Imagine that the Acme Auto Company has designed Device X to prevent death in a vehicle crash. You and I have completed the first randomized controlled trial of Device X; we studied 400 drivers who were randomly assigned to X or a placebo device at the time of their crash and ascertained which drivers died.

The risk of death among the drivers with Device X was 0.075 compared with 0.15 for drivers without X (table): risk ratio 0.5, 95% confidence interval 0.28 to 0.90. Our draft manuscript concludes: “We found that drivers who crashed in vehicles with Device X had a risk of death which was half that of similar drivers in similar vehicles without X. If our findings represent the casual effects of X, this device can prevent about half the driver deaths that would otherwise occur in a crash.” After reading our draft, a colleague suggests we make a policy recommendation in the discussion section of our paper. Should we?

Interpreting results or stating preventive implications does not require a statement about policy; the two sentences in quotations above provide an adequate interpretation. Calls for more data or research are often unnecessary, but they are not my topic.¹ A policy recommendation is advice that some action should be taken by someone to promote health: a behavior should be adopted, advice should be given, a public education campaign should begin, a product should be purchased, a law should be enacted, and so on.

I will offer three reasons for not giving a policy recommendation …