| Re: Statistic codes: Significance level [message #38287 is a reply to message #38274] |
Tue, 02 March 2004 01:32   |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
"Andry William (Please remove \".spam\")" <andry@ya.com.spam> writes:
> Dear all:
>
> I am processing some long term time series of various data.
> I computed the long term trends using the REGRESS fuction.
> However, I need to also give the significance level of the
> trends. I have been looking around for a code to do it but
> have not found anything.
What do you mean by "significance level of the trend?" Do you mean
you want to estimate a confidence interval of the slope parameter to
your fit? Can you supply uncertainty estimates for your time series?
If yes to both, then routines like LINFIT or MPFITFUN can do this (Bob
points you in the right direction). If you can't supply uncertainty
estimates, then any estimated significance levels are irrelevant.
Also, if your trend model is non-linear, then estimated confidence
regions provided by these routines are not reliable, and you need to
make a chi-square confidence grid, as described in Bevington or
Numerical Recipes.
But "significance level" can have other meanings too. Sometimes this
can mean the goodness of the fit, in which case you would want to
compute the probability of chance occurrence of having a chi-square
value greater or less than a certain value (usually you would want to
do this when the fit is bad). MPCHITEST can help you there.
"Significance level of the trend" could also mean a comparison of the
fit with and without a trend. In that case you will want to perform
an F-test for that addition of the trend parameters. You can read
about this in Bevington, and use MPFTEST to compute the probabilities.
You will need two fitted chi-saure values: one with, and one without,
the trend.
Good luck,
Craig
P.S. MP* routines are at
http://cow.physics.wisc.edu/~craigm/idl/idl.html (under fitting)
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
