Re: Statistical modeling of novel chemical systems( AKA Chemist with lots of stats questions)
- From: "ghostwriter" <ghostwriter25@xxxxxxxxxxxxxxxx>
- Date: 30 Aug 2006 04:58:16 -0700
Jeremy Boden wrote:
On Tue, 2006-08-29 at 13:06 -0700, ghostwriter wrote:
Let me summerize.How were these things done before the introduction of computers?
I managed to figure out a good statistical model to one of the major
reactions in my work. Everyone is extremely happy with that fact,
except for the PhD that I bumped from this years call for papers.
Unfortunatly said PhD is also the official go to guy for stats
problems, and while he hasnt snubbed me completely getting my questions
answered has been pulling teeth.
I am trying to iterate in order to maximize the fit of my curve to the
data, but what equation should I use to determine when I have the best
fit? Maximizing for R^2 gives systematic weaknesses in the model due
to the fact that the reaction is exponential, so I don't wish the
points weighted equally thoughout the entire reaction. Ideally I want
to maximize the confidence intervals, and while getting enough repeats
for each time isnt difficult, the curve gives only a single data point
for each time so a T-test isnt possible.
The equation is
f(x)=exp(a*x^b)
a is always negative and b is usally between -1 and -2
f(x) is therefor always between 0 and 1 and approaches 1 as time goes
to infinity.
Any help greatly appreciated.
--
Jeremy Boden
Much slower.
Actually I figured out the model simply by repeatly graphing different
transformations of the data which is exactly how I imagine most
modeling was done prior to computers.
I am not a mathamatican, I am a chemist, if you want to discuss
reaction mechanisms, and energy diagrams of polymer crosslinking I can
reasonably claim some expertise.
Frankly in stats I beat the crap out of 90% of the population but you
guys are part of the 10% that beats the crap out of me. Thats the
simple truth.
My current assumption is that if I take multiple tests at each point in
time and then use that data to generate a hypothetical bell curve at
each of those times, I can then fit the curve to the different bell
curves not the individual points. Then R^2 will become the percent fit
to the therotical and be confidence interval of a sort.
That will likley work for me until I can one of the stats guys from
corporate to sit down with me. Or one of you guys to give me a straight
answer.
Ghostwriter
.
- Follow-Ups:
- References:
- Prev by Date: Re: Pluto needs your help
- Next by Date: Re: Am I a crank?
- Previous by thread: Re: Statistical modeling of novel chemical systems( AKA Chemist with lots of stats questions)
- Next by thread: Re: Statistical modeling of novel chemical systems( AKA Chemist with lots of stats questions)
- Index(es):
Relevant Pages
|
