On Wed, 21 Nov 2001, Lee Sieswerda wrote:
> To summarize the thread I started on Cytel's LogXact:
>
> Thanks to Bill Gould who pointed out that I should have used -blogit- and
> not -logistic- to take the "Cytel challenge". Cytel's ads say (in bold
> letters) that "Only LogXact can fit a logistic regression model to these
> data," which seems like a worthy Stata programming challenge... for some
> smart person. Because Stata is so flexible my first thought is that Stata
> must be able to do the calculation, but that perhaps no one has yet
> programmed that specific application. I personally don't know anything about
> "conditional permutation distributions" or "hybrid network-Monte Carlo
> algorithms". It seems a shame for someone to fork out US$1000 on a
> relatively narrow program like LogXact to do something that could be
> programmed in Stata. Richard Goldstein's response seemed to suggest that
> Stata cannot do what LogXact can do. Joseph Coveney submitted a solution to
> the "Cytel Challenge" using -glm-. Very nice, and very gratifying to see
> that Stata has passed the challenge in every important way. It is
> interesting that the Cytel site says that:
>
> "Of interest is our interpretation of the "age" variable - its p-value is
> .071. Without LogXact 4, we would have been forced to drop Cephalexin from
> the conventional asymptotic model, causing the p-value for "age" to drop to
> .04 and leading to an overestimation of its significance."
>
> With Joseph's solution, he gets almost exactly the same p-value (0.073) as
> LogXact and Cephalexin is not dropped. (All of the other estimates come out
> about the same too). With Bill's solution, Cephalexin was dropped but it
> still managed a p-value of 0.064 (not 0.04 as predicted by Cytel). Also,
> while Stata's -glm- gives an OR of 1.39e+09 for Cephalexin, it doesn't seem
> to me that LogXact's OR of 207 with an upper ci of infinity is any more
> helpful. Therefore, it would seem that LogXact is not the only software
> capable of fitting a logistic model to this data, despite their claim to the
> contrary.
>
> Anyway, thanks to Bill et al for inventing Stata. I use it everyday.
>
> Lee Sieswerda
>
>
> > -----Original Message-----
> > From: Joseph Coveney [SMTP:[email protected]]
> > Sent: Wednesday, November 21, 2001 11:32 AM
> > To: [email protected]
> > Subject: Re: st: cytel challenge
> >
> > Lee Sieswerda wrote:
> > -------------------------------------beginpost----------------------------
> > --
> > Cytel makes LogXact for doing "exact" logistic regression (www.cytel.com).
> > In their ads they have something called the "Cytel Challenge" where they
> > ask
> > people to try to fit a logistic regression model to the following data:
> >
> > Diar AB Age Hosp
> > 0 0 0 0
> > 6 0 0 1
> > 1.9 0 1 0
> > 2.9 0 1 1
> > 100 1 1 1
> >
> > The percentage of patients with diarrhea (Diar) is the outcome and the
> > other
> > three variables are predictors: Antibiotic Use (AB; yes/no),
> > Age(young/old),
> > Hospital Stay(Hosp;short/long). Taking the challenge using the -logistic-
> > in
> > Stata fails to produce a converged model. I really don't know the details
> > of
> > how LogXact manages to fit this model, but my question is: would it not be
> > possible to program Stata to do "exact" logistic regression and be able to
> > fit this model? Or is there something inherently different about Cytel's
> > software that it can accomplish this and Stata cannot?
> > -----------------------------endpost--------------------------------------
> > --
> >
> > How about:
> >
> > --------------begin Cytel.do---------------
> > insheet using http://www.cytel.com/examples/diarrhea.dat
> > generate cou0=totno-diarrhea
> > rename diarrhea cou1
> > reshape long cou, i(cephalex clindomy sex age los) j(diarrhea)
> > glm diarrhea cephalex clindomy sex age los [fweight=cou], irls /*
> > */ jknife1 family(binomial) link(logit) eform nolog
> > exit
> > --------------end Cytel.do----------------------
> >
> > The estimates for the parameter and its confidence interval for the odds
> > ratio for Cephalex are silly, but they're silly in Cytel's solution
> > (www.cytel.com/new.pages/LX.ch2a.html), too. The advertisement claimed to
> > be particularly interested in the p-value for the estimate for age. From
> > the looks of Stata's results and those posted on their website, I'd say
> > Stata matched their results, and not only for age, but for the estimates
> > for
> > the other parameters, as well.
> >
> > Joseph Coveney
> >
> > P.S. Have a look at -permute- and -perm- for a fuller explanation of
> > Stata's flexible capabilities in permutation/randomization/exact tests
> > custom-made for user-chosen test statistics.
> >
> >
> >
> >
> >
> > *
> > * Help is available at
> > * http://www.stata.com/support/statalist/faq
> > * http://www.ats.ucla.edu/stat/stata/
> *
> * Help is available at
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
