[R-sig-Epi] Epidemiology Task View
BXC (Bendix Carstensen)
bxc at steno.dk
Mon Nov 24 17:36:49 CET 2008
David,
I do not think an epidemiology task view should put so much emphasis on a historically outdated procedure as MH, particularly since the reason that it is outdated is the advent of computing facilities. After all, MH is just a first step approximation to a (simple) regression estimate, designed for a pocket calculator. It's all in glm, and much better and simpler there.
It is widely implemented because it is a nice little starting exercise to do if you want to practice constructing R-packages, not because of its relevance. Only code consistency, not relvance or correctness is checked on CRAN.
If details on how the various ROC implentations work should be mentioned we will soon be writing an entire book.
The purpose of a task view is merely to point the reader to those manual pages where she might find useful functions. On the assumption that the reader is sane. And hence in the case of an Epi Task View knows about glms for epidemiology.
Bendix
> -----Original Message-----
> From: David Winsemius [mailto:dwinsemius at comcast.net]
> Sent: 24. november 2008 17:11
> To: BXC (Bendix Carstensen)
> Cc: r-sig-epi at stat.math.ethz.ch
> Subject: Re: [R-sig-Epi] Epidemiology Task View
>
> Thanks, Bendix;
>
> I haven't been involved in teaching R and so have not gone
> looking for stuff I don't use. I think my suggestion to add
> links to functions or packages with ability to create ROC
> curves should be taken in the spirit of responding to Neil's
> request for ideas about this notion:
> "Thus I feel that it would be useful to mention which
> functions within a given package are useful for a particular
> task." My personal hope would be that the ROC producing
> functions which label the cutpoints on the curves by default
> should be given favorable notice. The widespread practice of
> publishing ROC curves which do not have cutpoint labels
> should be deplored.
>
> I just looked at the Distributions Task View (prompted by the
> announcement of a revision). I notice that it first listed
> the facilities in its domain that are part of the base R and
> later had sections for available packages that extended the
> functionality.
>
> If such an organization were chosen, I would nominate
> mantelhaen.test as a base "epi" function of at least
> classical epidemiologic note.
>
> http://sekhon.berkeley.edu/stats/html/mantelhaen.test.html
> (It is interesting that so many other elaborations of that
> function have been written in the epi-packages.)
>
> The coin package deserves mention for its provision of "exact" tests.
>
> In searching further on the extensions or elaborations of the
> Mantel- Haenszel test I also found the package MDD (minimal detectable
> difference) and TSHRC (two stage hazard rate comparison)
> which look worthy of inclusion.
>
> Because their vignettes start out with longitudinal data
> analysis and also include a chapter on meta-analysis as well
> as many worked examples of meat-and-potatoes methods in
> epidemiology, I think the HSAUR package and accompanying
> vignettes could have separate links.
> http://cran.r-project.org/web/packages/HSAUR/
> http://cran.r-project.org/web/packages/HSAUR/vignettes/
>
> Best;
> David Winsemius, MD, MPH
> Heritage Labs
>
>
> On Nov 24, 2008, at 3:20 AM, BXC (Bendix Carstensen) wrote:
>
> > In the Epi package is a function ROC.
> > Bendix
> >
> >> -----Original Message-----
> >> From: r-sig-epi-bounces at stat.math.ethz.ch
> >> [mailto:r-sig-epi-bounces at stat.math.ethz.ch] On Behalf Of
> >> David Winsemius
> >> Sent: 24. november 2008 05:52
> >> To: r-sig-epi at stat.math.ethz.ch
> >> Subject: Re: [R-sig-Epi] Epidemiology Task View
> >>
> >> Dear EpidemiRogists;
> >>
> >> " ... this I now have the following categories with suggested
> >> packages...
> >>
> >> Case-Control : Epi, epiR epibasic, epicalc, epitools Cohort :
> >> Epi epiR, epibasic epicalc epitools Incidence/Mortality :
> >> Demography : demogR "
> >>
> >>
> >> I notice that there were no suggested entries for
> >> Incidence/Mortality.
> >> I would have thought that glm (for Poisson regression),
> >> survival, and Harrell's Design/Hmisc packages would fit
> >> there, although I also saw that there are also plans for a
> >> cross-reference to the Survival Task View. I do not see a
> >> problem in having some overlap with other Task Views at least
> >> in "obvious" slots.
> >>
> >> I have also notice that analysis of ROC curves comes up in
> >> r-help fairly commonly. I am not a big fan of these and do
> >> not have any recommendations to propose as a result, but
> >> maybe some notice of the potential problems could surround
> >> references to packages that offer that desired facility?
> >>
> >> --
> >> David Winsemius, MD
> >> Heritage Labs
> >>
> >> _______________________________________________
> >> R-sig-Epi at stat.math.ethz.ch mailing list
> >> https://stat.ethz.ch/mailman/listinfo/r-sig-epi
> >>
>
>
More information about the R-sig-Epi
mailing list