[R] parameter asterisks

Ben Bolker bolker at ufl.edu
Tue Aug 4 14:48:56 CEST 2009




alexander russell-2 wrote:
> 
> Hello,
> Is there a clearcut answer as to why R prints 'NA' sometimes instead of
> standard errors?
> 
> 
> mle2(minuslogl = nlikfun4, start = list(a = 1, c = 1, d = 0.2,
> 
> b = 0.1, b1 = 0.1), method = "Nelder-Mead")
> 
> Coefficients:
> 
> Estimate Std. Error z value Pr(z)
> 
> a 3.83845751 0.47320236 8.1117 4.993e-16 ***
> 
> c 0.95545367 NA NA NA
> 
> d -0.22509015 NA NA NA
> 
> b 0.04260199 0.00743892 5.7269 1.023e-08 ***
> 
> b1 0.00212538 0.00031189 6.8145 9.459e-12 ***
> 
> 

  This is in the bbmle package so I'll bite ...
The standard errors in the summary() for mle2
are based on inverting the Hessian (the second-derivative
matrix computed at the maximum likelihood estimate).
This sometimes runs into numerical problems, because
these derivatives have to be computed by finite differences.
My best guess (in the absence of a repeatable example,
hint, hint) is that you might  be able to get around the problem
by setting "parscale" (i.e. something like
control=list(parscale=c(a=4,c=1,d=0.2,b=0.05,b1=0.002)) --
you just need to get the order of magnitude right) to let
R know that the expected values of the parameters vary
by an order of magnitude.

  Ben Bolker

-- 
View this message in context: http://www.nabble.com/parameter-asterisks-tp24804823p24807838.html
Sent from the R help mailing list archive at Nabble.com.




More information about the R-help mailing list