[R-sig-ME] lme4, Irtrees, and Outcomes Measured at Multiple and Nested Levels of Geography and Time

[David Duffy]

On Thu, 21 Jun 2012, Rex W. Douglass wrote:

> The Long Version: A Vietnam era survey measures territorial control over 
> time in 12,000 hamlets nested in 2,500 villages. There was a 
> hamlet-month survey, a hamlet-quarter survey, a village-month survey, 
> and village-quarter survey. Using Irtrees, lme4, and their 
> documentation, I can put together a model for hamlets and villages 
> separately over time. Two problems, however, stump me.
>
> Responses at different time intervals
>
> Questions asked monthly and quarterly about villages, which hamlets were 
> nested into: Ideally, information about the village would propagate down 
> into the thetas for hamlets, and information about the hamlets would 
> propagate up into the thetas for the villages.
>
> I appreciate any advice you might have.

There's a lot of stuff here - you could almost write a thesis about it ;)

The only comment I could make is that you want a flexible approach that 
allows different measurement errors from your different instruments, so 
structural equation modelling (eg the sem package) or in BUGS.  Those can 
also incorporate your time series, which is harder in lme4.