[R-sig-ME] Solution to missing ID??

[David Duffy]

On Tue, 1 Jan 2013, Luciano La Sala wrote:

> I have a dataset consisting of weights collected from two flocks of 
> chickens; treatments (infected with some bug) and controls (given a 
> placebo). The same two groups were followed over time and the weight of 
> each chicken was taken at seven equally spaced time points. I need to 
> assess the effect of treatment and time on body weight, but I may have 
> pseudo-replication problem here since seven measurements were collected 
> from each bird overtime the course of the experiment. Having said that, 
> I thought that including chicken ID as random intercept would help me 
> circumvent this issue, but unfortunately, the chickens were not 
> individually identified and the repeated measures can't be allocated to 
> each of the birds. I guess this will preclude me from implementing a 
> mixed effects model. Is there a way / analytical approach to circumvent 
> this problem, i.e. get rid the pseudo-replication issue?

You might consider a delete-d jackknife approach. This should 
help to improve your variance estimates even though you can't identify 
the true clustering in the data - presumably you have retained individual 
measurements even though you don't know who the individuals are.  An 
alternative might be to collect longitudinal data on a new subgroup, and 
fit to both datasets as a missing data type problem.

| 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