Multiple Bridging Projects Using Multiple Imputation and Related Techniques

Comparability is sometimes desired between data collected using one measurement system and other data collected using a second measurement system. Examples include situations in which data are to be combined or compared over time but the measurement systems have changed, and situations in which one of the measurement systems is considered more accurate than the other. In such problems, if a “validation sample” exists with measurements under both systems for each sample unit, a model can be fitted to the sample that “bridges” the transition between one system and the other. The bridging model can then be used to impute measurements based on one system to data collected using the other system. The use of multiple imputation can facilitate analyses that reflect the extra variability due to uncertainty in the bridging process. This talk will give a brief introduction to multiple imputation and then will discuss three bridging projects that used multiple imputation and related techniques: one addressing the transition in the industry and occupation coding system for the 1970 census to that for the 1980 census; one addressing the transition from single-race to multiple-race reporting in Federal data collections; and one adjusting for differences between self-reported data in a large health survey and clinical measurements. The talk will highlight some similarities and differences among the three projects, discuss various statistical issues that arose, and give some theoretical results related to the efficiency of estimates and the validity of inferences based on bridged data.