[Fwd: Precision]

[Tim Bergsma]

David,

Below is a fragment of the original conversation.  "Precision" is one of
those terms fraught with user bias.  That's okay, as long as we say what
we mean in the context of EML.  Peter's intention was that eml-precision
is defined operationally:  it specifies the smallest interval that the
instrument (some entity capable of quantifying) can distinguish.  That's
a great piece of metadata, and it's fine by me that we call it
"precision".  Precision in the classical sense, or any other sense for
that matter, gets bumped into the "accuracy" category, where it can be
independently defined.

I like the operational definition of precision, because it solves the
problem of what happens to precision under format conversions.  For
example, if my rain gauge measures to the nearest quarter inch, but my
web page reports in decimal inches, 1.25 inches of rain gives the false
impression that my rain gauge measures to the nearest 0.01 inch.  Not a
problem if operational precision is explicitly stated:  0.25 inches.

As for how to handle missing precision data...this is a bit of a
quandary.  Establishing a default "missing value" has the same logical
effect as allowing an empty element, except that it forces the provider
to confront the fact that the metadata is deficient.  That said, I think
zero is a nice lie, since it is always valid but never correct.  :)

Tim.

-------- Original Message --------
From: Tim Bergsma <tbergsma at kbs.msu.edu>
Subject: Precision
To: Peter McCartney <peter.mccartney at asu.edu>
CC: "Eml-Dev (E-mail)" <eml-dev at ecoinformatics.org>

Okay Peter.  I understand the concept you're talking about, I agree that
it is an important one, and I agree that Precision is an acceptable term
for it.  Precision is also used in two other ways, neither of which are
synonymous with accuracy.  In computing environments, precision can mean
"least significant digit".  I believe Java uses it thus.  Your use of
precision is a superior superset of that usage, because it covers -- but
is not limited to -- a decimal frame of reference.  The classic use of
precision is to indicate repeatability, in contrast to accuracy which
indicates general correctness.  If my arrows always hit 3 inches left of
the bulls eye, I have great precision, but limited accuracy. Accuracy is
deviation of the sample mean from the true value.  Precision is related
to variance within the sample population. 

So,  what do we recommend to eml users?  put classic accuracy and
classic precision in "accuracy" and distinguish them with qualifying
text?  To my chagrin:  I bought a laser straight line tool last week,
that made claims about "accuracy" at 30 m, translated into Spanish and
French as something like "precision".

Tim.

> Peter McCartney wrote:
> 
> There is meaning to the classification accuracy of nominal and ordinal
> values.
> 
> i think standard deviation of the average of values is a really a
> statement about accuracy not precision. precision is used to indicate
> when the miminmum value distinction being reported is not in integrals
> of the units - that is, measuring meters to the nearest tenth, or an
> instrument that returns degrees, but only in intervals of 5 degrees.
> its a qualifier to units in my opinion. in the first case, i could
> have reported the units as decimeters with a precision of 1. in the
> second case, i dont have a unit definition for 5 degrees, so the best
> i can do is say its in degrees in increments of 5.
> 
> -----Original Message-----
> From: Tim Bergsma [mailto:tbergsma at kbs.msu.edu]
> Sent: Thursday, October 31, 2002 10:13 AM
> To: Eml-Dev (E-mail)
> Subject: comments on attribute.xsd
> 
> Hi.
> 
> I just read thru eml-attribute.xsd.
> 
> Typos:
> canbe
> which specify ordered...  (specifies?)
> domain violoations
> character (in the context of missingValueCode, perhaps character(s))
> thatthe
> 
> Comments:
> It looks like all attributes have accuracy, regardless of scale.
> However, only interval and ratio scales have precision.  That's okay
> with me, if there is meaning to the accuracy of categorical values.
> 
> However, precision needs work.  We agreed that precision is a
> statistical thing, and is pretty useless without a descriptive
> interpretation.  Accuracy has a place to describe the test that
> produced
> the value, but precision doesn't have anything equivalent.  This is
> bad.  Consider, for example, that better precisions usually have lower
> 
> numeric values!  There's got to be a way to describe what the
> precision
> means.  (Often it is the standard deviation of repeated measurements.)
> 
> Tim.
> --
> Tim Bergsma
> LTER Information Manager
> W.K. Kellogg Biological Station
> Michigan State University
> Hickory Corners, MI   49060
> 616/671-2337
> tbergsma at kbs.msu.edu
> http://lter.kbs.msu.edu
> _______________________________________________
> eml-dev mailing list
> eml-dev at ecoinformatics.org
> http://www.ecoinformatics.org/mailman/listinfo/eml-dev

-- 
Tim Bergsma
LTER Information Manager
W.K. Kellogg Biological Station
Michigan State University
Hickory Corners, MI   49060
616/671-2337
tbergsma at kbs.msu.edu
http://lter.kbs.msu.edu
_______________________________________________
eml-dev mailing list
eml-dev at ecoinformatics.org
http://www.ecoinformatics.org/mailman/listinfo/eml-dev