5 ms·

Everybody’s Lying about the Link Between Gun Ownership and Homicide

11 points by wskinner 9y ago

vannevar 9y ago
The author's criticism of other analysis have some merit, though his chief complaint is that the others include suicide and accidental death. But isn't his point that you shouldn't selectively exclude contrary data? Since those who die by accident or suicide are just as dead, it's hard to argue for their exclusion except as another way of "lying" about the relationship between gun ownership and death by firearm.
comex 9y ago
This article accuses others of selectively choosing what to graph, but does much the same itself. It graphs gun ownership vs. firearm homicide rate for: 1. All countries 2. Countries with low firearm homicide rates 3. European countries but neglects to graph rich countries, which would make the United States stick out like a sore thumb in the top-right corner. It goes on to cite a graph from Vox of precisely that, and criticize it on various grounds, including its lack of R^2 value - but without providing a substitute graph, or a substitute R^2 value. Admittedly, the R^2 would probably still be low (though I don't know how low). And even if the graph has the United States in the top right, a single data point that's an outlier in two measures would not itself establish a link between those measures. But neither does a lack of correlation between the measures in other data points, by itself, rule out that they're related in the case of the outlier point. For example, consider the amount of time I spend using my phone in a particular hour, versus the amount of time I've already spent using it since its last charge. Most of the time, my guess is those factors are positively correlated (revealing how non-busy I am that day) or uncorrelated. But on rare occasions, the latter value approaches my phone's battery life, at which point the former drops to zero. In other words, it's an outlier case where the correlation suddenly becomes negative. To determine whether a connection might exist, it's useful to ask: - Whether there is a plausible mechanism for causation (there's a pretty obvious one in this case), or more generally for the two measures to reflect aspects of the same underlying processes (i.e. gun culture). - Whether there's independent reason to think those processes would act unusually for that data point (there is, because the U.S.'s gun culture is unique).

11 points by wskinner 9y ago

The author's criticism of other analysis have some merit, though his chief complaint is that the others include suicide and accidental death. But isn't his point that you shouldn't selectively exclude contrary data? Since those who die by accident or suicide are just as dead, it's hard to argue for their exclusion except as another way of "lying" about the relationship between gun ownership and death by firearm.

comex 9y ago

This article accuses others of selectively choosing what to graph, but does much the same itself. It graphs gun ownership vs. firearm homicide rate for: 1. All countries 2. Countries with low firearm homicide rates 3. European countries but neglects to graph rich countries, which would make the United States stick out like a sore thumb in the top-right corner. It goes on to cite a graph from Vox of precisely that, and criticize it on various grounds, including its lack of R^2 value - but without providing a substitute graph, or a substitute R^2 value. Admittedly, the R^2 would probably still be low (though I don't know how low). And even if the graph has the United States in the top right, a single data point that's an outlier in two measures would not itself establish a link between those measures. But neither does a lack of correlation between the measures in other data points, by itself, rule out that they're related in the case of the outlier point. For example, consider the amount of time I spend using my phone in a particular hour, versus the amount of time I've already spent using it since its last charge. Most of the time, my guess is those factors are positively correlated (revealing how non-busy I am that day) or uncorrelated. But on rare occasions, the latter value approaches my phone's battery life, at which point the former drops to zero. In other words, it's an outlier case where the correlation suddenly becomes negative. To determine whether a connection might exist, it's useful to ask: - Whether there is a plausible mechanism for causation (there's a pretty obvious one in this case), or more generally for the two measures to reflect aspects of the same underlying processes (i.e. gun culture). - Whether there's independent reason to think those processes would act unusually for that data point (there is, because the U.S.'s gun culture is unique).