glmfit - logistic regression

[Conor Walsh]

hi

I am new to this forum and new to statistics in Matlab. I am
analyzing some data and am using glmfit to perform a
logistic regression. I use the command as follows

[b,dev,stats] =
glmfit(lesionsize,totalcomp,'binomial','link','logit')

My x vector is contrinuous and represents the size of
lesions. My y variable is binary and represents whether
there was a complication or not.

In reading the help, I read that matlab automatically
inserts a column of 1's into X. In my case it is just 46*1
matrix and after that I guess it is 46*2. That is why glmfit
returns to Beta values and 2 corresponding p values.

As I am not a stats expert, I was wondering if someone could
help me interpret the results from glmfit (i.e. what each p
value represents). I know there is a way to force matlab to
remove the colum of 1's but I'm not really sure what it
represents and why it is needed.

Thanks in advance

Conor

[Conor Walsh]

[Peter Perkins]

Conor Walsh wrote:

> As I am not a stats expert, I was wondering if someone could
> help me interpret the results from glmfit (i.e. what each p
> value represents). I know there is a way to force matlab to
> remove the colum of 1's but I'm not really sure what it
> represents and why it is needed.

Conor, the constant term makes it so that the expected response at x=0 is something other than exp(0)/(1+exp(0)) = .5. It's exactly analogous to the intercept term in a linear regression.

The p-values are also analogous to the same quantities in linear regression.

You might take a look at the books by Doson or Collett (a good one) in the references; they will give you a lot of good information.

[Conor Walsh]

Thanks Peter

I'll take a look, I guess what is really confusing me is
that when i do a linear regression i only get one p value

for my analysis with glmfit the two p values i get are
0.2608
0.0392

The first is > 0.05 but the second is <0.05. I'm a little
confused as to whether I can say there is a significant
relationship or not

Thanks again for the assistance

conor

Peter Perkins <Peter.Perki...@mathworks.com> wrote
in message <g7i5dh$kd1$1...@fred.mathworks.com>...

[Peter Perkins]

Conor Walsh wrote:
> Thanks Peter
>
> I'll take a look, I guess what is really confusing me is
> that when i do a linear regression i only get one p value
>
> for my analysis with glmfit the two p values i get are
> 0.2608
> 0.0392
>
> The first is > 0.05 but the second is <0.05. I'm a little
> confused as to whether I can say there is a significant
> relationship or not

Ah, OK. In a linear regression, there's an F statistic which is sort of "for the whole fit", but there are also t statistics for each coefficient. GLMs have the same sort of thing, using the (difference in the) deviance. The fact that one of your t stats is small and the other big indicates that maybe you don't need one of the coefs in your model.