|
It is not appropriate to use the bootstrap commands in RADL. To use bootstrap commands appropriately a user needs to be able to identify the primary selection units (PSU). In the case of ABS CURFs, all respondents in the same CD (Collection District) belong to the same PSU. However, grouping respondents into PSUs significantly increases the risk of a respondent being identified so the ABS cannot release this information. To enable researchers to produce accurate variance estimates, the ABS releases the replicate weights which take the sample design into consideration, as well as any other steps in the estimation process such as non-response adjustments.
For STATA users, the ABS has made available the user-supplied svr suite of commands to assist users in calculating standard errors using replicate weights. Users must first use svrset to set the key survey replication variables for use by the other svr commands.
The following parameters should be specified:
- Replication method is the delete-one jackknife: svrset set meth jk1
- Main analysis weight is usually the person weight: for example svrset set pw FINWT (or equivalent)
- Variables denoting the replicate weights need to be specified: for example svrset set rw REPWT1-REPWT30 (or equivalent)
- Degrees of freedom for variance calculation are one less than the number of replicate weights: for example svrset set dof 59 (or equivalent)
A useful summary of replicate weights is available in the STATA help documentation: http://www.ats.ucla.edu/stat/stata/Library/replicate_weights.htm
This page first published 27 November 2009, last updated 20 April 2012
|
