It looks like there is a problem when the trendline calculates the
coefficient for datasets that are way off from the origin. I found the same
problem. The fitted curve looks great but the coefficients are way off. The
regression algorithm does much better. If the dataset is translated around
the means; i.e., use (x-xaverage) and y-yaverage), the trendline predicted
values also comes out pretty good. The regression algorithm will presumably
differences and perform the regression on the differences, just like the
regression add-in. the fact that the trendline curve is usually pretty good
especially if you go to the sixth order polynomial, means that the regression
is ok, but the reporting of the coefficients when it translates back to the
original dataset is in error. The same results are obtained whether in XP,
2003 or 2007. Microsoft should pay attention!

"John1791" wrote:

I am trying to work with trendlines, and I can match my data with a
polynominal trendline. The plotted trendline matches the data points very
closely, but when I apply the formula generated to actual data, my projected
values are way off. I believe the problem is that the displayed formula is
rounding constants off and not displaying to the number of significant
figures required to generate a workable formula.

Are there any ways to change any settings to allow Excel to display more
figures in the diplayed trendline formulas?

Thank you in advance.