Excel 2007 Polynomial Order Incorrect

  • Thread starter Thread starter Darren
  • Start date Start date
D

Darren

We recently upgraded to Office 2007. When we attempt to do a trendline in
Excel, the order of polynomials is always one below what it is stated to be
and the highest ordered polynomial is always wrong. (i.e. on a fourth order
polynomial, it will only display polynomials through the third order and in
the equation, the third ordered polynomial value is wrong, the others are
correct).

We tried installing the hot fix from KB 938541 as well as right clicking on
the equation for the chart and adding significant digits under Format
Trendline Label.

A sample file showing the problem can be downloaded from:

https://share.chicagogsb.edu/seos/1000/mpd/12012008730e596e383fd90e46fe4b9af27f96a6

Thank you in advance for any assistance provided.
 
SP1 did not fix the problem.
Click to expand...

If you run the same dataset under 2003 do you get the roughly the same
coefficients for a 4th order fit as are displayed in 2007 but with one
value hidden. Or is 2007 actually doing a 3rd order fit when you ask
for a 4th order?

If you post an ASCII example of test data and observed fitted
equations 2003 vs 2007 for orders 1,2,3,4 then someone might look at
it.

I am a bit surprised at your claim because I have not seen it
misbehave on 1st order linear fits - the display is y = mx + c as
expected. OTOH most of my stuff was created in 2003 and ported to 2007
so any bugs in the Chart Wizard or user interface would not affect me.

The new polynomial fit algorithm used in charts from 2007 is as dumb
as hell and significantly less numerically stable than previously
(borderline at 2nd or 3rd order polynomials depending on the data).
Previously the charts used a clever regularised algorithm that worked
much better than the generic polynomial fit LINEST in the
spreadsheet.

That may have something to do with your problem.
Bin 2007 and ask for your money back. The product is not fit for
purpose. XL2003 charts are way better.

Regards,
Martin Brown
 
I also have trouble with the formula from adding a trendline to a 4th order polynomial. I graphed data in both 2003 and 2007 and added a 4th order trendline and displayed the formula on the graph. In both versions, I got the same formula, specifically, y = 0.00000095x4 - 0.00753977x3 + 22.51174905x2 - 29,872.75x + 14,865,177. This yielded 46690 for my text x value of 1980, but I expected a number less than 1.

The formula using linest was: y = -3.3105E-10x^4+1.77394E-06x^3+-0.002673x^2+0x+1798.06, which yields the expected 0.0065 for the x value of 1980.

Quite a bug!

Note: I also tried a 2nd order polynomial. The results were the same using linest and using the formula on the chart from the treadline. No bugs.





Martin Brown wrote:

Re: Excel 2007 Polynomial Order Incorrect
14-Dec-07

If you run the same dataset under 2003 do you get the roughly the same
coefficients for a 4th order fit as are displayed in 2007 but with one
value hidden. Or is 2007 actually doing a 3rd order fit wh

Previous Posts In This Thread:

Excel 2007 Polynomial Order Incorrect
We recently upgraded to Office 2007. When we attempt to do a trendline in
Excel, the order of polynomials is always one below what it is stated to be
and the highest ordered polynomial is always wrong

Have your downloaded Office 2007 SP1 and tried trendlines again?
Have your downloaded Office 2007 SP1 and tried trendlines again?
Download info: http://support.microsoft.com/kb/936982

I was unable to get to your file - my email address was rejected
best wishes
--

Re: Excel 2007 Polynomial Order Incorrect
SP1 did not fix the problem.

:

Re: Excel 2007 Polynomial Order Incorrect
Martin,

Here is a link to the file: http://www.mediafire.com/?7g4abmw30wn

In 2003 we were getting the correct values. 2007 is doing a 3rd order fit
when we click 4th order (and the 3rd order is wron

Re: Excel 2007 Polynomial Order Incorrect
If you run the same dataset under 2003 do you get the roughly the same
coefficients for a 4th order fit as are displayed in 2007 but with one
value hidden. Or is 2007 actually doing a 3rd order fit wh

EggHeadCafe - Software Developer Portal of Choice
ASP.NET Color Tool Derived from Reflection
http://www.eggheadcafe.com/tutorial...04-bb778fb8bfb7/aspnet-color-tool-derive.aspx
 
I'll leave out any comments about the suitability of a 4th order trendline.

Two things come to mind.

First, you should format the trendline equation using a scientific
number format with lots of digits.

Second, are you using an XY Scatter chart, or a line chart? The two
differ not in formatting of the plotted data but in the treatment of the
X variable.

- Jon
 
Back
Top