If you do not know whether a real-world intervention will work, consider a randomised controlled experiment

Abstract

In a randomised controlled experiment, some of the experimental units (e.g., people, schools, intersections, or suburbs) are randomly assigned to treatment and the others to no treatment. Randomisation is considered very important because harsh experience in many fields has shown how easy it is for biases to occur if any other method is used. Examples are discussed of realworld interventions that either were implemented in randomised experiments or where issues of methodological quality were prominent. Some of these are from road safety (road marking, areawide traffic calming), and some are from other areas of public works (including filtering of drinking water, performance of personnel, better street lighting for deterring crime, and project cost underestimation). Systematic reviews are an even stronger form of evidence than individual randomised experiments, yet there are serious difficulties in identifying randomised experiments in bibliographic databases, particularly as research methodology has not in the past been considered important in the indexing of engineering research. How to locate information on past trials of interventions is discussed --- both self-conducted searches and the services a library can provide.