Andrew Caplin is visiting Northwestern this week to give a series of lectures on psychology and economics.  Today he talked about some of his early work and briefly mentioned an intriguing paper that he wrote with Kfir Eliaz.

Too few people get themselves tested for HIV infection.  Probably this is because the anxiety that would accompany the bad news overwhelms the incentive to get tested in the hopes of getting the good news (and also the benefit of acting on whatever news comes out.)  For many people, if they have HIV they would much rather not know it.

How do you encourage testing when fear is the barrier?  Caplin and Eliaz offer one surprisingly simple, yet surely controversial possibility:  make the tests less informative.  But not just any old way.  Because we want to maintain the carrot of a positive result but minimize the deterrent of a negative result.  Now we could try outright deception by certifying everyone who tests negative but give no information to those who test positive.  But that won’t fool people for long.  Anyone who is not certified will know he is positive and we are back to the anxiety deterrent.

But even when we are bound by the constraint that subjects will not be fooled there is a lot of freedom to manipulate the informativeness of the test.  Here’s how to ramp down the deterrent effect of bad result without losing much of the incentive effects of a good result.  A patient who is tested will receive one of two outcomes:  a certification that he is negative or an inconclusive result.  The key idea is that when the patient is negative the test will be designed to produce an inconclusive result with positive probability p.  (This could be achieved by actually degrading the quality of the test or just withholding the result with positive probability.)

Now a patient who receives an inconclusive result won’t be fooled.  He will become more pessimistic, that is inevitable.  But only slightly more pessimistic.  The larger we choose p (the key policy instrument) the less scary is an inconclusive result.  And no matter what p is, a certification that the patient is HIV-negative is a 100% certification.  There is a tradeoff that arises, of course, and that is that high p means that we get the good news less often.  But it should be clear that some p, often strictly between 0 and 1, would be optimal in the sense of maximizing testing and minimizing infection.