income non-respondents are often different from those who do provide income
information. Therefore, any analysis of income undertaken without considering the
type of people who do not provide income data could produce biased estimates.
Much has been written about the quality of income data obtained in surveys. A
selection of these include Miller (1953) who compared the income data on the 1950
US census with that of the 1950 US Current Population Survey (CPS). He found that
income is usually under-reported in surveys as respondents often forget minor or
irregular sources of income. Miller (1953) also found that those who were self-
employed were more likely to misreport their earnings. The self-employed were
asked about their earnings separately to the employed at both sweeps one and two
of the MCS. In addition MCS respondents were asked to report the income from
their main job, which we shall consider in this paper, as well as being asked about
earnings from second/occasional jobs.
Weinberg et al (1999) compare the CPS benchmarks from the National Income and
Product Accounts supplemented with data from the Internal Revenue Service tax
returns and the Social Security Administration. They consider the income data from
the CPS from 1947 to 1997. They claim that the tendency to under-report income is
largely from sources other than wages or salaries, for example asset income, and
interest and dividend payments.
Siminski et al. (2003) compare the Australian Bureau of Statistics (ABS) Household
Income Data with the Australian System of National Accounts, ABS population data
and the Department of Family and Community Services expenditure data. They note
that the under-reporting of income is not restricted to the bottom end of the income
distribution. They give various reasons why survey data sets and their external data
sources provide different reports. These include (i) a problem with the external data
source, (ii) different scopes of the surveys, (iii) different definitions used to define
income groups, (iv) the appropriateness of the weights used and (v) the misreporting
of income.
Rodgers et al. (1993) consider measurement error in income data for the Panel
Survey of Income Dynamics (PSID) validation study by comparing employee and
employer reports of earnings. This study is limited to the male employees of a single
large manufacturing firm but it is found that hourly wages are the most likely to suffer
from measurement error. Jackle et al (2004) used a sample of low income
respondents from the European Community Household Panel Survey (ECHP) and
undertook a validation study. They compared the income data obtained from
employers’ records and government benefit data from the Department for Work and
Pensions (DWP). They found that obtaining consent from respondents to contact
their employer was more difficult than obtaining consent to contact the DWP about
benefit records.
This paper will, however, focus less on the quality of the data actually obtained in the
MCS and more on item non-response associated with the income data. Rodgers et
al. (1993) cite earlier work on the PSID data by Duncan et al. (1985) who found from
company records those who were unit non-responders had earnings 5.5% higher
than earnings respondents, and those who were item non-responders had earnings
11.3% lower than earnings respondents.