The SIPP has a very complex sample design. The Census Bureau provides a series of weights and guidance to users on how to apply them.(11) The basic components for all the different sets of weights are the same, namely:
- A base weight that reflects the probability of selection for a sample unit;
- An adjustment for subsampling within clusters;
- An adjustment for movers (in Waves 2 and beyond);
- A nonresponse adjustment to compensate for sample nonresponse; and
- A poststratification (second-stage calibration) adjustment to correct for departures from known population totals.
The weight variable used is WHFNWGT, for the fourth month of Wave 8. This weight represents "the population that the sample household represents in that reference month ?" (p. 8.4).
The Census Bureau recommends that standard errors be calculated using Fay's method of balanced repeated replications (BRR); however, the replicate weights are not publicly available on the Census Bureau website.(12) As a result, we used the method recommended by the Census Bureau for earlier SIPP panels.(13)