Re: OT: interesting global warming quote found elsewhwere

On Aug 4, 1:30 am, John Popelish <jpopel...@xxxxxxxx> wrote:
John Larkin wrote:
On Sat, 2 Aug 2008 23:49:06 -0700 (PDT), bill.slo...@xxxxxxxx wrote:
Paleontologists have exacly the same kind of problem. Climate models
are in fact testable - feed in last century's climate, and see how
well the model predicts this century's climate.

That's not simulation, that's curve fitting. It works fine backwards
in time, if you tweak the model to make it so. You can model
past-history Gaussian noise that way.

You just summed up the fallacy that the book, "The Black
Swan" is all about.

When I was doing serious curve-fitting (for my Ph.D. work, back in the
1960s) the variant of the Fletcher-Powell non-linear lwast squares
curve-fitting program that I wrote included a section that exploited
the data you accumulated during the fitting process to spit out 99%
confidence limits on each of the parameters fitted and the correlation
between the parameters. If you tried to fit more parameters than the
data could support, you could get a very good fit between the model
and the data, but the confidence limits on the individual parameters
became vey large, particularly where the parameters turned out to be
hghly correlated (effectively different ways of getting the model to
change in the same way).

I'd expect moderm climate modellers to be no less sophisticated. There
will always be people who think that data analysis is some kind of
mindless process - Mann and his hockey stick come to mind - but there
seem to be enough climate modellers around these days that a least
some of them should be a little more statistically sophisticated than
the average stock-broker (who are the people being jeered at in "The
Black Swan").

John Larkin couldn't be expected to know about stuff like that - it
isn't taught in electronic engineering course. I got the limited
sophistication I can claim by talking to the demonstrator in the
"Theory of Computation 1" course I did while I doing my Ph.D. - the
course didn't cover curve fitting, but the demonstrator had written a
master's thesis on the subject which I borrowed and read, along with a
bunch of papers that he had cited in the thesis. It drove my
supervisor nuts - he thought that I wasn't spending enough time at the
bench - but I think he was eventually happy with the result.

--
Bill Sloman, Nijmegen

.

Quantcast