[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: My apologies, but... physiology revisited

From: "Mickey P. Rowe" <mrowe@indiana.edu>

 > I did not.  Your objection lost all credibility with me when I looked
 > back at the paper.  I left in your statement including the part that
 > you were objecting to there being only four *birds*, but couldn't
 > believe that that would have bothered you if you'd actually looked at
 > the data.

It does, since the way the study is done assumes an *undemonstrated*
equivalence between mammalian and avian respiratory systems. One simply
cannot blithely lump birds and mammals together just because both are
endothermic, some additional evidence is needed to make such a move

 > > But even a sample size of 12 is still disturbingly small.  To get
 > > really good statistics one wants sample sizes of at least 40-50, and
 > > hundreds is even better.
 > Have you read the methods to find out how much work went into the
 > collection of each data point?

That doesn't matter.  The precision of the data on each taxon does not
alter precision of the *inter*taxon comparisons one whit.

 >  Have you looked at how closely the points fall along their
 > respective regression lines?

That isn't very surprising with such a small sample size.

As another recent post pointed out - just adding a fulmar would break
the endotherm regression line!  With such a small sample, sampling
artifacts are just too easy to get, either accidentally or due to
subconscious bias in sample selection.

 > > Furthermore, unciritcally lumping mammals and birds in a single
 > > regression line rather tends to beg the question. It is simply not
 > > neccessarily the case that mammals and birds fall on the same
 > > regression line.
 > This is precisely why I say I expect better of you.  *LOOK* at the
 > paper.  The birds most certainly do fall on the same regression line.

With only four birds there is *no* *way* *to* *tell* this.  You simply
cannot gert an adquate avian regression line to compare to the mammalian
one with so few birds.

 > Showing that was most likely part of the point of graphically
 > displaying the data.

Insufficient.  They need to perform a *statistical* *test* of
equivalence (and show that with such a small sample the test could
actually show non-equivalence - with so few points getting a p-value
above 95% can be difficult).

 >  The authors expected the data to show the trend
 > they do because they have a functional link between structure and
 > metabolism.

The problem is that they need to *show* there is a functional link.
Just expecting one doesn't cut it.

 >  As I said to Greg, why do *you* propose that the bird
 > data fall on the same regression line as the mammals?

A. sampling artifact


B. the slope is slightly different, but with only four birds the standard
   error of the slope includes the slope of the mammalian line.


B. the correlation is real, but the *variance* about the regression
   line is much higher than their small sample indicates.

 >  Alternatively
 > why are they well above the regression line of the crocodiles and
 > lizards?

A. Sampling artificat

B. the *average* RT size is higher in birds, but they show lots of
   variation not present in the small sample.

 > > Well, we don't really know that the samples were random anyhow. If
 > > there was any selectivity in choosing the organisms to include in
 > > the study, even an unconsious selectivity, this sort of significance
 > > could *easily* be a total artifact.
 > This is a pretty significant charge coming from a person who doesn't
 > appear to have even read the paper.

There is just about nothing they could say that would address this
issue, short of saying they used a random number generator to select
the four birds from a *much* larger list of candidates.  But if they
had such a large list of candidates, why didn't they just use them
all in the study and avoid the whole issue?

Unconscious selectivity is a *major* concern in statistical studies,
and *nobody* is immune. That is why there are standard randomization
procedures that should be used in acquiring any sample, unless the
sample is exhaustive.

Also you left out my *main* argument - the above was just a minor
point that may need to be considered.  My main argument was in regard
to unsample *variance*.  The main problem with the study is that it
does not in any way address the issue of *overlap* in RT size ranges.
It cannot do so with so few sample points (even the full 9+12 is
insufficient for that).

The argument they are trying to make is of the following form:

        1: Endothermy requires (implies) large RT's (E => RT)
        2: Dinosaurs lack large RT's
        Ergo: dinosaurs were not endothermic.

Now, for this to work they have to demonstrate that #1 is true
*without* *exception*.  It is NOT sufficient that it be true on
average.  At best that allows one to conclude "RT => E", which doesn't
help them at all.  Thus merely showing a regression line is not
adequate evidence. Instead, they must provide a robust estimate of the
*standard* *deviation* from the regression line, *and* show that
dinosaurs' RT's fall more than 1.94s below the endotherm regression
line.  You cannot get a robust estimate of s.d. from just 12 points.

 > I think Stan is representative of a
 > lot of people who are much happier thinking of dinosaurs as endotherms
 > a la mammals and birds rather than as anything else.

At present I am not fully decided on this issue, but I do not think
the RT issue settles it one way or the other, at least not yet.

swf@elsegundoca.ncr.com         sarima@ix.netcom.com

The peace of God be with you.