[R] Test for equivalence

[Mike Marchywka]

> From: Greg.Snow at imail.org
> To: mentor_ at gmx.net; r-help at r-project.org
> Date: Sat, 12 Feb 2011 18:04:34 -0700
> Subject: Re: [R] Test for equivalence
>
> Does it make sense for you to combine the 2 data sets and do a 2-way anova with treatment vs. control as one factor and experiment number as the other factor? Then you could test the interaction and treatment number factor to see if they make a difference.


I'm not a statistician and don't play one on TV but I'm not sure if the OP has a specific
approach of hypothesis in mind. I guess it could be a question about an equivalence or 
non-inferiority trial or about some notion of stationary statistics between the two
control and treatment groups ( do list A and B have same E(x^n) for example). More
likely, it sounds like a question about " do A and B appear to be drawn from the same
populations in terms of statistics I care about." So, I guess first I'd just re run
whatever analyses you did with lists A and B but run control vs control and also
treatment vs treatment, pool the results ( A+B combined ) etc. See what that returns
and do sensitivity tests, deleting points moving them a bit etc. Any anova, cox, aft etc
probably wouldn't hurt but hard to know without knowing real issue. 

FWIW, this issue was raised at a recent review of a drug where part of the FDA discussion 
concerned differences
in placebo survival between two studies. someone also earlier mentioned 
"the FDA doesn't accept such and such." In this review of Provenge that was linked to later threats
against some oncologists ( presumably disgruntled DNDN stock speculators LOL).
many types of information are considered including post hoc analysis.
Now "doesn't accept" doesn't mean they will "refuse to file" a BLA that uses
some analysis but this panel anyway was quite open to considering all the information
they had probably more so than the public ( stockholders LOL) that was often just quoting
some isolated statistics, 

http://www.fda.gov/ohrms/dockets/ac/07/transcripts/2007-4291T1.pdf

( you can find the info presented by DNDN in their briefing and the
responses, this is just a transcript of meeting)

This was probably so bizarre to a lot of people because the panel voted solidly 
that they thought the drug was effective
but the FDA rejected the thing largely due to efficacy concerns. 
Their "vote" was on a question that forced 
a bit of an unfortunate choice and it was easy to see how the descrepancy occured.
And in the final analysis the FDA question is, " should the sponsor be allowed to
collect money for claiming they have a drug to treat this condition." I'm not
citing this as a case of "how statistics should be done" by any means, just that it
is an interesting recent case of how it is done in "real life."




>
> --
> Gregory (Greg) L. Snow Ph.D.
> Statistical Data Center
> Intermountain Healthcare
> greg.snow at imail.org
> 801.408.8111
>
>
> > -----Original Message-----
> > From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
> > project.org] On Behalf Of syrvn
> > Sent: Saturday, February 12, 2011 7:30 AM
> > To: r-help at r-project.org
> > Subject: [R] Test for equivalence
> >
> >
> > Hi!
> >
> > is there a way in R to check whether the outcome of two different
> > experiments is statistically distinguishable or indistinguishable? More
> > preciously, I used the wilcoxon test to determine the differences
> > between
> > controls and treated subjects for two different experiments. Now I
> > would
> > like to check whether the two lists of analytes obtained are
> > statistically
> > distinguishable or indistinguishable
> >
> > I tried to use a equivalence test from the 'equivalence' package in R
> > but it
> > seems that this test is not applicable to my problem. The test in the
> > 'equivalence' package just determines similarity between two conditions
> > but
> > I need to compare the outcome of two different experiments.
> >
> > My experiments are constructed as follows:
> >
> > Exp1:
> > 8 control samples
> > 8 treated samples
> > -> determine significantly changes (List A)
> >
> > Exp2:
> > 8 control samples
> > 8 treated samples
> > -> determine significantly changes (List B)
> >
> >
> > Now i would like to check whether List A and List B are distinguishable
> > or
> > indistinguishable.
> >
> > Any advice is very much appreciated!
> >
> > Best,
> > beginner
> > --
> > View this message in context: http://r.789695.n4.nabble.com/Test-for-
> > equivalence-tp3302739p3302739.html
> > Sent from the R help mailing list archive at Nabble.com.
> >
> > ______________________________________________
> > 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.
>
> ______________________________________________
> 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.