Hi Ed,

I appreciate your advice.

As always, I am skeptical of trend fits. We are fitting a trend to
experimental, which looks quadratic (or cubic), and are using it to
interpolate between points, and not predictions. I did not notice the
equation was incorrect until I took the derivative and it did not match the
plotted results.

I agree that people go nuts with higher order polynomial terms, which can
end up with good R^2 values, but meaningless results.

Using computer results blindly is very dangerous, and I am glad to see that
you are discouraging it.



"Ed Ferrero" wrote:

Hi Dan,

Bernard has told you how to do what you asked for. Just a word of caution on
using polynomial regression.

I don't know what data you are trying to fit, but I usually discourage
people from using polynomial models unless they have a good reason to
suspect that the 'real life' data should fit a polynomial curve (e.g. a ball
thrown in the air).

If you are trying to fit business data, then polynomial curves are not very
useful IMO. Given a fixed number of data points, you can always find a
polynomial curve with exact fit just by adding enough coefficients. However,
this is not likely to be a good model for the underlying data.

If you have monthly data that may show a seasonal component and a trend,
then it may be better to try smoothing the data with a logarithmic filter
that does not remove too much of the underlying structure. You can download
a Henderson filter from
http://www.edferrero.com/ExcelCharts...2/Default.aspx

HTH

Ed Ferrero
www.edferrero.com