Note that you would probably need to modify your degrees of freedom
when that happens. Does -svy jackknife- do that automatically? In
Philip's example, it looks like the subpopulation is wholly contained
in the second wave with 60 PSUs.
On Fri, Nov 6, 2009 at 5:59 PM, Jeff Pitblado, StataCorp LP
<[email protected]> wrote:
> Philip Burgess <[email protected]> is using -svy jackknife- and
> noticed that there were 's' characters sprinkled in with the dots displayed
> for each jackknife replicate:
>
>> I have two independent complex survey designs, one with 30 JK
>> replicate weights (survey conducted in 1997), the other with 60
>> (survey conducted in 2007).
>>
>> I have combined these files in a manner similar to the Californian
>> Health Interview Surveys - http://www.chis.ucla.edu/methodology.html
>> (see the Methodology Paper on Examining Trends ... - this has detailed
>> instructions regarding how to combine such data with STATA).
>>
>> So, my combined file has 90 replicate weights.
>>
>> I svyset with the following command:
>>
>> - svyset [pweight=mhsfinwt], jkrweight(wpm0101 - wpm0190,
>> multiplier(1)) vce(jackknife) mse -
>>
>> When processing this design, for say, estimating propotions for the two surveys:
>>
>> - svy jackknife : proportion mhpsyo12, over(nsmhwb)
>> (running proportion on estimation sample)
>>
>> Jackknife replications (90)
>> ----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
>> .................................................. 50
>> ..........ssssssssssssssssssssssssssssss
>>
>> What does the 'ssss...' mean?
>
> 's' is one of those special codes like 'e' and 'n'; 'e' and 'n' are documented
> with the -dots- option in -[SVY] svy jackknife-.
>
> 's' indicates that the corresponding replicate contributes nothing to the
> variance estimate because the omitted PSU does not overlap with any
> observations in the subpopulation. For example, in the delete-1 jackknife
> this will occur when a PSU is in the estimation sample but is outside the
> subpopulation sample.
>
> When replicate weight variables are -svyset-, this can only happen if the
> values of the replicate weight variable are equal to the original sampling
> weight within the estimation sample.
>
> Given Philip's -svyset- command, it appears that -- within the estimation
> sample -- the last 30 replicate weight variables do not differ from the
> -mhsfinwt- variable.
>
> In a future update, we will modify the help file for -svy jackknife- to
> document the 's' character (like the 'e' and 'n' "dots").
>
> --Jeff
> [email protected]
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
