Privacy Externalities and Unraveling

I don’t smoke and if I went to buy life insurance tomorrow, I would want to disclose that fact to the insurance company. Insurance is priced based on a pool of risks, and as a nonsmoker, I want to be placed in a different pool than the smokers are in.

But when I reveal that I am not a smoker, I set in motion a chain of inferences which should, on average, have the consequence of revealing that smokers are smokers, even if they never say anything. This is a standard result in information economics—we call it unraveling—and creates what we might think of as a privacy externality: when I reveal information about me, it has the consequence of revealing something about you. My willingness to give up my privacy gives up your privacy too. This will bite most often when one group affirmatively wants to distance itself from a second group.

We might defend this in the smoking context on the notion that absent the waiver of privacy by the nonsmokers, we get pooled risks and non-smokers end up subsidizing the higher expected insurance costs of smokers. Smokers therefore won’t fully internalize the cost of smoking, and we will have too many smokers. Allowing nonsmokers to disclose information about their nonsmoking then turns out to be socially valuable in the way that it better channels the cost of smoking to smokers, but we need to figure out how to account for, if at all, the privacy loss of the smokers.

So a couple of questions. What are your preferences? Fewer smokers or more privacy for smokers? And, if you are a privacy fan, what is the best discussion in the literature of the way in which unraveling creates privacy externalities?

Comments

[minderbender]

Isn't Professor Strahilevitz writing about this, in the context of criminal records of young men? I thought he was.

For what it's worth, my sympathy is entirely with the non-smokers. Smokers create enough negative externalities as it is (though arguably taxes have actually reversed this). The harder case would be genetic diseases - if you allow people to disclose their DNA tests, it will unravel and people with a genetic predisposition to disease will find insurance very expensive. On the other hand, if you don't allow disclosure, it might heighten adverse selection problems.

Also, there was a case of rape (I think) in a small town, and they tried to get DNA from all males in the town. Seems similar - if all but one male volunteer their DNA, the holdout is pretty much implicated (at least, one would suppose, enough for a warrant for his blood). But how is this different from compelling a blood test in the first place?

Actually, wouldn't it be interesting to track down the suspect's siblings, parents, whatever, and try to extrapolate his DNA from that? Or, can an identical twin give HIS blood so the police can nail the guy that way? But I don't think that's what you were asking.

[Ruben Rodrigues]

I'm not familiar with the literature on unraveling and privacy externalities, but I was struck by the situation you described.

I immediately thought of the numerous popular social networking websites students (and others) use, such as 'facebook'. When I usually think of privacy issues in this context I'm generally comforted by the fact that anyone can control what information they wish to display and with whom they wish to display it. However, if you're capable of discerning information on a mass scale based on what other's have volunteered, it seems quite unsettling from a privacy stand point. At the same time, it would seem to me that if enough users can either avoid divulging certain information, or at the very least be ambiguous, then the prospect for 'unraveling' would be limited. But I don't know.

Of course, in the social network scenario you don't have the same incentive to divulge as with smoking, though perhaps 'peer pressure' could influence divulging, or in facebook's case you must agree to divulge your information to third-parties in order to use their respective add-on features.