[R] MM estimator

Martin Maechler maechler at stat.math.ethz.ch
Thu Nov 14 16:08:07 CET 2013

>>>>> "SZ" == Simon Zehnder <szehnder at uni-bonn.de>
>>>>>     on Thu, 14 Nov 2013 08:52:16 +0100 writes:

    SZ> Check the gmm package with a weighting matrix equal to
    SZ> the identity.  Best

    SZ> Simon

No.    "MM" is ambigous: The  gmm package is about
       "Generalized Method of Moments"

but Izhak is really asking about MM estimation in the field of
robust statistics, where an MM estimator is a special kind of 
"M-Estimator" (and these, introduced by Peter Huber (1964), Annals,
	       fortunately are well known unambigously,
	       as a generalization of ML estimators (= MLE)),
namely an M-estimator with redescending psi (==> non-convex
problem ==> solution typically depends on starting value),
started with a high-breakdown point initial estimate.

Izhak mentioned  nlrob() from package robustbase, and he is
right that this does not provide an MM estimator, but only an M
estimate started by Least Squares.

Fortunately, we have had contributions to the robustbase
package from Eduardo Conceição since last summer.
He provided alternative versions of  nlrob(),
notably a  nlrob.MM()  
currently *hidden* {and undocumented}.

I.e. you need

 install.packages("robustbase", repos="http://R-Forge.R-project.org")


and then use  robustbase:::nlrob.MM(....)

The reason for all this is that I plan to have one nlrob()
function  but with  a new argument  'method'
and so eventually

    nlrob.MM(...., method="MM")

will correspond to current to the R-forge version


Please for all more, use the dedicated mailing list
R-SIG-robust only (i.e. do *not* just reply-all to this

Martin Maechler,
ETH Zurich

    SZ> On 14 Nov 2013, at 04:37, IZHAK shabsogh
    SZ> <ishaqbaba at yahoo.com> wrote:

    >> hi
    >> I have a nonlinear regression model with two parameter, i
    >> have estimated using nls in R and i want to also find the
    >> estimate using MM, someone refer me to this function
    >> nlrob but this is main for only M estimate. can you
    >> please help me findout a function in R for MM nonlinear
    >> regression
    >> thanks
    >> [[alternative HTML version deleted]]
    >> ______________________________________________
    >> R-help at r-project.org mailing list
    >> https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do
    >> read the posting guide
    >> http://www.R-project.org/posting-guide.html and provide
    >> commented, minimal, self-contained, reproducible code.

    SZ> ______________________________________________
    SZ> R-help at r-project.org mailing list
    SZ> https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do
    SZ> read the posting guide
    SZ> http://www.R-project.org/posting-guide.html and provide
    SZ> commented, minimal, self-contained, reproducible code.

More information about the R-help mailing list