[R] Post-hoc tests in MASS using glm.nb

[Bill.Venables at csiro.au]

Amen to all of that, Bert.  Nicely put.  The google style guide (not perfect, but a thoughtful contribution on these kinds of issues, has avoiding attach() as its very first line.  See http://google-styleguide.googlecode.com/svn/trunk/google-r-style.html)

I would add, though, that not enough people seem yet to be aware of within(...), a companion of with(...) in a way, but used for modifying data frames or other kinds of list objects.  It should be seen as a more flexible replacement for transform() (well, almost).

The difference between with() and within() is as follows:

with(data, expr, ...) 

allows you to evaluate 'expr' with 'data' providing the primary source for variables, and returns *the evaluated expression* as the result.  By contrast

within(data, expr, ...) 

again uses 'data' as the primary source for variables when evaluating 'expr', but now 'expr' is used to modify the varibles in 'data' and returns *the modified data set* as the result.

I use this a lot in the data preparation phase of a project, especially, which is usually the longest, trickiest, most important, but least discussed aspect of any data analysis project.  

Here is a simple example using within() for something you cannot do in one step with transform():

polyData <- within(data.frame(x = runif(500)), {
  x2 <- x^2
  x3 <- x*x2   
  b <- runif(4)
  eta <- cbind(1,x,x2,x3) %*% b   
  y <- eta + rnorm(x, sd = 0.5)  
  rm(b)
})

check:

> str(polyData)
'data.frame':	500 obs. of  5 variables:
 $ x  : num  0.5185 0.185 0.5566 0.2467 0.0178 ...
 $ y  : num [1:500, 1] 1.343 0.888 0.583 0.187 0.855 ...
 $ eta: num [1:500, 1] 1.258 0.788 1.331 0.856 0.63 ...
 $ x3 : num  1.39e-01 6.33e-03 1.72e-01 1.50e-02 5.60e-06 ...
 $ x2 : num  0.268811 0.034224 0.309802 0.060844 0.000315 ...
> 

Bill Venables.

-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of Bert Gunter
Sent: Wednesday, 18 May 2011 12:08 AM
To: Peter Ehlers
Cc: R list
Subject: Re: [R] Post-hoc tests in MASS using glm.nb

Folks:

> Only if the user hasn't yet been introduced to the with() function,
> which is linked to on the ?attach page.
>
> Note also this sentence from the ?attach page:
>  ".... attach can lead to confusion."
>
> I can't remember the last time I needed attach().
>
> Peter Ehlers

Yes. But perhaps it might be useful to flesh this out with a bit of
commentary. To this end, I invite others to correct or clarify the
following.

The potential "confusion" comes from requiring R to search for the
data. There is a rigorous process by which this is done, of course,
but it requires that the runtime environment be consistent with that
process, and the programmer who wrote the code may not have control
over that environment. The usual example is that one has an object
named,say,  "a" in the formula and in the attached data and another
"a" also in the global environment. Then the wrong "a" would be found.
The same thing can happen if another data set gets attached in a
position before the one of interest. (Like Peter, I haven't used
attach() in so long that I don't know whether any warning messages are
issued in such cases).

Using the "data = " argument when available or the with() function
when not avoids this potential confusion and tightly couples the data
to be analyzed with the analysis.

I hope this clarifies the previous posters' comments.

Cheers,
Bert

>
> [... non-germane material snipped ...]
>
> ______________________________________________
> R-help at r-project.org 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.
>



-- 
"Men by nature long to get on to the ultimate truths, and will often
be impatient with elementary studies or fight shy of them. If it were
possible to reach the ultimate truths without the elementary studies
usually prefixed to them, these would not be preparatory studies but
superfluous diversions."

-- Maimonides (1135-1204)

Bert Gunter
Genentech Nonclinical Biostatistics

______________________________________________
R-help at r-project.org 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.