Home | Business News | Browse by Publication | M | Monthly Labor Review

Income imputation and the analysis of consumer expenditure data: the Consumer Expenditure Survey now provides imputed income data from 2004 forward for households that do not report a specific income value; an examination of how income imputation affects the analysis of expenditure data shows that the results most sensitive to imputation are statistics that focus on households with lower levels of expenditures.

Article Excerpt
The Bureau of Labor Statistics (BLS) Consumer Expenditure Survey (CE) began imputing income in its 2004 data. Imputation predicts income for households that reported receiving income but failed to report a specific value. Many national household surveys such as the Current Population Survey and the Survey of Consumer Finances impute missing income values. While imputation is common practice, researchers should take some precautions when using imputed data.

This article examines how income imputation affects analysis of the CE expenditure data. Most importantly, researchers who use both income and expenditures data from 2004 forward no longer have to restrict their sample to households that reported income. This study presents results for the restricted sample employed before imputation was introduced and compares them with results using the sample that should be employed after imputation. The study also compares the distribution of expenditures and measures of well-being--such as the Gini coefficient and the poverty rate--in the two samples.

The other large effect of adopting income imputation is that there may be a break in time series data that use multiple years of CE data. Because BLS will only provide imputed income data from 2004 forward, researchers who want to create a time series using income and expenditures will not have imputed income data for the period before 2004. This study uses data from 2002 to 2004 to show how the introduction of income imputation creates a break in the time series for some statistics (such as the poverty rate).

The data section of the article describes the relevant factors of the CE, and the imputation section provides an overview of the imputation procedure and how it interacts with expenditures. The methodology section describes the sample, defines the measure of expenditures, and defines other key variables. It then compares the distributions of expenditures before and after imputation and looks at how measures of well-being are affected by the introduction of income imputation.

Data

The CE Interview Survey is a continuing quarterly survey of consumer units. A consumer unit consists of members of a household who are related or share at least two of the three major expenditures: housing, food, and other living expenses. In each consumer unit, one individual is referred to as the reference person, who is the person who rents or owns the residence as designated by the respondent. Data are collected from consumer units and the individuals within these consumer units five times over a 13-month period. The first interview is used for bounding purposes and is not released publicly. The remaining four quarters of data are released publicly, and these are the data used in this analysis. By restricting the sample to consumer units that appear in all four quarterly interviews, a measure of yearly expenditures for each consumer unit can be created.

The CE has 18 income variables. The following 6 variables are collected for each individual in the consumer unit: wages and salaries, self-employment income (nonfarm), farm income, Social Security benefits, railroad retirement benefits, and Supplemental Security Income benefits. The remaining 12 variables are collected for the consumer unit as a whole: pension income, interest income, dividend income, royalty income, unemployment benefits, workers' compensation benefits, child support, alimony, income from roomers or boarders, income from other rental units, food stamp benefits, and other income.

BLS creates a complete income reporter designation to determine whether consumer units provided sufficient income data for use in official publications. (1) A consumer unit is designated a complete income reporter if it meets one of the following three criteria:

1. The reference person reports a nonzero amount for a major income source. BLS defines major sources of income as wage and salary, self-employment, farm income, Social Security benefits, railroad retirement benefits, or Supplemental Security Income benefits.

2. At least one other consumer unit member reports a nonzero amount for a major income source and reports valid zeros for all major income sources for the reference person.

3. The consumer unit reports a nonzero amount for at least one other income source and valid zeros for all major sources for all members.

A consumer unit could be classified as a complete income reporter and still not provide a full accounting of its income. For example, the reference person could report wage and salary income but fail to report a valid amount for its alimony income; this consumer unit would be classified as a complete income reporter under condition (1). In 2004, 87 percent of consumer units were complete income reporters and only 64 percent of complete...