Lies, darn lies and statistics

David MoonBlog

An old joke claims that 72.35 percent of all statistics are made up on the spot – and the more precise the conclusion, the more likely the claim is to have been contrived from thin air. That’s why I tend to question highly specific conclusions drawn from statistics that are buried multiple links away from the headline.

I recently read an article titled Gun Laws Save Lives, which described a study that compared per capita gun deaths with a compiled index of 50 different policies related to gun laws, each weighted by some factor determined by the study’s authors. As suggested by the article’s title, there was a correlation between the index and gun deaths; states that were deemed to have stronger gun laws had fewer gun deaths.

That did not prove, however, that the gun laws were responsible for the differences in gun deaths – only that they were correlated.

Just for mathematical fun, I took the study’s gun laws data and compared it to several other state-level data series to see what other correlations might exist. It is important to note that I am not taking any position on whether we should have more, fewer, or different gun laws. I’m not making any suggestions about guns. I am merely commenting on how people can use – and misuse – data.

Interestingly, deaths stemming from knives, carbon monoxide poisoning and smoking are each almost as highly correlated with state guns laws as are deaths from guns. (I am happy to provide correlation coefficients and significance testing data for any interested nerds.) I doubt anyone would claim that gun laws have any effect on deaths from carbon monoxide poisoning.

Curiously, states with higher percentages of obese people have more people killed with guns. The relationship isn’t nearly as strong or significant as the relationship between gun laws and deaths from knives, guns or smoking, but it is still very highly correlated.

I expected state-level deaths from car accidents and from guns to be correlated but was surprised to find they aren’t.

Of the dozen state-level data series I compared with gun deaths, the most highly correlated series was family income. That is, almost half of the variation in gun death rates between states could be statistically explained by differences in family income.

Interestingly, gun laws are more highly correlated with a state’s average family income than with gun deaths. Are people in low-income states more likely to commit gun murders? Or are individuals with higher incomes less likely to use a gun to kill someone?

Statistics don’t tell us which are the chickens and which are the eggs, but it doesn’t prevent agenda-driven people from pretending that they do.

David Moon is president of Moon Capital Management. A version of this piece originally appeared in the USA TODAY NETWORK.