Since we're getting technical, and to address Jerome's point. Consider

• Type I error: false positive - attributing a cause to a random event
  
• Type II error: false negative - attributing a real effect to chance
  

And now perform a cost-benefit analysis

• Effect of a type I error: an innocent organisation pays a settlement [cost] is paid to people as compensation [benefit] for a condition which is the result of chance, and to their ambulance-chasing lawyer [cost], and safety regulations are strengthened [benefit] unnecessarily [cost].
  
• Effect of a type II error: people damaged by a criminally negligent (or worse) organisation fail to get compensation and safety regulations are not strengthened as would be necessary.

There is no benefit to a type II error, only costs, and manifest moral iniquity to boot. The type I error is a mixed bag.

Therefore, the probability of a type II error should be minimised, with a limit to the expected cost of type I errors.

In other words, the false positives should be turned into an estimated "expected cost of doing business", and the compensation possibly capped by statute, and then the probability of false negatives should be minimised.

Can the last politician to go out the revolving door please turn the lights off?