huangyhg

model reliability
hy folk,
i'm intersting in knowing how to assess reliability of a model. for example if i have a set of experimental data i can find some correlation (polynomial or other) to approximate these. which parameter i can use to evaluate if model is good or not? excell give "r^2" and i understand that is good when is near 1 but i don't know what is it and why.... i heard also f,s, and so on....how can adress myself into so much parameters...where i can find some information about all this?
thanks a lot for consideration.
gelpino
check out our whitepaper library.
the square root of r^2 gives you the correlation factor. a good rule of thumb is: >0.9 means strong positive correlation, >0.7 means positive correlation (but there might be some additional factors that need to be controlled / identified), < 0.7 means no siginifigant correlation.
r can also go negative (<= 1.0). all this means is that as your independant variable increases, your dependant variable decreases, i.e. y = -x.
the model is always reliable, it's the data that's not
your term "reliable" should be more properly called "fit." what you seem to be asking is "how well does the model fit the data?"
for a linear regression, the aforementioned regression coefficient is the measure of how well the data fits the model or vice-versa.
the regression coefficient is essentially the square root of the variance of the fit from the mean (ssr) divided by the variance of the data from the mean (sst).
the other measure is the variance of the data from the fit (sse). if sse is zero, then all the error is due to the model and if the model's variance from the mean matches the data's variance from the mean, the fit is good. if the variance of the data from the fit is large, e.g., you have a huge cloud of data, then the regression will be poor, as there will be equally likely models that could fit the data.
ttfn

__________________
借用达朗贝尔的名言：前进吧，你会得到信心!
[url="http://www.dimcax.com"]几何尺寸与公差标准[/url]