R help - May 2013 - Fwd: binomial glm and sole effect of the treatment

Hiya,

I'm using simple glm binomial models to test the effect of treatment
(factor, 3 levels) on infection prevalence (infected/uninfected):

ad3<-glm(Infection~ecs, family=binomial, data=eilb)

but summary() function returns for each of the factor-level coefficients
against the control treatment:
> summary(ad3)
Call:
glm(formula = Infection ~ ecs, family = binomial, data = eilb)

Deviance Residuals:
    Min       1Q   Median       3Q      Max
-1.4006  -1.0383  -0.9005   1.3232   1.4823

Coefficients:
            Estimate Std. Error z value Pr(>|z|)
(Intercept)  -0.3365     0.5855  -0.575    0.566
ecsminus2    -0.3567     0.8473  -0.421    0.674
ecsplus2      0.8473     0.9361   0.905    0.365

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 43.860  on 31  degrees of freedom
Residual deviance: 42.162  on 29  degrees of freedom
AIC: 48.162

Number of Fisher Scoring iterations: 4

> str(eilb)
'data.frame':    32 obs. of  59 variables:
(...)
 $ ecs               : Factor w/ 3 levels
"control","minus2",..: 2 3 3 3 2
2 1 1 1 1 ...
 $ Infection         : Factor w/ 2 levels
"Infected","Uninfected": 1 1 1 1
1 1 1 1 1 1 ...


What I want to know is whether the treatment in general had an effect on
infection prevalence, not the difference between respective factor levels.
If it was a general linear model I could switch between using lm() and
aov() functions, but how can I proceed here? I sense I'm missing something
obvious, so I'll appreciate your help!


cheers,
kasia

	[[alternative HTML version deleted]]

Katarzyna Kulma <katarzyna.kulma <at> gmail.com> writes:
> 
> Hiya,
> 
> I'm using simple glm binomial models to test the effect of treatment
> (factor, 3 levels) on infection prevalence (infected/uninfected):
> 
> ad3<-glm(Infection~ecs, family=binomial, data=eilb)
> 
> but summary() function returns for each of the factor-level coefficients
> against the control treatment:
> 
> > summary(ad3)
> 
> 
> What I want to know is whether the treatment in general had an effect on
> infection prevalence, not the difference between respective factor levels.
> If it was a general linear model I could switch between using lm() and
> aov() functions, but how can I proceed here? I sense I'm missing
something
> obvious, so I'll appreciate your help!
> 

  either

ad3B <- update(ad3,.~1)
anova(ad3,ad3B)

or

drop1(ad3,test="Chisq")

should do what you want (a likelihood ratio test between the model
containing the treatment and the model without the treatment).

Bonus, the likelihood ratio tests will be more accurate than
the Wald tests provided by summary().