Don’t we need two control groups in large Randomized Controlled Trials?
Résumé
Evidence-based practices have been irrigating research in education for decades, raising great hopes of improved efficiency of educational systems. However, translating teaching practices proven effective in RCTs conducted by researchers into ordinary classrooms don’t often achieve significant results with large effect sizes (James-Burdumy et al., 2009). A literature review on this issue shows that new study designs have emerged, involving teachers in the design and/or the implementation phases, process evaluations complementing the quantitative measurement of the intervention (Connolly et al., 2018). Researchers collect information on how teachers appropriate the experimental teaching practices. Their experiment can be adjusted directly to real working conditions and target the teachers’ zone of proximal development more accurately (Bryck, 2015). Translating those experiments then has greater chances to be efficiently implemented in ordinary practices. This paper specifically focuses on how to select participants in this type of study design, where the experiment is facilitated by regular teachers. Participants are often volunteer teachers, randomly assigned to an experimental or a control group (Banerjee & Duflo, 2009). A paired control group can also be recruited among non volunteer teachers, in order to reflect the ordinary practices. This paper explores the pros and cons of these two solutions to constitute a control group and suggests a promising third: comparing the experimental group to two control groups: (a) a group of volunteer teachers, randomly assigned (b) a non volunteers group, paired to the experimental group. We describe a study using this method with 43 classrooms and 982 middle school students.