The Employment Retention and Advance (ERA) Demonstration piloted between 2003 and 2007, was randomly offered to eligible lone parents and long-term unemployed individuals; the treatment consisted of financial incentives and assistance to find, retain and advance through employment. It has been evaluated experimentally by a research consortium involving IFS. We propose to add to the predominantly US literature which assesses non-experimental evaluation methods against an experimental benchmark.

The analysis will be done by implementing some of the main non-experimental methods that have been used in UK evaluations, including area controls, difference-in-differences techniques and matching. The aim is to present estimates of the bias arising from using a selection of non-experimental approaches, ranging from the most basic to the more robust techniques. Conducting a non-experimental analysis of a randomised intervention requires identifying a non-experimental comparison group from another data source. We intend to exploit the existing datasets that are the most amenable to this purpose, the Labour Force Survey and the Families and Children Study. In this sense, the eventual 'best' non-experimental approach should be viewed as the one arising from the 'best' use of the available data, so that the results of the random assignment will be compared against the best alternative that can be devised with the data and resources available.