Randomised control trials (RCTs) – sometimes just called ‘trials’ – have recently come into their own as a preferred method of evaluating public policies. Policymakers around the world now use them much more regularly than they used to, evaluating a range of policies with RCTs, such as development aid, educational practice innovations and measures to promote the growth of firms, just to name a few of the applications currently in play. These examples join established trials of health interventions, such as on exercise, smoking cessation and attendance at medical clinics, and follow on from long-running programmes of randomised evaluations in welfare and employment that go back to the 1960s. There has been a gradual maturation in the skills of using the method and policymakers have gained more experience in designing and implementing trials (see Torgerson and Torgerson, 2008). The growth of official interest in trials has run in parallel with their diffusion across the academy. Whereas RCTs used to be restricted to a few areas of academic study, like health, many disciplines, such as political science (see Druckman, 2011), have seen a growth in the use of trials to answer important questions that were hitherto hard to appraise with observational data, such as the effect of canvassing on voter turnout (Green et al, 2013). The other main driver has been the recent interest in behavioural public policy: using ideas from the behavioural sciences and behavioural economics to redesign the tools of government (Oliver, 2013). Nudges and other forms of behavioural redesign, especially when directed at government communications, are particularly amenable to testing by RCTs, whereby each nudge is evaluated in a treatment arm (John et al, 2011).

RCTs have the benefit of simplicity, at least on first inspection, particularly where there is only one intervention to evaluate compared to a control group (see John, 2016). The idea rests on creating a fair comparison between something that a public agency or researcher does and a different state of affairs, which might be no intervention at all or where the recipient group just gets a normal package of services or a comparable intervention.