Type I Errors Reconsidered

The concept of Type I and Type II errors is often used to describe the incentives of government regulators. The claim is that the bureaucrats will want to avoid Type I errors (false positive), as these are quite embarrassing to the bureaucracy. Therefore they will err on the side of caution and commit more Type II errors (false negatives).

A typical example is the FDA: approving a harmful drug (Type I error) kills people and looks bad publicly. Not approving a safe drug (Type II error) kills people but is mostly unseen. This incentive structure will tend to produce more Type II errors and also more deaths than if the appropriate trade-offs were taken. That is the standard story, as I understand it.

There is at least one other factor to consider which isn't part of the above story. Whenever there is a Type I error, the very bureaucracy that committed the error will often gain more power, and that power is above and beyond what is needed to correct similar errors in the future. This is due to various related factors, such as media hysteria, Congressional desire to "do something," etc.

The Thalidomide case would be a perfect example, except that the FDA had not yet approved it when the resulting birth defects became known (this was due to red-tape unrelated to the possibility of birth defects). However, after the drug was approved in Europe and the results became known, the expected hysteria ensued. But the FDA already had the power to regulate the safety of drugs. The resulting legislation also gave the FDA the new power to regulate the effectiveness of drugs. But the effectiveness of Thalidomide was never in question, only the safety, which the FDA already had the power to regulate.

The big problem with my hypothesis is that we don't see this going on more often. Regulatory agencies do seem to bias towards Type II errors, although they don't seem to mind benefiting from Type I errors. Perhaps there are institutional structures that prevent deliberate Type I errors, perhaps it is just dumb luck, or perhaps the examples are just not visible. I am leaning towards the last two explanations.