Description of Income Imputation Beginning With 2004 Data
In a continuing effort to produce high quality data, the Consumer Expenditure Survey (CE)
has implemented multiple imputation of income data starting with the publication of the
2004 tables. Prior to 2004, CE has only published income data collected from complete income
reporters (see glossary). However, even complete income reporters do not provide information
on all sources of income for which they report receipt. This problem is reduced, but not
eliminated, with the collection of bracketed income data starting in 2001. However, even
bracketed data only provide a range in which income falls, rather than a precise value for
that income. In contrast, imputation allows income values to be estimated when they are not
reported. In multiple imputation, several estimates are made for the same consumer unit,
and the average of these estimates is published.
The introduction of multiply imputed income data affects the CE published tables in
several ways, because income data are now published for all consumer units, instead of
complete reporters only. The most obvious result of this change is seen on the tables
showing expenditures categorized by income before taxes, including by quintile. Starting in
2004, columns describing income, expenditures, and characteristics for "Total complete
reporting" and "Incomplete reporting of income" no longer appear in these tables, and the
column entitled, "All consumer units" appears on all income tables. This occurs because
income quintiles and income ranges are no longer defined using only on data collected from
complete income reporters, but instead are defined using the average of the multiply imputed
values for each consumer unit. Also, in the tables showing expenditures by demographic
characteristic, such as age of reference person, the footnote indicating that "Income before
taxes" refers to "Components of income... derived from 'complete reporters only'" no longer
appears.
Because of the implementation of income imputation, data for 2004 are not strictly
comparable to those of prior years, especially for the income tables. In the 2003 CE tables,
for example, nearly 16 percent of all consumer units are defined to be incomplete reporters.
Income imputation allows expenditure data for these consumer units to be included in income
categories from which they were previously excluded. To the extent that the incomplete
reporters have different demographic characteristics, such as age of reference person, or
expenditure patterns than the complete income reporters who were previously assigned to these
categories, averages for demographic characteristics and annual expenditures will change.
In addition, some complete income reporters who do not provide values for all sources of
income for which they reported receipt will be classified in different income categories
(ranges or quintiles) after imputation estimates these missing values. Again, to the extent
that their demographic characteristics or expenditure patterns differ from the complete
reporters previously assigned to these categories, the data in the tables will change.
Furthermore, certain expenditures, such as personal insurance and pensions, are computed using
income data. As a result of imputation, average annual values for these expenditures may be
substantially different in the 2004 CE tables than in tables for previous years.
In addition, multiply imputed microdata require special methods for analysis.
Users of the public use microdata will find guidance in the 2004 Public Use Microdata
Documentation for both the Interview and Diary Surveys. Included in these documents is a
description of the method used to obtain proper variance estimates for multiply imputed data.
These procedures are also used to produce estimates of standard errors and coefficients of
variation for multiply imputed income data in the published standard error tables available
on this website.
Last Modified Date: November 29, 2005