Deriving State-Level Estimates from Three National Surveys: A Statistical Assessment and State Tabulations. D. Possible Indirect Model-Dependent Estimators


The vast majority of Federal statistics are produced using direct estimates. In some situations, the Government finds it necessary to produce indirect estimates of characteristics for which there are insufficient data at the desired level of aggregation. Schaible et al. (1993) discuss eight current Federal Government programs that use indirect estimators. Two of these programs involve income estimation (state, metropolitan area, and county per capita income estimation by the Bureau of Economic Analysis (BEA) and median income for four-person families by state by the Bureau of the Census), and another uses the NHIS (model-based state estimates).

Fay's (1993) write-up of the Bureau of the Census' estimates of median family income describes how indirect estimators are used to determine inter-censal eligibility for the Low Income Home Energy Assistance Program (LIHEAP). Multivariate regression estimators combine data from the most recent census, the most recent March supplement to the CPS, and BEA data on per capita income.

Malec's (1993) write-up of model-based estimates from NHIS describes how state disability estimates have been published from the NHIS three times, always using indirect estimators. These estimators have involved synthetic estimation, ratio and regression adjustments, and composite estimation. It then describes an ongoing effort to produce estimates of physician visits in the last year using a Bayesian hierarchical approach.

Malec also mentions two other efforts to improve the NHIS' ability to produce state-level estimates. Elston, Koch, and Weissert (1990) used a regression model to stabilize the subgroup means used in synthetic estimates of disability rates. Marker (1995) and Marker and Waksberg (1994) placed a Bayesian prior distribution on the subgroup means to improve synthetic estimates of number of doctor visits in the past year and of self-reported poor health.

Some of these indirect estimators make use of a variety of administrative records maintained by government agencies. For example, the Bureau of the Census has recently developed sub-state poverty estimates that incorporate food stamp and IRS records. To be useful, such administrative records must be available for all states.

An advantage of indirect estimators is that sometimes when it is impossible to accurately produce estimates for individual states, it is still possible to develop useful models that describe the differences observed across a set of states. Thus, if groups of states implement similar programs it may be possible to model the effect of different types of programs, even while not being able to make accurate state-level estimates.

A limitation on the current use of indirect estimators for measuring the effect of the devolution of programs is that the only data that can be used in developing the models is pre-devolution. Models are much better at predicting the future in a steady-state environment. Thus, the utility of indirect estimators may increase in the future as states have a few year's experience implementing their new programs.

There are a wide range of indirect estimators that could be examined for producing state-level estimates. It would be very important for ASPE to evaluate any models that are used to produce indirect estimates, including determining measures of accuracy for these estimators.

View full report


"deriving.pdf" (pdf, 2.83Mb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®