ISER Working Paper Series 2006-29
Measurement error in models of welfare participation
01 Jun 2006
It is believed that a substantial proportion of individuals entitled to receive means-tested welfare benefits do not claim them, thus reducing the effectiveness of government programmes designed to reduce poverty. Existing qualitative research on welfare participation emphasises claim costs arising from the difficulty and hassle of making a claim and other intangible costs such as distaste for welfare participation and social stigma associated with dependence on benefits. In Britain, non-take-up is particularly serious for pensioners. Official estimates report that, although approximately 2 million pensioners were living in low income households in 2000-01, between a third and a quarter of them did not claim the Income Support / Minimum Income Guarantee payments to which they were entitled.
One of the most serious difficulties faced by researchers in trying to understand non-take-up behaviour is that we cannot observe directly the level of entitlement that a person would be judged to have if they were to apply for benefit. Researchers attempt to overcome this problem by using household survey data giving details of the income, assets and other circumstances of representative individuals, then to use the known rules of the benefit system to simulate the entitlement that would result from a successful claim. However, this simulation process is imperfect because individuals' responses to survey questions may be subject to reporting error.
The existence of measurement error in income and simulated benefit entitlement causes systematic distortion of the results produced by statistical analysis of take-up behaviour. Moreover, this bias is technically very complicated and difficult to remove. We point out a technical flaw in previous influential analyses of benefit take-up in the presence of measurement error and indicate how it can be corrected.
We apply the method to a model of the take-up of Income Support by older British pensioners during the period 1997-2000 and investigate the effect of measurement error bias by estimating a generalised model with explicit allowance for the distortionary effect of measurement error. We find that a failure to take account of measurement error would cause the researcher to over-estimate the extent to which potential claimants respond to financial incentives and, consequently, the magnitude of the claim costs (arising from 'stigma' or 'hassle') borne by claimants. These findings should be seen as exploratory rather than definitive but they indicate the very important role that measurement error plays in studies of take-up behaviour.