[R] NLS results different from Excel

Marc Schwartz marc_schwartz at me.com
Wed Feb 20 21:18:29 CET 2013


Just as an FYI, there is the NISTnls package on CRAN by Doug Bates:

  http://cran.r-project.org/web/packages/NISTnls/index.html

There have also been threads over the years touching on some of the issues in replicating the NIST results, for example:

  http://tolstoy.newcastle.edu.au/R/devel/06/07/6331.html

Regards,

Marc Schwartz

On Feb 20, 2013, at 9:58 AM, Bruce McCullough <bdmccullough at drexel.edu> wrote:

> The idea that the Excel solver "has a good reputation for being fast and 
> accurate" does not withstand an examination  of the Excel solver's 
> ability to solve the StRD nls test problems.  Solver's ability is 
> abysmal.  13 of 27 "answers" have zero accurate digits, and three more 
> have fewer than two accurate digits -- and this is after tuning the 
> solver to get a good answer.  For details see
> 
> B. D. McCullough and Berry Wilson
> "On the Accuracy of Statistical Procedures in Microsoft Excel 2000 and 
> Excel XP,"
> /Computational Statistics and Data Analysis/ *40*(4), 713-721, 2002
> 
> The situation is the same for Excel 2003 and Excel 2007.  The alleged 
> "improvements" for Excel 2010 have had not much practical effect.  Excel 
> solver does have the virture that it will always produce an answer, 
> albeit one with zero accurate digits.
> 
> To see an extended example of precisely how solver fails:
> 
> B. D. McCullough
> "Some Details of Nonlinear Estimation," Chapter Eight in
> /Numerical Methods in Statistical Computing for the Social Sciences, /
> Micah Altman, Jeff Gill and Michael P. McDonald, editors
> New York: Wiley, 2004
> 
> I am unaware of R being applied to the StRD, but I did apply S+ to the 
> StRD and, with analytic derivatives, it performed flawlessly.
> 
> 
> On 02/19/2013 08:38 PM, r-help-request at r-project.org wrote:
>> May I be allowed to say that the general comments on MS Excel may be alright,
>> in this special case they are not.  The Excel Solver -- which is made by an
>> external company, not MS -- has a good reputation for being fast and accurate.
>> And it indeed solves least-squares and nonlinear problems better than some of
>> the solvers available in R.
>> There is a professional version of this solver, not available from Microsoft,
>> that could be called excellent. We, and this includes me, should not be too
>> arrogant towards the outside, non-R world, the 'barbarians' as the ancient
>> Greeks called it.
>> 
>> Hans Werner
> 
> 
> -- 
> B. D. McCullough, Professor
> Department of Decision Sciences
> LeBow College of Business



More information about the R-help mailing list