IZA DP No. 3177: How Reliable Are Income Data Collected with a Single Question?
revised version published in: Journal of the Royal Statistical Society, Series A (Statistics in Society), 2010, 173 (2), 409-430
Income is an important correlate for numerous phenomena in the social sciences. But many surveys collect data with just a single question covering all forms of income. This raises issues of quality, and these are heightened when individuals are asked about the household total rather than own income alone. Data are typically banded, implying a loss of information. We investigate the reliability of ‘single-question’ data using the ONS Omnibus and British Social Attitudes (BSA) surveys as examples. We first compare the distributions of income in these surveys – individual income in the Omnibus and household income in the BSA – with those in two other much larger UK surveys that measure income in much greater detail. Second, we investigate an implication of restricting the single question to individual income and interviewing only one adult per household: total income in respondents’ households is unobserved. We therefore examine the relationship between individual and household income in one of the comparator surveys. Third, after imposing bands on comparator survey data, we measure the information loss from banding with Generalised Entropy indices. We then assess its impact on the use of income as a covariate. Disaggregation by gender proves fruitful in much of the analysis.