[R] betareg help

[Achim Zeileis]

On Sun, 13 Mar 2011, Vlatka Matkovic Puljic wrote:

> http://dl.dropbox.com/u/21595123/Book1.csv

Thanks. As I suspected. Due to the many 0.001 values, the starting value 
selection does not work well. In the development version of betareg() I've 
added a more useful warning message and an ad hoc workaround. Until it's 
available, you need to select your own starting values (as suggested in my 
previous e-mail).

For example:

## data
d <- read.csv2("Book1.csv")

## starting values
cf <- coef(lm(qlogis(cond) ~ alcoh + drug, data = d))

## basic regression model
m <- betareg(cond ~ alcoh + drug, data = d,
   start = list(mean = cf, precision = 1))

## with precision regression
m2 <- betareg(cond ~ alcoh + drug | alcoh + drug, data = d,
   start = list(mean = cf, precision = c(1, 0, 0)))

## estimate only a threshold model
m3 <- glm(factor(cond > 0.001) ~ alcoh + drug, data = d, family = binomial)

All three models show some positive but non-significant association 
between the response and the regressors.

hth,
Z

> 2011/3/13 David Winsemius <dwinsemius at comcast.net>
>
>> Nothing came through. You need to read the posting guide.
>>
>>
>> On Mar 13, 2011, at 12:59 PM, Vlatka Matkovic Puljic wrote:
>>
>>  Sorry, here is my data (attached).
>>>
>>> 2011/3/12 Ben Bolker <bbolker at gmail.com>
>>>
>>>  Vlatka Matkovic Puljic <v.matkovic.puljic <at> gmail.com> writes:
>>>>
>>>>
>>>>> That was also my first  thought.
>>>>> But I guess  it has something to do with W and phihat
>>>>> (which I'm struggling to check
>>>>>
>>>>
>>>>  Again, it would help to post a reproducible example ...
>>>> hard to debug/diagnose by remote control.  If you can't
>>>> possibly post the data to the list, or put them on a web
>>>> site somewhere, or randomize them a bit so you're not
>>>> giving anything away, or find a simulated example that
>>>> shows the same problem, you could as a last resort send them
>>>> to me.
>>>>
>>>> Ben Bolker
>>>>
>>>
>>>
>>
>> David Winsemius, MD
>> West Hartford, CT
>>
>>
>
>
> -- 
> **************************
> Vlatka Matkovic Puljic
> +32/ 485/ 453340
>
> 	[[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.
>