The National Science Foundation sponsors a ton of STEM programs across the US, but far too rarely does it include a randomized evaluation of whether anything actually works.
I disagree. This debate has been going ever since Hargeaves' speech some quarter century ago. IMHO RCTs in education are much less relevant than in e.g. medicine because the intervention usually doesn't target the individual but a social group of learners. So a study that does an intervention in two classes of students, where one was the control, is best thought of as having N=1 in both.
Hence, doing a sufficiently powered RCT is incredibly difficult and likely has to draw its sample from beyond the local context where the investigated curriculum is even relevant.
To be clear, I'm not saying we should throw out RCTs in education. But I certainly do not see them as the same kind of golden tool that they are in other disciplines.
Cluster randomized trials in education aren't uncommon, nor are they difficult (in fact, they are easier than randomizing individual students). And the sample size isn't the number of classrooms, but a number in between the number of students and classrooms (it depends on the intracluster correlation coefficient).
I disagree. This debate has been going ever since Hargeaves' speech some quarter century ago. IMHO RCTs in education are much less relevant than in e.g. medicine because the intervention usually doesn't target the individual but a social group of learners. So a study that does an intervention in two classes of students, where one was the control, is best thought of as having N=1 in both.
Hence, doing a sufficiently powered RCT is incredibly difficult and likely has to draw its sample from beyond the local context where the investigated curriculum is even relevant.
To be clear, I'm not saying we should throw out RCTs in education. But I certainly do not see them as the same kind of golden tool that they are in other disciplines.
Cluster randomized trials in education aren't uncommon, nor are they difficult (in fact, they are easier than randomizing individual students). And the sample size isn't the number of classrooms, but a number in between the number of students and classrooms (it depends on the intracluster correlation coefficient).