Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

Re: The air track experiment:!Cooked Data



At 11:26 8/26/99 -0400, you wrote:
>>... Eddington had to be
>>very selective (just drop off that anomalous data point!)

>... I suspect that it happens a
>lot more than we are willing to admit. Most of us, I'm sure, have, on
>looking at a data point that just looks awful (i.e., doesn't fit with
>the rest of the data) have at least been tempted to say, "That can't
>be right. I'll just ignore that point and use at the rest."....

>Hugh Haskell

I exclude the bulk of this splendid review to note that the exclusion of
outriders is not a sneaky-pete issue, but an element of a most
respectable way of handling experimental data statistically.

It can be shown that the principal result of using median like values
instead of mean like values is to require rather more numbers in the
sample. Justifications are available for excluding rather large proportions
of data points at the skirts in rigorous analyses.
Needless to say, there is a right way.



brian whatcott <inet@intellisys.net>
Altus OK

I agree entirely. What I had intended to say, but didn't very well,
was that the exclusion of outliers should only be done after careful
consideration of the situation. If one excludes a data point or
points just because it or they don't fit into what you want to have,
then you run a real risk of missing the important part of your
experiment. Most of the time you won't, of course, but it would be
tragic to miss an important accidental discovery because one didn't
see the importance of an outlier.

I believe it was Pais, in his excellent history of modern physics,
"Inward Bound," who told about the American physicist who was doing
some work along the same lines as Becquerel (or was it Roentgen--my
copy of the book is at home and I'm not) obtained a set of fogged
photographic plates under much the same circumstances as his European
counterpart, and, thinking that they were ruined, discarded them. All
this happened in 1894 or early 1895, before the discovery he missed
was recognized. This isn't quite the same as discarding an actual
data point, but I'll bet he kicked himself a lot for not following up
on that "outlier."

Of course, throwing out all outliers and keeping them all are sins of
equal magnitude. If we know that an outlier resulted from a known
problem with that particular iteration of the experiment, then get
rid of it. If we don't know this, however, we should investigate what
happened to see if it can be reproduced and under what conditions.
Then we know whether to take it seriously or not.

The "monopole event" that Blas Cabrera recorded on his magnetic
monopole detection apparatus has never been adequately explained, but
has never been repeated. It is widely believed to have been some
anomalous event, perhaps even a prank, but we don't know for sure.
This may be the only recorded evidence for the existence of what
seems to be a much rarer object than the theory calls for, or it may
be just an outlier. But the possibility is there.

When an event that is very rare or difficult to detect is being
sought (gravity waves, proton decay), one must apply fairly strict
standards before accepting an event as real. Most events will be
considered outliers. When the effect sought is easily detected, the
presence of outliers in the data is correspondingly easier to
determine. This is the scientific corollary to the skeptic's motto:
"Extraordinary claims require extraordinary evidence."

Hugh


Hugh Haskell
<mailto://hhaskell@mindspring.com>

Let's face it. People use a Mac because they want to, Windows because they
have to..
******************************************************