[R-sig-ME] Welch's ANOVA

[David Winsemius]

On Aug 22, 2011, at 11:25 PM, rafleon at highstat.cl wrote:

> Hello everyone!
>
> Does someone know if there is a library that include Welch' ANOVA.
>
> I have data of catch rate (weight of lobster/pot lift, it is a  
> continuous
> variable). I have two factors (two and four levels each). I cannot get
> variances stabilized (I applied Box&Cox transformation), so Welch'  
> ANOVA
> could ids in this situation due to the fact that uses Weighted least
> squares. Alternatively, can I use GLM? Maybe binomial link? I'm not  
> sure.

Not sure I see a mixed model issue being raised here, but it is  
typical to use Poisson models when variances are proportional to means  
within groups. Could have been considered count data (lobsters per  
pot)  times a continuous variable (individual lobster wts) and then  
you might raise mixed model issues,   but I'm guessing you didn't  
count the lobsters and weigh them individually.

glm(wt.in.one.pot ~  fac1 + fac2 , family="quasipoisson")

(When the LHS variable is not discrete, the quasi- form needs to be  
used or an error will be thrown.)

-- 

David Winsemius, MD
West Hartford, CT