[R-sig-eco] Continuous distribution for zero or positive values (inGLM or GLMM)

[Ben Bolker]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 05/20/2011 04:11 PM, Fred Takahashi wrote:
>> You can actually fit a glmm with the 'tweedie' family in glmmPQL:
> 
> In fact I had tried this yesterday with my data, but as my residual
> plot had similar pattern that produced by a glmmPQL family=gaussian
> (despite of different scale) I assumed that the implementation of
> tweedie was incomplete. But now I seen that is only a peculiarity of
> my data (so, I will look for other options than Tweedie distribution).
> 
>> gg <-
>> glmmPQL(y~x,random=~1|f,data=d,family=tweedie(var.power=1,link.power=0))
>> ====
>>
>>  However, you would probably want to do this in a loop if you want to
>> estimate var.power as well.
> 
> For an analysis of real data, I will need to include the var.power in
> the minimization algorithm (or select the best model with different
> options of var.power)? I tried just to use tweedie.profile before
> analysis... it is not OK? Ex:
> 
> out<-tweedie.profile(d$y~d$x,p.vec=seq(1.1, 1.9,
> length=9),method="saddlepoint")#just for speed up the test...
> gg <-glmmPQL(y~x,random=~1|f,data=d,family=tweedie(var.power=out$p.max,link.power=0))
> 

  What you did is not ridiculous, but ideally you would estimate the
tweedie power parameter in the context of the mixed model fitting, not
beforehand.  It's analogous to estimating the parameters of the marginal
distribution of the data vs the parameters of the conditional
distribution given the covariates ...

  I think the discrete + continuous conditional ("delta") approach will
be easier, and probably just fine  ...

  Ben Bolker
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3W0BYACgkQc5UpGjwzenN5sgCgoDr8RRja9hiuaRur9bBVTSeR
v/kAn3c1pnZzh1RqdqT2Kpyh/oQR2SA/
=1O9/
-----END PGP SIGNATURE-----