Hello folks,
Thanks for your patience.
These issues have been discussed ever since I became involved in analysis
using computers.
Problems arise; for example:
In any analysis involving projections of future conditions, imho, there is
no (T)ruth. There is only the best estimate of appropriate models to use
and the associated data. The problem has been, and I sustpect still is,
that the data for input to environmental models are very difficult to
identify in a *consistent* basis across the model segments. Yields for
agricultural models, for example, need to be consistent across the
agricultural production possibilities, across the soil definitions, across
water use assumptions/per unit output and resource inputs. That means that
the researchers must arrive at some professionally defensible mutually
consistent data sets. The data has to be adjusted; that adjustment must be
professionally reviewed and accepted. Furthermore, for projection periods,
some way must be developed to recognize the potential change in
technologies. Again reviewed and accepted by appropriate professionals.
The first step for me was to run a model to see if we could reproduce
current, base year conditions. Check and refine with appropriate
professionals. Run a futures conditions without change in technology then
run another with changed technological assumptions. Run sensitivity
analyses to see which variables had significant impact on the results. And
so on.
At the end of the testing period, one had a model that was *valid*, that is,
one that is appropriate for the purposes at hand. It was not Truth, it was
not a model that could be expected to *reliably* reproduce the future
conditions exactly. I have always believed that is was important to use
language that would remind the analysts, the reviewers, the audience, that
these projections were best estimates of the *relative* future relationships
between conditions without the proprosed policies/programs compared to
*with* such proposals. A guide to informed judgement.
That is why I have used "valid" rather than such terms as "reliable" which
might infer to the audience that at the end of some 10-year period they
could expect to see the model results reproduced in "real life".
After all, one can only test the actuallity of the futures model when that
future period arrives; most hard decisions cannot wait that long. IMHO.
For the same reasons I have always used "projections" rather than
"predictions", "forcasts" etc.
I don't know how the climatologists and others working on climate issues
address these questions. I do believe that they are honourable folks and do
and will approach the problems they address with similar humility. At least
until I have evidence to the contrary. Not to say that there no scoundrels
among them; those types exist everywhere. :-)
So Steven B., I do *not* mean "reliable" or "Truth" when I use "valid". I
believe that you use too strict and limited definition of "validity". And
that Bush chief of staff was obviously not competent to judge and events of
the last 10+ years confound his judgment. IMHO.
I hope this helps to explain my views. As always I welcome your views. And
maybe my definitions do not fit today's language usage.
Ray
-----------------
Steve wrote:
>I'd personally like to see you provide some discussion on the difference
>between a prediciton and a projection. To me, and I do lots of
>forecasting, the two terms are pretty much synonymous. Further, from my
>readings on statistical models, the only real test of a model is how well
>its *predicitions* match up with the real data.
>
Steven B. wrote:
>Data validity cannot be determined by "trial runs" of a model. You are
>talking about data "reliability." Validity and reliability are very
>different issues. And, while models such as you mention can be validated by
>ground checks, most of the models for global weather and global warming
>cannot be verified against actual data. That is my problem with models.
>
Then Steven B. said in response to Steve:
>Perhaps. The only way a model can be 'validated' is to run the model using
>an 'independent' data set. This type of data set would have to be different
>than the data set used to 'calibrate' the model. Time can only tell which
>models are accurate and which ones are not.
>
>Ray, again you are using the term "valid" when you mean "reliable."
Validity
>refers to the "Truth" (with a capital 'T'). What you mean is how well the
>model runs using various data sets; i.e how reliable it is. I do not
>question the reliability of the computer models, indeed I know nothing of
>them at all, but I worry that policy will or will not be made based on this
>confusion of validity for reliability. If you recall in the George Bush's
>(the elder) administration there was a cessation of all work on global
>warming policy because his chief of staff (I forget his name, he was a
Ph.D.
>engineer from New Hampshire?) disagreed with the models so he said there
was
>no "evidence" for global warming. He mistook reliability for validity.
>
|