[R] Using betareg package to fit beta mixture with given initial parameters

[Santiago Bueno]

Tree dbh haut Btot
1 35.00 18.90 0.357
2 25.00 16.60 0.214
3 23.00 19.50 0.173
4 13.50 15.60 0.060
5 20.00 18.80 0.134
6 23.00 17.40 0.137
7 29.00 19.90 0.428
8 17.60 18.20 0.100
9 31.00 25.30 0.514
10 26.00 23.50 0.273
11 13.00 13.00 0.031
12 32.00 20.70 0.356
13 28.00 28.50 0.349
14 15.00 18.20 0.068
15 19.00 14.90 0.110
16 33.00 14.90 0.260
17 24.00 19.10 0.160
18 22.00 19.20 0.204
19 39.00 25.20 0.724
20 30.00 26.60 0.386
Hello dear all:

I am using above data to run the following code;

library(nlme)

start <- coef(lm(Btot~I(dbh**2*haut),data=dat))

names(start) <- c("a","b")

model1 <-(nlme (Btot~a+b*dbh**2*haut, data=cbind(dat, g="a"), fixed=a+b~1,
start=start,

groups=~g, weights=varPower(form=~dbh)))


I get regression parameters with the intercept being non-significant.
Therefore, I run the following code to obtain an equation without
intercept..


summary(nlme(Btot~b*dbh**2*haut, data=cbind(dat,g="a"), fixed=b~1,
start=start["b"], groups=~g, weights=varPower(form=~dbh)))


When I do run the last code, I get the following error:


Error in nlme.formula(Btot ~ b * dbh**2 * haut, data = cbind(dat, g = "a"),
 :

  step halving factor reduced below minimum in PNLS step

Can someone help??


Best regards,

Santiago

On Wed, Mar 22, 2017 at 2:26 AM, Michael Dayan <mdayan.research at gmail.com>
wrote:

> The method of setting the initial values given lambda, alpha1, etc. should
> not depend on the exact values of lambda, alpha1, etc. in my situation,
> i.e. it does not depend on my data.
>
> On Mar 22, 2017 04:30, "David Winsemius" <dwinsemius at comcast.net> wrote:
>
>
> > On Mar 21, 2017, at 5:04 AM, Michael Dayan <mdayan.research at gmail.com>
> wrote:
> >
> > Hi,
> >
> > I would like to fit a mixture of two beta distributions with parameters
> > (alpha1, beta1) for the first component, (alpha2, beta2) for the second
> > component, and lambda for the mixing parameter. I also would like to set
> a
> > maximum of 200 iterations and a tolerance of 1e-08.
> >
> > My question is: how to use the betareg package to run the fit with
> initial
> > values for the parameters alpha1, beta1, alpha2, beta2 and lambda? I saw
> in
> > the documentation that I would need to use the 'start' option of the
> > betareg function, with start described as "an optional vector with
> starting
> > values for all parameters (including phi)". However I could not find how
> to
> > define this list given my alpha1, beta1, alpha2, beta2 and lambda.
> >
> > The current code I have is:
> > mydata$y <- <my_data>
> > bmix <- betamix(y ~ 1 | 1, data = mydata, k = 2, fstol = 1e-08, maxit =
> > 200, fsmaxit = 200)
> >
> >
> > And I suspect I would need to do something along the lines:
> >
> > initial.vals <- c(?, ?, ?, ?, ?)
> > bmix <- betamix(y ~ 1 | 1, data = mydata, k = 2, fstol = 1e-08, maxit =
> > 200, fsmaxit = 200, control=betareg.control(start=initial.vals)))
> >
> > But I do not know what to use for initial.vals.
>
> If there were sensitivity to data, then wouldn't  that depend on your
> (unprovided) data?
>
>
> >
> > Best wishes,
> >
> > Michael
> >
> >       [[alternative HTML version deleted]]
> >
> > ______________________________________________
> > R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> > 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.
>
> David Winsemius
> Alameda, CA, USA
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.
>

	[[alternative HTML version deleted]]