7       145         86  0.0529
  kpros1  1     13.9       144         72  0.0002

S+>anova(zz0,test="F")
Analysis of Deviance Table

 

Binomial model

 

Response: kop

 

Terms added sequentially (first to last)
         Df Deviance Resid. Df Resid. Dev F Value  Pr(F)
    NULL                   148        102              
   mprae  1      7.7       147         94    12.1 0.0007
einviuni  1      4.2       146         90     6.5 0.0117
einvizya  1      3.7       145         86     5.9 0.0166
  kpros1  1     13.9       144         72    21.9 0.0000

------------------

All the variables involved are binary. They are relatively unbalanced, so that a number of combinations might not occur at all.

Here my questions:

- why are the some standard errors, see summary, so large?

- which are the correct, if any, P-values for the coefficientas?

- is the difference null.deviace-residual.deviance adequately described by the Chisquare distribution, here, when not?

- Can one confidently believe that variable kpros1 is relevant, or not relevant? There appear to be a contradiction with the t-value of the summary and the reduction in the deviance in the anova(zz0,test="Chi") command.

 

Thanks for helping,

L. Molinari