Re: curve fitting -- Excel Accuracy
- To: mathgroup at smc.vnet.net
- Subject: [mg41232] Re: curve fitting -- Excel Accuracy
- From: "Jerry W. Lewis" <post_a_reply at no_e-mail.com>
- Date: Thu, 8 May 2003 09:39:58 -0400 (EDT)
- References: <200304221043.GAA29936@smc.vnet.net> <200304221043.GAA29936@smc.vnet.net> <b8vsbv$dlc$1@smc.vnet.net>
- Sender: owner-wri-mathgroup at wolfram.com
Your original answer was in response to C B, who asked about curve fitting on Excel's chart platform. Excel's polynomial curve fitting from the XY chart platform is actually quite good (better than SAS, S-Plus, R, etc). McCullough, Knusel, etc, provide useful information about very real shortcomings in Excel, but they also overlook related things that Excel does well. C B specifically asked about the interpolation routine used by Excel's chart smoother: Excel's chart smoother appears to use Bezier curves. Brian Murphy has posted code for Bezier curves. http://www.xlrotor.com/excel_stuff.htm In most cases, the difference between Bezier curves and cubic splines is minimal. David Braden has posted code for cubic splines http://groups.google.com/groups?selm=3A11AD33.B1438202%40rochester.rr.com Jerry Ian Brooks wrote: > A number of people wrote and asked for the details of the paper I referred > to about Excel's accuracy. > > It is > > "On the Accuracy of Statistical Procedures in Microsoft EXCEL 97" (with > Berry Wilson), > Computational Statistics and Data Analysis 31(1), 27-37, 1999 > > > There is a new paper that I have not yet read > > "On the Accuracy of Statistical Procedures in Microsoft Excel 2000 and > Excel XP" (with Berry Wilson), > Computational Statistics and Data Analysis 40(4), 713-721, 2002 > > > I understand there are some differences between the Excel versions, but > that the underlying problems have not been addressed and Excel XP is not > significantly different than 97. > > Ian > > > >>>Forget comparisons with Excel. If you are trying to do anything remotely >>>non-trivial Excel is almost certainly giving you garbage. If you think I >>>am exaggerating, take a look at the papers by Bruce McCullough who ran >>>the NIST (National Institutes of Standards and Technology) statistics >>>benchmark suite through Excel. The results are appalling, but not >>>surprising. I don't have the paper to hand, but from memory of the 26 >>>non-linear curve fitting problems posed Excel gave COMPLETELY wrong >>>answers to 21 of them!!! By completely wrong, I mean that where the >>>benchmark value is something like 0.6 Excel gave the answer as 13!