Abstract:
|
The Consumer Expenditure Survey (CE) is designed to produce unbiased estimates of expenditures made by U.S. households at both the national and regional levels, but not at the state level. However, there is an increasing demand for state-level expenditure estimates and re-designing the survey to produce unbiased state-level estimates in all 50 states would be prohibitively expensive. This paper presents an experimental approach to calculating weights for selected states with a sufficient representation in the CE survey so that users can use the weights to calculate unbiased expenditure estimates for their research projects. The approach was first tested in New Jersey because in the sample design based on the 2000 Census, every county in the state was in CE's sample. New Jersey's sample was reduced by dropping counties, and expenditure estimates were calculated using both the full and reduced samples. There were no statistical difference in the expenditure estimates, indicating a successful approach. Then the approach was extended to Florida and California and again there were no statistical differences, so the state weighting algorithm was successful in all three test states.
|