Replicate weights allow users to generate empirically derived standard errors for estimates they produce. In theory, the standard error of an estimate measures the variation that can be expected in the estimated value of a statistic across multiple samples drawn from a given population. Researchers can use replicate weights to construct an estimate of the true standard error when only sample data are available.

RWT06 is available for all years. Selecting RWT06 adds 160 replicate weights to your data set (variables named RWT06_1 through RWT06_160). Please be aware that including these variables will make your data file quite large. The Census Bureau produced these replicate weights by using what is known as the Successive Difference Replication (SDR) method, which involves repeated implementations of the initial weighting algorithm.

To calculate standard errors for an estimate, users should generate 160 separate estimates using each of the 160 replicate weights RWT06_1 through RWT06_160. Along with the single full-sample estimate that can be generated using WT06, this information can then be used to compute the standard error of the estimate using the following formula provided by the Census Bureau:

where Y is the characteristic of interest,

Yo is the original estimate of Y,

the SUM is over 160 replicate estimates, and

Yi is the ith replicate estimate of Y.

Once calculated, the standard error is useful for constructing confidence intervals and in hypothesis testing. Both SAS and Stata include procedures that are designed to use the replicate weights to produce estimates of means, proportions and regression coefficients with correctly-calculated standard errors.

Additional information about the methodology used by the Census Bureau to create replicate weights can be found in Chapter 14 of CPS Technical Paper 66, available here.

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.

Four different sets of replicate weights are available for the ATUS. Users should select replicate weights that match the estimation weights they have chosen. That is, RWT03 should be used with WT03, RWT04 should be used with WT04, RWT06 should be used with WT06, and REHWT should be used with EHWT. Estimation weights should be chosen based on which samples are included in analyses. For most applications, WT06 is the preferred weighting variable. Weights derived using the 2003 weighting methodology should not be used in analyses that pool 2003 ATUS data with data for any other year. Users who pool 2003, 2004 and/or 2005 data may use WT04. Users whose analyses include data from 2006 or later data must use WT06. Users whose analyses include Eating and Health Module data must use EHWT. WT03, RWT03, WT04 and RWT04 can be made available to users of the ATUS Data Extract Builder by special request.

2003-2010: ATUS respondents.