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: [Phys-l] data analysis : pendulum period versus length



I see the value of John's neutral statement of the comparison between model and data: "How big
is the discrepancy between the model and the data?"
Daniel Crowe
Loudoun County Public Schools
Academy of Science
dan.crowe@lcps.org
John Denker 05/11/11 7:13 PM >>>
On 05/11/2011 02:57 PM, Dan Crowe wrote:
John Denker said, in part: "The real question IMHO is how well the
data fits the model."


Having been trained as an experimentalist, I prefer to phrase the
question in the opposite direction: "How well does the model fit the
data?"

Interesting point.

It could go either way. For example, I consider myself a
card-carrying (and scar-carrying) experimentalist ... but if
a high-school kinematics experiment finds that F is not equal
to ma, I would bet that most of the discrepancy comes from
the data, not from the model.

In general though, I would argue against both extremes. In the
overwhelming majority of cases, the data is imperfect *and* the
model is imperfect. And you cannot attribute the discrepancy
to one or the other. You cannot even apportion it!
-- You could say that the experiment is non-ideal, and a
"better" experiment would have more closely agreed with
the model.
-- You could say that the model was incomplete, and a
"better" model would have accounted for the nonidealities.

So I recommend more neutral language. For example, "How big
is the discrepancy between the model and the data?"

=============

To clarify another point: You *always* need a model. A previous
question could be interpreted as asking about "some" data converging
to "some other" data ... but it never works that way. You have to
ask about data converging to a model. If you have no model, or
(equivalently) if you consider every imaginable model impartially,
the result is nonsense, guaranteed. This is an extreme case of
what is sometimes called "overfitting". That is, if you fit data
to data, you will get massacred by overfitting, guaranteed.

For more on this, see
http://www.av8n.com/physics/thinking.htm#sec-omit
_______________________________________________
Forum for Physics Educators
Phys-l@carnot.physics.buffalo.edu
https://carnot.physics.buffalo.edu/mailman/listinfo/phys-l