Predicting likelihood of long-term unemployment: the development of a UK jobseekers’ classification instrument (WP116)
Predicting likelihood of long-term unemployment: the development of a UK jobseekers’ classification instrument.
Documents
Details
Working paper no. 116
This paper describes work undertaken by the Department for Work and Pensions (DWP) to explore the feasibility of developing a profiling tool to predict, at the point of first claim, the likelihood of a new Jobseeker’s Allowance (JSA) claimant reaching long-term unemployment, defined as 12 months or more with a continuous claim.
The work was undertaken in response to the Department’s interest in the application of segmentation models to improve and refine service allocation and the emergence of such tools in other countries. Additionally, the 2007 review of the Government’s welfare to work strategy conducted by David Freud and the 2008 review on conditionality and support for those claiming benefits carried out by Professor Paul Gregg both recommended investigating the development of an accurate early identification model for all jobseekers based on the approach used in Australia.
A predictive model was built using logistic regression and based on data collected from a 2010 telephone survey of 5,600 new claimants combined with administrative data held by the Department.
The paper presents the variables that the modelling suggests are the most efficient predictors of future long-term unemployment. The paper then discusses model accuracy along with the implications for implementation in an operational context.
The results from this work have increased the Department for Work and Pensions’ understanding of predictive models and claimant segmentation and how they might function in practice. The Department is undertaking further work in this area to determine a high-level approach to claimant segmentation and to generate the necessary material to inform strategic decision-making about the use of segmentation approaches within this context.