[R-sig-ME] Kinship vs. Pedigreemm

[Jonas Klasen]

There is a difference if you have only one observation per individual. 
Then you have to make slight change in the lme4 package (cran) to get 
pedigreemm to work. See: 
https://stat.ethz.ch/pipermail/r-sig-mixed-models/2010q1/003340.html

So far as I know, in the pedigreemm package, there is only the internal 
function pedigreemm:::ZStar which can handle covariance relationship 
matrices. So there is no officially supported way for implementing a 
kinship in pedigreemm,or I'm wrong?

On 04/10/2012 11:54 PM, Joehanes, Roby (NIH/NHLBI) [F] wrote:
> Hi David and lmers:
>
> Thank you. Pedigreemm, whose backend is lmer, also can fit mixed model with a kinship matrix. I think both packages can.
>
> I did a little bit of experiment with both packages and found that the -2 log likelihood value (-2LL) of pedigreemm output is consistently smaller than that of kinship, suggesting better convergence. This is perhaps reflected by the backend numerical optimizers used by the package. Moreover, in pedigreemm, I could specify which random effects are affected by the pedigree factor and which ones are not. I could not find such option in kinship package.
>
> Other than these, I am not aware of any differences---especially on models that can or cannot be specified by either package. I wonder if any of you know of this.
>
> Thank you again,
> Roby
>
>
> On Apr 9, 2012, at 10:36 PM, David Duffy wrote:
>
>> On Mon, 9 Apr 2012, Joehanes, Roby (NIH/NHLBI) [F] wrote:
>>
>>> Yes, you are right. It appears that package kinship is now broken into
>>> several packages: coxme, kinship2, bdsmatrix, among others. Are you
>>> aware of the difference between them and pedigreemm?
>> Terry Therneau wrote in the lmekin vignette:
>>
>> "Let me emphasis this: most models that can be fit with the lmekin
>> function can also be fit with lme and/or lmer. For any such model the
>> lme/lmer functions will be faster and have superior support routines
>> (residuals, printing, plotting, etc.) The solution code for lmer is likely
>> also more reliable since it has been exercised on a much wider variety of
>> data sets.
>>
>> "However, there are models that lmekin will fit which lme will not. The
>> most obvious of these are models with a random genetic effect, e.g. a
>> kinship matrix. The second class will be models for which the user has
>> written their own variance extension, as described in the variance
>> vignette."
>>
>>
>> -- 
>> | David Duffy (MBBS PhD)                                         ,-_|\
>> | email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
>> | Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
>> | 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

-- 
__________________________________________________

  Jonas Klasen
  PhD student
  Genome Plasticity and Computational Genetics
  Max Planck Institute for Plant Breeding Research