Comments on: Trendline Fitting Errors

By: chip

chip — Tue, 23 Feb 2010 12:41:39 +0000

Thanks, Jon. That’s actually the problem. Once I changed to an XY it matched my LINEST calcs exactly.

By: Jon Peltier

Jon Peltier — Tue, 23 Feb 2010 12:15:05 +0000

Chip -

Another important point is that the chart has to be an XY chart, not a line chart. Using the wrong chart type means the algorithm uses 1, 2, 3, etc. instead of the actual X values.

By: chip

chip — Tue, 23 Feb 2010 05:37:29 +0000

Thanks, Jon. I used plenty of digits I think. The formula generated that supposedly represents the line is:

y = 2677.2Ln(x) + 1820.4

Plugging my raw data back into this, it’s way off. By a magnitude of at least 4.

Using LINEST according to the formulas on John Walkenbach’s spreadsheet page, I come up with:

y = 1036.569Ln(x)+ (-1751.76)

Obviously a very different formula! When I plot the data plugged back into this formula, it matches the trendline drawn by Excel.

I did not find any reference to this anywhere and I wonder if it’s something in the 2003 with the 2007 compatibility pack that breaks it. I did not intend to actually use the formulas written by the trendline anyway, but was pretty concerned that I was doing something very wrong.

By: Jon Peltier

Jon Peltier — Tue, 23 Feb 2010 05:14:01 +0000

Chip -

I’ve heard lots of complaints about accuracy of Excel’s statistical computations, enough that I can’t keep track of what’s supposedly wrong in which version.

Did you allow for sufficient sig figs to reduce rounding errors in your calculations?

By: chip

chip — Tue, 23 Feb 2010 05:08:53 +0000

Jon, have you heard of issues with XL2003 displaying incorrect formulas for the logarithmic trendline? I haven’t used these a lot. I plugged that formula reported by the Add Trendline option back against my data manually and it was nowhere close to what the trendline was showing. When I calculate the log trend using LINEST I can use the results to replicate the trendline Excel draws exactly. I have Excel 2003 loaded with the 2007 compatability pack.

By: John L

John L — Mon, 11 Jan 2010 21:10:59 +0000

Thanks very much!! In all the years I have been using the TRENDLINE function I have never run across this point; but it is the reason!

By: Jon Peltier

Jon Peltier — Mon, 11 Jan 2010 21:07:44 +0000

John -

This would not be present or absent on a workbook basis. Instead it depends on the data. If you have zero values in the series data, Excel does not allow some trendline types.

By: John L

John L — Mon, 11 Jan 2010 17:53:29 +0000

I have frequently used EXCEL TRENDLINE features. A recent new workbook does not show the TRENDLINE TYPE – POWER EXPONENTIAL or LOGARITHMIC . Any idea what has happened? Previous workbooks still display those options.

By: Jon Peltier

Jon Peltier — Fri, 26 Jun 2009 10:45:25 +0000

Hi Bas -

In my understanding, Excel 2003 had finally worked out most of the statistics bugs. For years, the trendline formula was (believe it or not) considered a world-class techniue for calculating poly fits, but 2003 finally brought this capability to the worksheet functions.

Excel 2007 broke the trendline formula for many cases, inappropriately changing coefficients within a certain range to zero due to an overzealous rounding error correction algorithm. SP1 made a few improvements, SP2 has made more. We all hope and wish that Excel 2010 will clean up all of the problems (and not just with statistics!).

By: Bas V

Bas V — Thu, 25 Jun 2009 23:44:17 +0000

One of my associates has found an instance of Excel 2007 SP1 displaying incorrect fit coefficients for a third order polynomial fit with a forced zero intercept. The value of the bad third order coefficient was not real small (same order as the other two) Moreover, the bad coefficient would not display at all in SP1 version once the document was saved and reopened. The problem went away with an update to SP2. SP1 users beware!