[R-sig-ME] glmmTMB: standard error of dispersion parameter

[John Maindonald]

> obj <- glmmTMB(deaths~I(NDAM2014^0.14), family=nbinom1(link='log'),
                            data=DAAG::hurricNamed)
> coef(summary(obj))[['cond']][,1:2]
                  Estimate Std. Error
(Intercept)      1.3003220  0.4764633
I(NDAM2014^0.14) 0.7702897  0.1283899

I would have expected coef(summary(obj))[[‘disp’]] to give the information you want,
but it returns NULL.  I judge that this is an oversight. If one specifies a dispformula
that is more than the default ~1, coef(summary(obj))[[‘disp’]]  does give the output
that is desired — see below.  Try however:

> obj[['sdr']]
sdreport(.) result
       Estimate Std. Error
beta  1.3003220  0.4764633
beta  0.7702897  0.1283899
betad 4.5529957  0.2116238
Maximum gradient component: 1.733627e-07

betad is the scale parameter, with a log link function.

Observe that exp(4.5529957) = 94.91632, which is what summary(obj) gives as
'Overdispersion parameter for nbinom1 family’.

-----------------------------------------------------------------------------------------------------------

Note that if one specifies a dispformula that is more than the default ~1,
coef(summary(obj))[[‘disp’]]  does give the required information:

> obj2 <- glmmTMB(deaths~I(NDAM2014^0.14), family=nbinom1(link='log'), data=DAAG::hurricNamed,
+ dispformula=~poly(NDAM2014,2))
> coef(summary(obj2))[['disp']]
NULL
> coef(summary(obj))[['disp']]
                    Estimate Std. Error   z value     Pr(>|z|)
(Intercept)         3.389720  0.2040211 16.614557 5.467767e-62
poly(NDAM2014, 2)1  7.737363  1.7277783  4.478215 7.526974e-06
poly(NDAM2014, 2)2 -1.780921  1.6305625 -1.092213 2.747396e-01

John Maindonald             email: john.maindonald using anu.edu.au<mailto:john.maindonald using anu.edu.au>

On 8/12/2018, at 10:18, Dennis Ruenger <dennis.ruenger using gmail.com<mailto:dennis.ruenger using gmail.com>> wrote:

In a fitted glmmTMB object, where do I find the standard error of the
dispersion parameter estimate for a negative binomial model? Thanks!

[[alternative HTML version deleted]]

_______________________________________________
R-sig-mixed-models using r-project.org<mailto:R-sig-mixed-models using r-project.org> mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models


	[[alternative HTML version deleted]]