R help - Oct 2006 - logistic regression with a sample missing subjects with a value of an independent variable

Dear R-help,
I am trying to make logistic regression analysis using the R function
"glm", with the parameter family set to binomial, in order to use a
logistic regression model.
I have 70 samples. The dependent variables has two levels (0 and 1) and
one of the independent variables has too two levels (0 and 1). 
The variables associate in the way shown in the table:

            Dependent   0       1
Independent 0          55      10

            1           0       5

This gives a strong association evaluated by the fisher test (p-value
0.0002481), but with the logistic regression it gives a p-value of 0.99
with very high values of estimate and standard error (respectively and
-19.27 and 1769.26).

Is there any way (other function, different setting of a parameter) to
perform logistic regression analysis with these data with R?

Thank you.

Gabriele Stocco
University of Trieste

On Sat, 21 Oct 2006, Gabriele Stocco wrote:
> Dear R-help,
> I am trying to make logistic regression analysis using the R function
> "glm", with the parameter family set to binomial, in order to use
a
> logistic regression model.
> I have 70 samples. The dependent variables has two levels (0 and 1) and
> one of the independent variables has too two levels (0 and 1).
> The variables associate in the way shown in the table:
>
>            Dependent   0       1
> Independent 0          55      10
>
>            1           0       5
>
> This gives a strong association evaluated by the fisher test (p-value >
0.0002481), but with the logistic regression it gives a p-value of 0.99
> with very high values of estimate and standard error (respectively and
> -19.27 and 1769.26).
Please see the comment at the bottom of this message, as your claims are 
not supported by any code.
> Is there any way (other function, different setting of a parameter) to
> perform logistic regression analysis with these data with R?
fit <- glm(matrix(c(55,0,10,5), 2, 2) ~ factor(c(0,1)), binomial())
fit0 <- glm(matrix(c(55,0,10,5), 2, 2) ~ 1,  binomial())
anova(fit0, fit, test="Chisq")

   Resid. Df Resid. Dev Df Deviance P(>|Chi|)
1         1     16.929
2         0  2.208e-10  1   16.929 3.880e-05

is a reasonable way to do this.  Beware the Hauck-Donner phenomenon (see 
e.g. MASS, the book) for t-tests of coefficients, although I do not get 
the values you quote.  Since the expected values are low, you should not 
take the p value too seriously.
> Thank you.
>
> Gabriele Stocco
> University of Trieste
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595