The problem you are asking is of discrimination of separated families
(lognormal regression x gamma regression ).A problem first dealt by Sir
David Cox (1961,1962)
For your problem see from a frequentist view
PEREIRA, B. de B. . Tests and Efficiences of Separate Regression Models.
Biometrika, inglaterra, v. 68, n. no. 1, p. 345-345, 1978.
or you can use the Intrinsic Bayes Factor of Berger and Pericchi or the
Patial Bayes Factor of O'Haggan
Basilio de Braganca Pereira
Lars Chi Escreveu:
> Dear All,
>
> I'd like to ask the following simple question. I have a continuous dependent
> variable which is highly right-skewed. I attempted to model this using a GLM
> with a Gamma distribution and log-link function. Alternatively, I can apply
> a variance-stabilizing transform (such as the log) to the dependent variable
> and model this using a Normal ~ instead of a Gamma. I'd like to ask, what is
> the main difference between the two approaches mentioned above? I mean,
> which one is "more appropriate" to use in which situation.
>
> Many thanks in advance for you help!
>
> Regards,
>
> Lards.
Basilio de Bragança Pereira
*Titular Professor of Bioestatistics and of Applied Statistics
*FM-Faculty of Medicine and COPPE-Posgraduate School of Engineering and
HUCFF-University Hospital Clementino Fraga Filho.
*UFRJ-Federal University of Rio de Janeiro
*Tel: (55 21) 2562-2594 or /2558/7045
www.po.ufrj.br/basilio/
*MailAddress:
COPPE/UFRJ
Caixa Postal 68507
CEP 21941-972 Rio de Janeiro,RJ
Brasil
|