When I first saw an IRB at work, it reminded me immediately of a creature that is familiar to Administrative Law teachers: the agency that has incentives to ignore Type 2 errors in favor of Type 1 errors. The classic example is the Food and Drug Administration. The FDA knows that if it wrongly approves a dangerous drug for sale (Type 1 error), then when somebody dies after using the drug (say, using thalidomide as a sleep remedy for pregnant women), there is hell to pay. Public outrage, congressional hearings, you know the drill. On the other hand, if the FDA wrongly fails to approve -- or wrongly delays approval -- of a safe and useful drug (Type 2 error), the costs are more diffuse and less visible. There are exceptions (think of recent pressure on the FDA over AIDS treatment drugs), but the culture of the FDA pushes everyone in the agency to be most alert for Type 1 errors, even though both types of errors are costly to health.
The same incentives operate within a university IRB. The consequences of Type 1 errors -- such as wrongly approving harmful or non-complying research protocols -- are disastrous, including suspension or loss of funding for the entire institution. The IRB, on the other hand, tends to ignore the consequences of Type 2 errors, such as the delays in research, the research projects that never occur, the research that is needlessly scaled back in ambition.
This old chestnut from regulatory theory leaves a sobering message about IRB reform. It is not enough simply to point out the possible Type 2 errors and expect the IRB to account for them from now on. There are powerful reasons why Type 1 errors get the most attention. The strategies have to focus on ways to make Type 2 errors more visible and salient across many different bureaucratic cultures and across research proposals from many different parts of the academic world.
What we need are ideas that are universal amplifiers for Type 2 errors, loud enough to make stifling of research a routine worry for the IRB, on an equal footing with the Type 1 errors. Got any?