[R] survey weights

[Thomas Lumley]

On Sun, 4 Sep 2005, A Das wrote:

> Just: "missing values in object". That would imply the
> object was created. But then I write "dchina", and it
> says "object dchina not found".

No, it would not imply the object was created.  If it was an error message 
(rather than a warning) the object would not have been created.

I presume the full message was
  Error in na.fail.default(object) : missing values in object

If so, it sounds as though you have missing values in the id, weights, or 
strata variable.
    summary(China[,c("psu","stata","weight0x"])
will verify this.

Stata will just have dropped these observations (use -svydes- to verify 
this).  If you want to drop the observations in R you need to do this 
explicitly. Having missing data may be unavoidable, but if you have 
observations in a sample it seems that you should know how they were 
sampled.
To drop these observations you could use

obsChina <- subset(China, !is.na(psu) & !is.na(strata) & !is.na(weight0x))

and then use obsChina rather than China in the svydesign() function.

 	-thomas



>                          -Bobby
>
> --- Thomas Lumley <tlumley at u.washington.edu> wrote:
>
>> On Sun, 4 Sep 2005, A Das wrote:
>>
>>> Thanks, Thomas.
>>>    Yes, that's exactly what happened: the warnings
>>> came first after "data(China)", and then after
>>> "dchina<-svydesign..." So the design object isn't
>>> being produced? The dataset is very large, and the
>>> weights were already set in Stata before
>> importing.
>>> Would either of those cause problems?
>>
>> Probably not.  What was the error message from
>> svydesign()?  That is what
>> will say what went wrong.
>>
>>  	-thomas
>>
>
>
>
>
> ____________________________________________________
> Start your day with Yahoo! - make it your home page
> http://www.yahoo.com/r/hs
>
>

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle