Will Big Data Kill Creativity?

Analytics has turned its attention toward ebook readers. Companies can now track the way you read; it may have implications for the way authors write. From the NY Times:

The longer a mystery novel is, the more likely readers are to jump to the end to see who done it. People are more likely to finish biographies than business titles, but a chapter of a yoga book is all they need. They speed through romances faster than religious titles, and erotica fastest of all…

At Oyster, a top book is “What Women Want,” promoted as a work that “brings you inside a woman’s head so you can learn how to blow her mind.” Everyone who starts it finishes it. On the other hand, Arthur M. Schlesinger Jr.’s “The Cycles of American History” blows no minds: fewer than 1 percent of the readers who start it get to the end…

He contrasted two romance novels. One had few Amazon reviews and little promotion, but Scribd’s data showed 6 out of 10 readers were finishing it — above average for the genre. Another romance had hundreds of reviews on Amazon, but only about 4 out of 10 readers bothered to finish it. They began closing the book, the data showed, when the writer plunged deeper into fantasy. Maybe this was not a good idea.

Some writers, of course, might not be receptive to hearing this.

“If you aren’t careful, you could narrow your creativity. You won’t take risks,” said Ms. Loftis, the young adult novelist. “But the bigger risk is not giving the reader what she wants. I’ll take all the data I can get.”


The Social Cost of Gun Ownership

Via Marginal Revolution, comes this 2005 paper examining the social cost of gun ownership. I’ve posted the good bits below.

In sum, gun prevalence is positively associated with overall homicide rates but not systematically related to assault or other types of crime. Together, these results suggest that an increase in gun prevalence causes an intensification of criminal violence—a shift toward greater lethality, and hence greater harm to the community. Of course, gun ownership also confers benefits to the owners and possibly other members of the household. The benefits are associated with the various private uses of guns—gun sports, collecting, protection of self and household against people and varmints. But if our estimates are correct, the net external effects appear to be negative…

It is important to distinguish between gun types. While handguns make up only about one-third of the private inventory of guns, they account for 80% of all gun homicides and a still-higher percentage of gun robberies. Handguns are also used in most gun suicides. Hence the social costs of handgun ownership are much higher than ownership of rifles and shotguns. Unfortunately, it is difficult to distinguish between the prevalence of long-gun ownership and handgun ownership in aggregate data, since they are very highly correlated across jurisdictions…

What would be the optimal license fee per household? Answering this question requires monetizing the social costs of the additional homicides that appear to be generated by widespread gun prevalence. One possibility would be to assign each homicide the value per statistical life that has been estimated in previous research, a range of $3 to $9 million (Viscusi, 1998), which come primarily from studies of workplace wage-risk tradeoffs. But even the lower end of this range may overstate the dollar value required to compensate the average homicide victim for a relatively higher risk of death, given that (as noted above) such a large proportion of homicide victims are engaged in criminal activity that entails a high risk of death. For example, a study of the wage premium paid to gang members engaged in selling drugs suggests a value per statistical life on the order of $8000 to $127,000 (Levitt and Venkatesh, 2000).

Suppose that given local conditions with respect to violence and gun ownership, we estimate a ratio of 10,000 handgun-owning households per annual homicide (approximately what holds at the national average for gun prevalence and homicide with an elasticity of homicide to gun prevalence of + 0.1) Given a conservative value of life, $1 million, then the appropriate license fee for a household would be $100 per year. That license fee would increase with the homicide rate, and in some jurisdictions, such as Washington, DC, would become so high that as to be the practical equivalent of a ban on ownership (a ban on handgun acquisition is currently in place in Washington, Chicago, and some other cities). Of course, this calculation ignores the problem of compliance.

This calculation will understate the optimal license fee per gun-owning household if our assumption about the average value per statistical life for homicide victims is too low, or if, as seems likely, gun violence imposes costs on society that are not well captured by any study of the value per statistical life.

Contingent valuation estimates intended to capture the complete social costs of gun violence indicate a value of around $1 million per assault-related gunshot injury (Cook and Ludwig, 2000; Ludwig and Cook, 2001). On average one in six assault-related gunshot injuries results in death (Cook, 1985; Cook and Ludwig, 2000). Under the assumption that this case-fatality rate is stable across time and space, then at the national averages for gun prevalence and homicide our baseline estimate of a guns/homicide elasticity of + 0.10 implies that each additional 10,000 gun-owning households leads to around 6 additional crime-related gunshot injuries. If these contingent valuation estimates are approximately correct, the optimal license fee per gun-owning household would be on the order of $600. If the true elasticity of homicide with respect to gun prevalence is on the order of +0.30 rather than +0.10, as suggested by some of our estimates that are based on modifications intended to reduce measurement error, the optimal license fee may be as high as $1800 per household.