WHY CRIME DROPPED In “How New York Beat Crime,” Franklin E. Zimring refers only incidentally to a decline since 1990 in the “percentage of the population in the most arrest-prone bracket, between 15 and 29,” in both New York and the nation. The nationwide decline in that age group must be a contributing factor to the crime drop in that city and the U.S. as a whole. The book Freakonomics, by Steven D. Levitt and Stephen J. Dubner, attributes the nationwide decline in crime to the 1973 U.S. Supreme Court decision in Roe v. Wade to legalize abortion. The logic is that as more unwanted pregnancies are terminated, fewer unwanted (and unloved) children are born, so fewer will grow up to be criminals. And the timing is perfect for the decline in the arrest-prone age bracket. Steven Stone Cupertino, Calif. Zimring’s analysis of the period of the New York City crime drop was excellent but failed to refer to what came before. As a result of the 1970 Knapp Commission investigation into police corruption in the city, the New York Police Department instituted rules and policies designed to stamp out corruption that had the unintended effect of encouraging an uncommonly docile police force. This docility continued through the 1990s until the appointment of William J. Bratton as police commissioner. The Bratton-led department ushered in novel changes in tactics and policies as well as raising the level of aggressive policing in the rank and file. Bob Vializ Mahopac, N.Y. Near the end of Zimring’s article, he mentions that even New York City’s much reduced homicide rate is far higher than that of most major European cities and Tokyo. He suggests that New York must address social issues to further reduce its crime rate but seems to ignore a major pachyderm in the parlor: namely, that in these foreign nations gun ownership is far smaller. George Schuttinger Mountain View, Calif. ZIMRING REPLIES: The famous theory that Roe v. Wade reduced U.S. crime in the 1990s that Stone refers to is discussed in my book The Great American Crime Decline (Oxford University Press, 2008). I am skeptical about the decision having a major impact on nationwide U.S. crime in the 1990s because it did not strongly affect the births of children considered to be at high risk of becoming criminals. But a major influence from legalized abortion on the New York City difference is particularly implausible for three further reasons: First, what separates New York from other cities is a decline from 2000 to 2007. Why should the effects of legalization last longer in Gotham? Second, crime in New York State’s other cities did not follow the New York City trend, yet abortion is a function of state law. Last, the same situational and contingent features of crime that explain police effectiveness in stopping it argue against the deterministic view that one generation’s births will control the volume of the next generation’s crime. Vializ’s logic that policing after 1990 had such powerful effects in New York in part because the police were not very effective before 1990 is impeccable. But because we do not know whether the aggression of more recent efforts added value to strategies such as hotspots, there is no way to test the contribution of unaggressive prior efforts to the larger marginal changes over time. Finally, Schuttinger is no doubt correct that handgun use inflates the rate of American homicide. All the more remarkable, then, is the more than 80 percent drop in New York City killings despite this handicap. BIGGER BORDER It has been my understanding for a while that the radius of the observable universe is roughly 13.7 billion light-years. Yet the box entitled “What Lies Beyond?” in “Does the Multiverse Really Exist?” by George F. R. Ellis, says that astronomers see out to a distance of about 42 billion light-years, our cosmic horizon. How can light travel more than 13.7 billion light-years in 13.7 billion years? William B. Keith Houston THE EDITORS REPLY: Space is expanding, carrying objects such as galaxies and photons with it, so light travels a greater distance than a simple calculation (such as speed multiplied by time) might suggest. An object that emitted light 13.7 billion years ago is now 42 billion light-years away. This figure depends on the values of cosmological parameters. STATISTICAL SUGGESTION As a biostatistician, I concur with Charles Seife’s critical comments about the abuses of the so-called p-value as a measure of statistical significance of data in “The Mind-Reading Salmon” [Advances]. Statisticians have criticized this methodology before, sometimes even recommending banning it. I would temper such criticism, however, by pointing out that there are a variety of adjustments to p-values to take into account the kind of multiple-testing artifact Seife refers to, and they are often used (though perhaps not as much as they should be). Another issue relevant to this topic is the publication bias of many journals, which often give greater weight toward publishing articles that report statistically significant findings over those that don’t. I have advocated before that one way to mitigate problems with null-hypothesis significance testing is for editors of scientific journals to employ “results blind” decision making in determining whether to publish and make it be known that they are doing so. Articles should be accepted for publication based primarily on the judged importance and relevance of the reported study, which is usually stated and defended in the “introduction” section of the manuscript, and whether the methodology (including that of the data analysis) is sound, which can be assessed via the “methods” section of the manuscript. With this kind of review process, if 20 studies of the effectiveness of a truly ineffective drug are conducted, and one of them shows a significant effect with a p-value of 0.05 because of chance alone, investigators for the other 19 studies not showing any effect would presumably not be inhibited from writing up and submitting reports of these for publication out of fear that they’ll be denied publication because of their nonsignificant results. Publishing of those results would then cause the scientific audience to be rightly skeptical of the one significant finding amid the many reports not demonstrating it. Joseph J. Locascio Instructor in Neurology Harvard Medical School
Zimring’s analysis of the period of the New York City crime drop was excellent but failed to refer to what came before. As a result of the 1970 Knapp Commission investigation into police corruption in the city, the New York Police Department instituted rules and policies designed to stamp out corruption that had the unintended effect of encouraging an uncommonly docile police force. This docility continued through the 1990s until the appointment of William J. Bratton as police commissioner. The Bratton-led department ushered in novel changes in tactics and policies as well as raising the level of aggressive policing in the rank and file. Bob Vializ Mahopac, N.Y.
Near the end of Zimring’s article, he mentions that even New York City’s much reduced homicide rate is far higher than that of most major European cities and Tokyo. He suggests that New York must address social issues to further reduce its crime rate but seems to ignore a major pachyderm in the parlor: namely, that in these foreign nations gun ownership is far smaller. George Schuttinger Mountain View, Calif.
ZIMRING REPLIES: The famous theory that Roe v. Wade reduced U.S. crime in the 1990s that Stone refers to is discussed in my book The Great American Crime Decline (Oxford University Press, 2008). I am skeptical about the decision having a major impact on nationwide U.S. crime in the 1990s because it did not strongly affect the births of children considered to be at high risk of becoming criminals.
But a major influence from legalized abortion on the New York City difference is particularly implausible for three further reasons: First, what separates New York from other cities is a decline from 2000 to 2007. Why should the effects of legalization last longer in Gotham? Second, crime in New York State’s other cities did not follow the New York City trend, yet abortion is a function of state law. Last, the same situational and contingent features of crime that explain police effectiveness in stopping it argue against the deterministic view that one generation’s births will control the volume of the next generation’s crime.
Vializ’s logic that policing after 1990 had such powerful effects in New York in part because the police were not very effective before 1990 is impeccable. But because we do not know whether the aggression of more recent efforts added value to strategies such as hotspots, there is no way to test the contribution of unaggressive prior efforts to the larger marginal changes over time.
Finally, Schuttinger is no doubt correct that handgun use inflates the rate of American homicide. All the more remarkable, then, is the more than 80 percent drop in New York City killings despite this handicap.
BIGGER BORDER It has been my understanding for a while that the radius of the observable universe is roughly 13.7 billion light-years. Yet the box entitled “What Lies Beyond?” in “Does the Multiverse Really Exist?” by George F. R. Ellis, says that astronomers see out to a distance of about 42 billion light-years, our cosmic horizon.
How can light travel more than 13.7 billion light-years in 13.7 billion years? William B. Keith Houston
THE EDITORS REPLY: Space is expanding, carrying objects such as galaxies and photons with it, so light travels a greater distance than a simple calculation (such as speed multiplied by time) might suggest. An object that emitted light 13.7 billion years ago is now 42 billion light-years away. This figure depends on the values of cosmological parameters.
STATISTICAL SUGGESTION As a biostatistician, I concur with Charles Seife’s critical comments about the abuses of the so-called p-value as a measure of statistical significance of data in “The Mind-Reading Salmon” [Advances]. Statisticians have criticized this methodology before, sometimes even recommending banning it. I would temper such criticism, however, by pointing out that there are a variety of adjustments to p-values to take into account the kind of multiple-testing artifact Seife refers to, and they are often used (though perhaps not as much as they should be).
Another issue relevant to this topic is the publication bias of many journals, which often give greater weight toward publishing articles that report statistically significant findings over those that don’t. I have advocated before that one way to mitigate problems with null-hypothesis significance testing is for editors of scientific journals to employ “results blind” decision making in determining whether to publish and make it be known that they are doing so. Articles should be accepted for publication based primarily on the judged importance and relevance of the reported study, which is usually stated and defended in the “introduction” section of the manuscript, and whether the methodology (including that of the data analysis) is sound, which can be assessed via the “methods” section of the manuscript.
With this kind of review process, if 20 studies of the effectiveness of a truly ineffective drug are conducted, and one of them shows a significant effect with a p-value of 0.05 because of chance alone, investigators for the other 19 studies not showing any effect would presumably not be inhibited from writing up and submitting reports of these for publication out of fear that they’ll be denied publication because of their nonsignificant results. Publishing of those results would then cause the scientific audience to be rightly skeptical of the one significant finding amid the many reports not demonstrating it. Joseph J. Locascio Instructor in Neurology Harvard Medical School
