Re: (PR#8824) wishlist: summary for regression models to report

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: (PR#8824) wishlist: summary for regression models to report

Prof Brian Ripley
  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--27464147-1097610713-1146500823=:15100
Content-Type: TEXT/PLAIN; CHARSET=ISO-8859-1; FORMAT=flowed
Content-Transfer-Encoding: 8BIT
Content-ID: <[hidden email]>

You are apparently unaware that 'na.action' is an argument to lm (which
defaults to the value of an option), and need not only take the value
'na.omit'.  Your request does not make sense for other possible
'na.action's except 'na.exclude' (for example those which impute).

However, there is already a mechanism provided for giving a suitable
message, naprint(), so all you need to do is to instruct the audience of
your sermon how to make use of it.  You might do better to tell them to
use na.action=na.fail (as S does and as I advise my students to do), and
perhaps also to discuss developments in the field in the last 50 years.

For 2.4.0-to-be I have added a line of output in the print.summary.[g]lm
methods that will add to the information on degrees of freedom based on
naprint().

On Mon, 1 May 2006, [hidden email] wrote:

> Full_Name: Ulrike Grömping
> Version: 2.3.0
> OS: Windows
> Submission from: (NULL) (84.190.150.205)
>
>
> Whenever any observations are excluded from a regression analysis (lm,
> glm, and other similar procedures) because of missing values, I would
> find it very useful if this fact is directly visible from the output. I
> think that the information should not only be available (I can e.g. look
> at length of the na.action element of the lm object) but that a serious
> statistical software should draw users' attention to the fact that
> observations have been excluded.

R is nothing like so dictatorial, but does already provide the tools for
this viewpoint (as well as for others).

> For convenience, it would also be nice in general if the number of
> observations used in the analysis is indicated (for lm it is of course
> possible but a bit awkward to find this number in case of many
> parameters).

(It is in fact very easy to find: see e.g. the code for logLik.lm.)

> I hope that this will be implemented because it is quite easy to do (as
> far as I can see). It would make it easier for students and applied
> researchers to comply with my preaching to always report on the number
> of valid observations and the portion of values excluded for
> missingness.

If you want something added to R, please be prepared to contribute a patch
for it. (I believe you could have learned a lot from doing so, including
about what is already provided.)

--
Brian D. Ripley,                  [hidden email]
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
--27464147-1097610713-1146500823=:15100--

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel