Monday, December 17, 2012

Statistics and sensibility should guide policy, not emotion.

Statistical significance should guide policy decisions, not emotions. Emotional policy decisions are how we end up with laws where someone who murders 7 people can serve less time in prison than someone who pulls down their pants in front of a child.

The data from this link: http://www.guardian.co.uk/news/datablog/2012/jul/22/gun-homicides-ownership-world-list states that in the USA, 3 out of 100,000 people die in homicides from firearms per year. But we'll use the CDC's data for a moment and call it 3.7 people. All things remaining constant, it would take 270 YEARS to just reach a total of 1% of that population. But in just 120 years, 99% or more of that population will already be dead from other causes.

The CDC states that the general death rate for 2011 was 740 in a population of 100,000. Whether that 740 includes homicides by firearm or not is inconsequential as even if it did, it would only account for 0.5% of the deaths!

To me, the incredible thing of all of this is... accuracy of data. Try to find the error rating for the statistical data gathered by the US Census, or the deaths by cause statistics from the FBI. There is none. And if there were any, these would be error rates based on the counted data and would often, only if audited at all, be determined by sampling. In a simple survey, often the accuracy is said to be +/- 3%. The accuracy of a typical automobile gas gauge is said to be +/- 7% but that's just someone's guess that most people agree with. The reason it's not published is because it's probably impractical to calculate.

But, the US Census Bureau did state it thought it's most recent error rate was only 0.4%. That's the agency rating itself (fox, hen house). Consider its basis for its data - it depends on surveys answered by us to be accurate. But lets take their number anyway. That would mean that any component statistic which is less than 0.4% by itself is just noise. Anything computed "per capita" starts with US Census data. and 3.7 gun deaths per 100,000 is a "per capita" rating. So, compare 0.4% accuracy by the agencies own rating with deaths by firearm, 0.5%. In-other-words, the population in focus could be as much as 100,400 or it could be as little as 98,600 to which we are comparing 740 against - the noise eclipses the feature data. If the same accuracy were applied to the CDC's number of 740 deaths, that would be 2.96 people, that's a very large chunk of 3.7 people so the statistic could really vary quite a bit in priority ranking relative to its neighbors from error alone.

According to the CDC, 3.2 times as many people take their own lives as those who died by firearms in a homicide. Given the tree of problems causing death, that's a much larger fruit to think about. Then there's the larger DUI fatalities, cell phone fatalities. There are some serious medical conditions which take more lives per year than homicides, suicides and DUIs combined. And yet, the public money spent on solving those problems is far less than that spent in pursuit of law enforcement and infrastructure for arms related crimes and control.

Whether you're for guns or against them, there's obviously some larger, lower hanging fruit to go after than guns if you're trying to reduce deaths. Going for new gun laws and all the infrastructure to enforce new gun laws to the extent that some wish right now, because of an emotional motive, and ignoring the statistical priority of the problem would be like trying to outlaw airplanes because you're afraid of dieing in an airplane accident.