[R] Help for "calibration"

[Thomas Lumley]

On Fri, 7 Jan 2005 Ted.Harding at nessie.mcc.ac.uk wrote:

> On 07-Jan-05 NDIKUMAGENGE Alice wrote:
>> Hello,
>>
>> I've tried stl decomposition but I made a mistake
>> I want to make a "calibration on a sample", in french we called it
>> "Redressement d'un échantillon"
>>
>> Thank u
>
> Are you referring to the sort of thing decsribed, for example,  in
>
>  http://statbel.fgov.be/studies/cal2002d.pdf
>
> and
>
>  http://statbel.fgov.be/studies/cal_en.asp
>
> ?
>
> If so, then you may find useful things in Thomas Lumley's
> "survey" package, though I'm not expert enough to judge how
> extensively it covers this area.
>

Not very extensively.  The procedure described on those websites is a 
extension and generalisation of what I would call raking to create 
poststratification weights.  The survey package will do raking by 
iterative proportional fitting, which means using a main-effects loglinear 
model instead of the more general formulation.

If you just want to get good poststratification weights you could try 
rake() and postStratify() in the survey package, but if you actually want 
to generalised calibration you are out of luck.


 	-thomas