s-news
[Top] [All Lists]

Re: glm.nb, t-values and p-values

To: <msc0235@geo.ed.ac.uk>
Subject: Re: glm.nb, t-values and p-values
From: <ripley@stats.ox.ac.uk>
Date: Tue, 27 Aug 2002 12:15:30 +0100 (BST)
Cc: <s-news@wubios.wustl.edu>
In-reply-to: <002f01c24db7$869dd600$7b54d781@geo.ed.ac.uk>
On Tue, 27 Aug 2002 msc0235@geo.ed.ac.uk wrote:

> Dear all,
>
> apologies if this is a very trivial question :-
>
> I have been using the glm.nb procedure, a typical output of which is shown
> below. the model returns t-values for each explanatory variable, but I am
> interested in the p.values. Is that just to be compared with a t-tables (and
> would there be a way to automate that in Splus?), or is there anything more
> subtle to do ?

Same problem as summary.glm.  The t-ratios are not t-distributed.  They
are asymptotically normally distributed, so you can use normal tables as a
rough guide.  But *only* a rough guide.  If you want to be more accurate,
use anova after dropping terms, as in the examples in MASS (the book).



>
> Many thanks!
>
> Helene
>
>
> Call: glm.nb(formula = nymphs ~ y.coord + aspect + code + heatherheight,
> data = splitdata, maxit = 50, link = log)
> Deviance Residuals:
>        Min         1Q     Median         3Q      Max
>  -1.080031 -0.7430399 -0.6535381 -0.4116674 2.537494
>
> Coefficients:
>                           Value                Std. Error
> t value
>   (Intercept)     -3.0162209384     0.58795492675         -5.1300207
>       y.coord      0.0002604797     0.00009523648          2.7350836
>        aspect      0.0017502784         0.00111105168      1.5753348
>         code1      0.0713570397     0.19279615354          0.3701165
>         code2     -0.2234691871     0.12541783923         -1.7817975
> heatherheight     0.0398734732     0.01703626392          2.3405057
>
> (Dispersion Parameter for Negative Binomial family taken to be 1 )
>
>     Null Deviance: 252.0698 on 358 degrees of freedom
>
> Residual Deviance: 233.3688 on 353 degrees of freedom
>
> Number of Fisher Scoring Iterations: 1
>
> Correlation of Coefficients:
>               (Intercept)    y.coord     aspect      code1      code2
>       y.coord -0.5715094
>        aspect -0.6128368  -0.0672375
>         code1  0.5101737  -0.1430343 -0.2791874
>         code2  0.5821637  -0.0920575 -0.3402160  0.5749168
> heatherheight -0.7844883   0.2159454  0.3703342 -0.6709585 -0.7514616
>
>               Theta:  0.685
>           Std. Err.:  0.212
>
>  2 x log-likelihood:  -520.199
>
>
>
>
>
> --------------------------------------------------------------------
> This message was distributed by s-news@lists.biostat.wustl.edu.  To
> unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with
> the BODY of the message:  unsubscribe s-news
>

-- 
Brian D. Ripley,                  ripley@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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


<Prev in Thread] Current Thread [Next in Thread>