[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re[2]: Bird Counts and Statistical Validity



Steve,

You are arguing apples and oranges.  I never said that
each and every CBC is unable to spot huge trends (even
a blind hog finds a root once in awhile).  I said the
CBC and GBBC were not statistically valid / hard
scientific data.  This does not mean they are of zero
use at all.  I wouldn't have participated on 5 CBCs if
I had thought that.

By the way - the barriers to being statistically valid
I referred to are NOT obviated by a careful compiler. 


Maybe a good compiler can weed out Tropicbirds and
Wagtails but as Josh pointed out, how can you weed out
several mistakenly identified Hairy Woodpeckers from
actual Downy Woodpeckers by (semi)kooky observers?  Or
Winter Wrens from House Wrens by novices?  Or Chipping
from Field Sparrows?

How can a compiler correct for inconsistent area
coverage?  Case in point:  this past Charleston CBC, I
went to great lengths to get a seaworthy boat to carry
Burton Moore, John Cottingham, Tripp Smith and me
around outer Bull Bay (southern half) and up and down
the the north end of Bull Island (4/5 of a mile
offshore).  This coverage is not usually obtained. 
Since the count circle ends 3/4 of a mile offshore, we
got the sun at our backs while scanning the water/sky
between us and the beach.  I guarantee the 6 Brant we
saw would not have been spotted by a shore party --
the sun would have been too bad and distance too far. 
Continuing this practice would probably turn up Brant
more years than not.   This doesn't mean Brant are
increasing; it means someone spotted the ones there
all along.  How is a count compiler supposed to
eliminate a false trend like that?  
Also: last year and a few before that, one party of
(non-elite) birders tried to cover the whole of Bull
Island while being driven around in a noisy truck.  
This year, our boat dropped off Burton and me; we
walked a majority of the North end.  Our boat and the
USFWS one also dropped folks off at the dock; these
folks split up and went North and south by truck/on
foot, covering sub-areas we decided upon at the
landing.  I'd be willing to bet a lot that this year
the Bull Island parties (2 instead of 1) got a lot
more birds of certain types.  Continue this improved
coverage a few years and one has a false trend --
results say significantly more birds when in reality
there are about the same number.  I don't see how any
count compiler can overcome inconsistent coverage of
count areas.

I won't continue, but re-read my post.  No count
compiler in the world can make studies based on
nonrepresentative samples into statistically valid
ones.  However, good compilers and participants do
help eliminate some of the crazy stuff + turn in more
accurate numbers and make the counts more useful.

Nathan Dias - Charleston, SC


--- Steve Compton <scompton@sc.rr.com> wrote:
> Nathan,
> 
> As a 15 year Breeding Survey participant and a
> Christmas birder for 22 years, and as the former
> compiler
> of the Charleston count, I have to say I agree with
> Charlotte. The "errors"
> you refer to are largely
> obviated by the careful methods under which a
> complier
> must submit the data for a Christmas Count.
> Remember,
> we are looking for population trends, not small
> differences
> in rare or accidental occurances. The BBS certainly
> has
> a more "reliable" methodogy, and there are
> weaknesses
> in certain counts, but I would maintain the
> Charleston
> count data is very reliable for trends of local bird
> popualations over the
> past 40 years. Just test the data, compare the data
> to any other measure and
> you will see.
> It is remarkable how predictable the results for a
> Christmas count can be.
> This for me is a strong indication
> of statistical reliability.
> 
> Nathan, I would recommend you compile a Christmas
> count for a few years and
> test your theories. In any event,
> it is fun to read your opinions and to be with you
> in the field. I look
> forward to that opportunity again.
> 
> Yours for truth and science,
> Steve Compton
> scompton@sc.rr.com
> 
> 
> ----- Original Message -----
> From: "Nathan Dias" <diasn@yahoo.com>
> To: "Charlotte Goedsche" <goedsche@mindspring.com>;
> "carolinabirds"
> <carolinabirds@duke.edu>
> Cc: <amw25@cornell.edu>
> Sent: Tuesday, February 18, 2003 3:41 PM
> Subject: Bird Counts and Statistical Validity
> 
> 
> > Charlotte -
> >
> > Just because so-called "scientific" studies use
> > Christmas count results does not mean they would
> > withstand rigorous evaluation by a statistician.
> >
> > Lookit:
> > When considering studies:  in general, samples are
> > taken and inferences about the overall population
> are
> > made by comparing samples.  If one cannot trust
> the
> > samples themselves, one cannot place much trust in
> the
> > inferences drawn from their comparison.  The name
> of
> > the game in applied statistics is to minimize the
> > uncertainty attached to these inferences.  I shall
> > demonstrate how this cannot be done to an
> acceptable
> > degree with data resulting from Christmas Counts
> and
> > GBBC.  My claim was not irresponsible, Charlotte;
> it
> > is founded in statistical bedrock.
> >
> > Here are some barriers to studies based on GBBC
> and
> > Christmas Count results being statistically valid:
> >
> > - Too many novice/untrained/kooky observers (who
> might
> > misidentify birds).  Both the spatial distribution
> and
> > the number of observers of varying competence is
> > uneven and non-random from one count to the next.
> > STRIKE ONE.  Example:  A phone survey/poll asks
> > respondents questions with cut-and-dry answers. 
> The
> > folks compiling the results "know what they are up
> > against" (they know for sure that the respondent
> said
> > they favored the U.S. going to war with Iraq). 
> But
> > folks using Xmas counts or GBBC results don't know
> for
> > sure if Suzy Dimwit really saw a given Loggerhead
> > Shrike or misidentified a Mockingbird.  This is a
> > hindrance to having a representative sample.
> >
> > - Grossly inaccurate numeric data (see my previous
> > post on double-counting, under-counting,
> > mis-estimation of flocks, etc.).  Everyone knows
> that
> > Christmas count species totals that folks turn in
> are
> > ballpark guesstimations (way too high or too low)
> in
> > many cases.  This numeric data is also
> inconsistent
> > (and non-random) in its inaccuracy.  STRIKE TWO. 
> In
> > and of itself, this doesn't mean a study based on
> the
> > data is not statistically valid, but it does mean
> the
> > uncertainty attached to inferences drawn is so
> high as
> > to make the inferences of much less use.
> >
> > - Uneven/inconsistent coverage of the same
> physical
> > area(s) from year to year.  STRIKE THREE.  This
> can
> > make comparing data from different years like
> > comparing apples to oranges.   NONREPRESENTATIVE
> > SAMPLES result from this practice (a cardinal sin
> if
> > one is seeking statistical validity).  Example:
> many
> > christmas counts have sub-areas that are not
> > "glamorous."  They receive little or no coverage
> some
> > years, half-hearted coverage in others and good
> > coverage by heroic experts in others.  This
> coverage
> > is uneven and nonrandom.  The same argument
> applies to
> > zip codes in the GBBC.  Another uneven coverage
> > problem results from the need for boats to cover
> water
> > areas of Christmas Counts.  Another uneven
> coverage
> > problem (on Christmas Counts) happens when land
> > (especially plantations) gets sold to
> > non-birder-friendly landowners.  Having coverage
> > depend on the whims of a cranky landowner is
> > definitely inconsistent and nonrandom.
> >
> > - Due to the relatively low number of observers
> versus
> > total bird population size, the high number of
> > non-expert observers, inconsistent sample area
> > coverage, inconsistently inaccurate data and many
> > other factors: Xmas Counts and GBBC produce
> > unacceptably high probabilities of alpha error and
> > beta error.  STRIKE FOUR.  -- Alpha error is when
> one
> > sees a statistically significant difference in the
> > samples, when no difference actually exists in the
> > populations.  Beta error is when one does not see
> a
> > statistically significant difference in the
> samples,
> > when a difference actually does exist in the
> > populations.  When probabilities of alpha error
> and
> > beta error are too high, one can throw inferences
> > drawn out the window.
> >
> > I could go on and on with the strikes.  The
> > nonrepresentative sample issue alone means NOT
> > STATISTICAL VALID and it keeps cropping up due to
> > various considerations.  QED.
> >
> > Breeding Bird Atlas studies are much better (in
> terms
> > of statistical validity) than Xmas counts and GBBC
> -
> > they use highly competent observers, identical
> routes
> > (areas of coverage) over time, identical time
> spend
> > traversing the route from year to year, and more.
> >
> > Nathan Dias - Charleston, SC.
> >
> > PS   My father (who has a PhD in Biostatistics)
> also
> > gets red in the face at the thought of folks
> claiming
> > that studies based on Christmas Count and GBBC
> results
> > are statistically valid.  As someone with degrees
> in
> > (pure) math and Computer Science, as well as A
> grades
> > in graduate level courses in Statistics, I agree
> > whole-heartedly.
> 
=== message truncated ===


__________________________________________________
Do you Yahoo!?
Yahoo! Tax Center - forms, calculators, tips, more
http://taxes.yahoo.com/