|
Abstract:
|
Creation of replicate weights for the jackknife procedure should reproduce the adjustment patterns of the main weighting. Often this is not possible because all the information is not available or the process is too cumbersome. When nonresponse adjustments using propensity scores are done, and there is no post-stratification, there are four possible ways of adjusting replicate weights. Thus complete reproduction of the weighting procedure may not be necessary. One can build replicates with the initial sample and reproduce the propensity categories, accept the categories but recalibrate the adjustments, estimate the change in adjustment coefficients from the respondents only, or apply the same adjustments as were applied in the original sample. Using simulated data, estimates from the four approaches were compared under different conditions.
|
