[LTER-im] measurmentScale/precision - what definition? howtohandle?
Tim Bergsma
tbergsma at kbs.msu.edu
Tue Aug 5 09:16:20 PDT 2003
Matt, others,
I agree with the decision to make precision not required, but I do not
agree with the proposed definition. Two concepts are being confused
here: analytical precision and operational precision. Analytical
precision is statistical, subjectively defined, and cannot be evaluated
unless a measurement has been repeated. Operational precision is
observational, objectively defined, and can be predicated even of a
single measurement.
The proposed definition favors analytical precision. The word "most" is
ambiguous. If standard deviation is provided as a measure of precision
(and if the data are normally distributed) then "most" means about 67%.
I agree that it is overly prescriptive to interpret precision uniformly
as standard deviation.
When drafting EML, we agreed that precision was interpreted
operationally. Thus, it was acceptable to have no element to hold a
description of the test that generated the value. We agreed that
analytical precision should be deferred to the accuracy element.
eml-accuracy is repeatable and has an element for storing a description
of the test.
I propose an alternative definition:
"Precision is the resolution of a method that generates a numerical
observation, and is reported in the same units as the observation
itself. Often it is the smallest interval that some measuring device
can distinguish. For example, if fish lengths are measured with a meter
stick marked at millimeter, centimeter, and decimeter intervals, the
precision of the observations could be 1 millimeter. However, if
observers are instructed to evaluate fish lengths to the nearest
centimeter, then the precision is 1 centimeter, even if the measurement
device has finer divisions. Precision is defined operationally, and
should not be confused with analytical precision, i.e. some estimate of
the variation among repeated observations (see eml-accuracy). Precision
is a statement about the way in which numerical values were generated;
it is somewhat independent of the format in which those values are
stored or reported."
best regards,
Tim.
P.S. forgot my bugzilla password...
Matt Jones wrote:
>
> Hey,
>
> OK, I confess to writing the text surrounding precision in EML. Sorry
> if it has been confusing. I agree that we need to rewrite it to clarify
> our intent. This is a long note (sorry), but it deals with two basic
> things: 1) how do we define precision, and 2) should it be required?
>
> 1) How do we define precision
> -----------------------------
> Earlier versions of EML defined precision as the number of significant
> digits, but our discussions on this revealed that this was inadequate
> for two reasons (iirc): 1) you can only have precsions to the closest
> order of magnitude, and 2) it is exceedingly difficult to compare
> precision of measurements using two different units. For example, body
> size in centimeters with precision of 3 significant digits is very
> different from body size in meters with precision of 3 significant
> digits. That was the basic motivation for an expression involving the
> unit of measurement, which is the common way it is done in physics.
>
> So, here's what I think we are trying to capture: a measure of how
> repeatable a measurement is (as opposed to how close a measurement is to
> its true value, which is accuracy). A common expression of precision is
> standard deviation. Precision can be calculated without reference to a
> known standard, while accuracy can not. Precision is generally a
> property of the measuring device.
>
> Here's a explanation of precision and accuracy from a high school
> physics tutorial:
>
> From
> http://www.carlton.paschools.pa.sk.ca/chemical/Sigfigs/accuracy_and_precision.htm:
> ----------
> Precision indicates how close together or how repeatable the results
> are. A precise measuring instrument will give very nearly the same
> result each time it is used.
>
> Accuracy indicates how close a measurement is to the accepted value.
> ----------
>
> From http://www.angelfire.com/stars/dhsphysics/mathskills.html:
> ----------
> Precision is how fine of a measurement that the measuring instrument is
> marked off for. A typical meter stick or metric ruler has millimeter
> marks as its smallest marking. A yard stick’s smallest marks may be 1/8
> inches. By estimating between the marks we can measure to half
> millimeters and to 1/16 of an inch with these tools. This is the limit
> of our precision with meter sticks and yard sticks. But if we use these
> measuring tools in a sloppy fashion, say by letting the position of the
> end slip, the measurements will not be accurate to anywhere near a
> millimeter or an 1/8 of an inch.
>
> Accuracy is how correct or true that a measurement is. If I estimate
> the width of a room to be 4 meters, and it is, than that is an accurate,
> but not precise measurement. If I let the ends slip on my meterstick,
> and measure the width of the room to be 4.2165 meters, that is a precise
> but inaccurate measurement. Only if I am very careful, I can get an
> accuracy that equals my precision when using a meter stick.
> ----------
>
> or, from http://webphysics.iupui.edu/NH/Projects/TEAMS%5B2%5D/err6.htm:
> ----------
> Precision is the degree to which several measurements provide answers
> very close to each other. It is an indicator of the scatter in the
> data.The lesser the scatter, higher the precision.
>
> Accuracy descibes the nearness of a measurement to the standard or true
> value, i.e., a highly accurate measuring device will provide
> measurements very close to the standard, true or known values.
> ----------
>
> Links to many more definitions like these can be found at:
> http://www.chemistrycoach.com/math_skills_for_chemistry_tutori.htm#Accuracy%20and%20Precision
>
> So from these definitions I surmise that precision is a measure of
> repeatability, and has nothing to do with accuracy (ie, a measurement
> can be highly precise yet extremely inaccurate).
>
> So, our current "precision" field in EML is a straightforward interval
> showing the spread of the measurements in the unit of measurements.
> This is common in physics (e.g., 17 +/- 1 cm). I don't think we need to
> change the fundamental usage in EML, but we do need to clarify its
> definition. Unfortunatley, our definition doesn't state how the
> interval was calculated. A more exact measure of precision would be a
> standard deviation, in that we specify exactly how the precision is
> calculated. I think this is, however, overly prescriptive, and that the
> looser sense of precision as an interval in which measurements will fall
> is better.
>
> Proposed definition for precision in EML:
> Precision indicates how close together or how repeatable measurements
> are. A precise measuring instrument will give very nearly the same
> result each time it is used. This means that someone interpreting the
> data should expect that if a measurement were repeated, most measured
> values would fall within the interval specified by the precision. The
> value of precision should be expressed in the same unit as the
> measurement. For example, for an attribute with unit "meter", a
> precision of "0.1" would be interpreted to mean that most repeat
> measurements would fall within an interval of 1/10th of a meter.
>
> 2) Should precision be required?
> --------------------------------
> Originally we felt that precision was fundamental to the expression of
> measured values, so much so that it should be required. It really is
> fundamental information about the measurement, but it is clear from
> David's experience that it is commonly not available. By making it
> required, we essentially prohibit people from providing other attribute
> metadata if precision isn't available, because their EML documents can
> not validate.
>
> Thus, I propose we change precision to be optional, so that we capture
> more metadata now than we did before. However, this does not mean that
> precision is not fundamental -- it just means that we would rather have
> the other metadata about an attribute than require precision. We should
> strongly recommend that a measure of precision be provided without
> requiring it for valid EML. This change would be backwards compatible
> with EML 2.0.0 because all existing EML 2.0.0 documents would still be
> valid under the new cardinality rule.
>
> Comments?
>
> In order to track this issue and its resolution I am going to open a new
> bug in bugzilla for the precision field.
>
> Matt
>
> --
> -------------------------------------------------------------------
> Matt Jones jones at nceas.ucsb.edu
> http://www.nceas.ucsb.edu/ Fax: 425-920-2439 Ph: 907-789-0496
> National Center for Ecological Analysis and Synthesis (NCEAS)
> University of California Santa Barbara
> Interested in ecological informatics? http://www.ecoinformatics.org
> -------------------------------------------------------------------
>
> _______________________________________________
> 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
269/671-2337
tbergsma at kbs.msu.edu
http://lter.kbs.msu.edu
More information about the Eml-dev
mailing list