Why is it the list always talks about the fun topics when I'm not
watching closely? Now you went and had a discussion about statistics
without me. No fair. As another statistician on this list, let me offer
my 2 cents on sampling. It gets around to Zuikos eventually, so buckle in
and hang on. As has been mentioned earlier, a representative sample is
almost always considered to be a *random sample,* the holy grail of
statisticians everywhere. A random sample is nothing more than a subset
(of the entire population of whatever you're studying) where every
subject or data point has *an equal chance of being included in the
sample*. If this rule isn't followed, you can still get interesting and
informative data, but data of which one must be cautious in a variety of
ways. (Kudos to Chris O'Neill, who has done a good job of educating us
about this limitation in the sampling of auctions he tracks.) The
ability to generalize gets more important, of course, as the stakes of
the study get higher. For example, there was a medical study done of
aging couples where the admitting nurse was told to randomly select
either the husband or the wife into either the experimental or the
control group. The nurse said, "OK, whoever comes through the door first
goes in the experimental group, and whoever comes through the door second
goes in the control group. Since I have no idea of who's likely to come
through the door first or second, that's random." Unfortunately, this
expensive study was scuttled because of her non-random selection
strategy, because it allowed the possibility of a biased sample.
Specifically, researchers were worried that: 1) the healthy spouse made
it through the door first (since the office was on a second story with
access by stairs), meaning that all the healthy spouses ended up in the
experimental group (which would allow an alternate explanation for the
treatment effects), or 2) that elderly husbands were likely to open the
door for their wives, meaning that women would be overrepresented in the
experimental group. These are the sorts of dangers that arise with what
is known as "convenience sampling," aka sampling by an easy selection
rule, rather than true random selection, which is difficult and expensive
to do. ? That being said, convenience sampling is wildly popular in the
corporate world because it's, well, convenient. And inexpensive. And
often yields data which, even if you can't rely on it with assurance, is
nonetheless valuable and often better than no data at all. And heck,
regarding OM prices, I'd rather have less-than-statistically-sanctioned
sampling than no sampling at all, so a big thanks from all of us goes to
Skip Williams and Chris O'Neill for their databases, and all the hard
work they put into compiling them. (An a personal thanks to Skip for
putting it in Excel, and to Chris making it available in tab-delimited,
the lingua francas of data). ? OK, so much for sampling. Some
heavily-funded studies (such as the Census) don't sample at all. They go
for the WHOLE POPULATION, a daunting task that bypasses the need for
statistics altogether! You no longer have to estimate, when you have data
for the entire population. This allows for extremely accurate data. But
as I said, getting data from a POPULATION instead of a SAMPLE is a huge,
expensive task, and I doubt anyone on the list would be up to recording
every single OM sale on eBay (unless she or he had a large, healthy
alternate source of funding to do so, such as winning the lottery or
something of that nature). That's why the issue of sampling, and how it's
done, becomes important. ? Regarding rating systems, I can't speak to
being able to rate conditions from pictures and descriptions, that's
beyond my abilities as a hobbyist, but some people seem to do it quite
well. Regarding the inability to interpret certain ratings from pictures
and descriptions, it could be possible to simply mark the data as
questionable when a reasonable certainty regarding a condition doesn't
exist. However, condition is certainly desirable information to have,
especially in cases as Reese states, when the auction value of an item
becomes confounded with the condition: it's likely that late, rare lenses
will be in good shape and common, early bodies in average shape, so here
the average and the item merge, yielding information that's difficult to
interpret when you've got a pristine, early, but common body (like a C10
OM-1) or a late, rare lens that has heavy wear (like a C7 8mm or 180mm).
I have run statistical regressions on Zuiko prices, then subsequently fit
moving-average Lowess curves to the data, and (coming as no surprise to
anyone) the variable of condition has a major impact on the price paid.
If I recall my graphs correctly, the jump from C7 [KEH=UGLY] to C8
[KEH=BGN] is often a big one, and of course the jump from C9 [KEH=Ex] to
C10 [KEH=LN] is usually quite large, too. ? Finally, regarding the
sentiment that "statistics can be used to prove anything," don't I wish!
If that were true I'd have a heckuva lot more publications on my vitae
than I do. I have longed, prayed, and bargained with various deities and
demons to get statistical significance at the p < .05 level on studies,
and when the numbers don't support your hypothesis, you're just out of
luck, usually with months of time and effort lost. (Your best hope in
these cases is to find something ELSE in the study that's interesting and
significant, and comment on that in a post hoc fashion.) Oh, sure, it'd
be easy to pull the wool over the eyes of the non-statistical, but 1)
that's not at all satisfying and 2) that's never your audience for a
scientific paper. You haven't seen savage till you've faced a crowd of
statisticians picking at your methodology! And in finishing this missive,
I only hope that I'm not eviscerated by another, smarter statistician on
the list who can shoot holes in what I've written. Cowering in fear as
Zuikomethodologists across the world prepare stinging rebuttals (but that
is, after all, the rough game of science,) -- Kelton
================================================================
Kelton Rhoads, Ph.D.
University of Southern California kelton@xxxxxxx
================================================================
< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >
|