355 – Speed Session #4: Topics in Epidemiology and Survey Research Methods, Part 2
Protecting Survey Design Information by Combining Strata and Accounting for the Realized Sample Selection
Vladislav Beresovsky
NCHS
Most survey designs contain revealing information about the geographic locality of the sampled items. When combined with other sources, such information could violate the confidentiality of the survey data. We propose the modification of existing methods for masking design information by constructing combined strata variance estimators. In our approach variance estimation error is minimized conditionally on the realized PSU selection, which is often different from the optimal way of selecting PSU with probability proportional to their size (PPS). Advantages of the proposed method for combining strata over other methods are demonstrated by comparing relative deviations between standard errors of the estimates of totals and means calculated using grouped and ungrouped strata. Coverage of the finite population target variables by their estimated confidence intervals is analyzed in a simulation experiment for different ratios between variability of the target variables and PSU sizes. Optimal properties of the proposed method for grouping strata are validated in application to real survey data.