- Questionnaire Text
- UnHarmonized Variables
160 sets of 9-digit replicate weights are included in extracts where this selection is made. The variables will be named REPWT1 through REPWT160. Selecting replicate weights will dramatically increase the size and processing time of extracts; users should request them only if they plan to use them.
REPWT1-REPWT160 are 9-digit numeric variables with four implied decimal places.
REPWT provides 160 separate household-level weights that allow users to generate empirically derived standard errors. Person-level replicate weights are available in REPWTP; the household-level replicate weights have been taken from the person-level replicate weights of the reference person.
More information about replicate weight creation and use is available on the IPUMS-CPS replicate weights FAQ page and in the Census Bureau's "Estimating ASEC Variances with Replicate Weights" document.
Calculating the standard error of an estimate enables the construction of a confidence interval around the sample estimate of interest and may also be used in hypothesis testing. In theory, the standard error of an estimate measures the variation of a statistic across multiple samples of a given population. Researchers can use replicate weights to mirror this theoretical approach when only sample data is available, and the resulting standard errors have a higher degree of precision than standard asymptotic standard errors.
The 2005-onward CPS samples contain 160 replicate weights at the person level (variables named REPWTP1 through REPWTP160). For self-representing strata (i.e., areas with populations large enough to be represented with certainty in the sample), the replicate weights were produced using what is known as the successive difference replication (SDR) method. For non-self-representing strata (areas that are not represented with certainty in the sample), the replicate weights were produced using the modified half-sample technique. Both methods involve repeated implementations of the initial (full-sample) weighting algorithm, such that full information about the CPS sample is available in the replicate weights. Nevertheless, users should use these replicate weights only for generating variance estimates, not for obtaining unique parameter estimates.
Estimates on the entire population are prepared by projecting forward the resident population from the last available census. These projections are derived by updating the demographic census data from a number of other data sources that account for death, births and net migration. About 3 years after every census (i.e. 2003 for the 2000 Census and 2013 for the 2010 Census), the Census Bureau updates its independent population control and provides a new weight for the relevant years.
Two important points should be noted here. First, the lag between when the Census is conducted and when the CPS weights are updated is about 3 years. While the Census data are being processed, the CPS files are made available using the weighting scheme from the US Census prior to the latest Census. Second, once the files are updated, the old weights become obsolete and are replaced in the IPUMS data extract system. Published estimates from the lag years that use the old weights are not always updated. For example, 2010 poverty estimates were released in March using the 2000 population controls. Once the 2010 population controls were made available, IPUMS-CPS replaced the March 2010, 2011, and 2012 weights that are based on the 2000 population control with weights that are based on the 2010 population controls.
IPUMS-CPS makes available only the most up-to-date weights. The old values are available here: Old SPM and Weights Values.
User Note: The successive difference replication approach (SDR) is different from other methods for creating replicate weights such as balanced repeated replication (BRR) and jackknife estimation, and standard statistical software packages have no built-in method to handle them. However, Stata's jackknife standard error program can be adapted to calculate replicate standard errors for CPS data; see the IPUMS-CPS replicate weights FAQ for details.
Additionally, it is possible for replicate weights to take negative values for certain cases; again, users should use these weights only for variance estimation purposes and not to obtain independent estimates.
Replicate weights are only available for the 2005-2009 samples of the CPS.
Comparability with IPUMS-USA
The person-level replicate weights are, broadly speaking, comparable to the person-level replicate weights in the ACS and PRCS samples 2005 and later. However, the details of the replicate weight calculations differ for non-self-representing CPS strata. Additionally, ACS/PRCS replicate weights are integers, while CPS replicate weights are not.
- All households.
|2005 – 2021||-||-||X||-||-||-||-||-||-||-||-||-|