A Numbers Game

One of my pet peeves is the way people abuse statistics (or numbers, in general). When it's done deliberately, for economic or political gain, mild irritation gives way to extreme contempt. A good example would be President Bush's oft-repeated assertion that 14 out of the 18 Iraqi provinces are safe. This number was widely reported in the media, but very few reports mentioned the fact that the other 4 provinces hold half the population of Iraq.

Paull Krugman reports on another in an editorial for the Times today.

This week, in a closed meeting with African-Americans, Mr. Bush asserted that Social Security was a bad deal for their race, repeating his earlier claim that "African-American males die sooner than other males do, which means the system is inherently unfair to a certain group of people." In other words, blacks don't live long enough to collect their fair share of benefits.
...
First, Mr. Bush's remarks on African-Americans perpetuate a crude misunderstanding about what life expectancy means. It's true that the current life expectancy for black males at birth is only 68.8 years - but that doesn't mean that a black man who has worked all his life can expect to die after collecting only a few years' worth of Social Security benefits. Blacks' low life expectancy is largely due to high death rates in childhood and young adulthood. African-American men who make it to age 65 can expect to live, and collect benefits, for an additional 14.6 years - not that far short of the 16.6-year figure for white men.

Second, the formula determining Social Security benefits is progressive: it provides more benefits, as a percentage of earnings, to low-income workers than to high-income workers. Since African-Americans are paid much less, on average, than whites, this works to their advantage.

Finally, Social Security isn't just a retirement program; it's also a disability insurance program. And blacks are much more likely than whites to receive disability benefits.

Put it all together, and the deal African-Americans get from Social Security turns out, according to various calculations, to be either about the same as that for whites or somewhat better.

Of course, this form of intellectual dishonesty is hardly unique to the current administration; pretty much every group with an agenda to push is guilty.

The BBC runs a weekly programme on the subject called More or Less. From the website:

The programme was an idea born of the sense that numbers were the principal language of public argument. And yet there were few places where it was thought necessary to step back and think in the way we often step back to think about language, about the way we use figures.
What do they really measure? What kind of truth, if any, do they capture?

Open the pages of any newspaper and you will see risks of this, targets for that, new spending and new cuts, arguments about productivity, performance indicators, measurements, statistics and quantification of every kind.

We all use numbers in so many ways to argue about, understand, help make sense of the world around us. More or Less hopes to make that task easier, more entertaining, more surprising.

More or Less often reports on figures that have featured prominently in recent news; this week's programme describes why a recent survey claiming that a quarter of English boys had committed criminal acts was seriously flawed (Listen here). For example, teenage boys who once pushed a sibling hard enough to leave a bruise or scratch were guilty of 'serious assault.' In fact, merely pushing someone six times a year (even if the victim never suffered injury) makes you a prolific offender!

Here's an amusing example: STATS, an American organization (apparently based at George Mason university) that seeks to"hold U.S. journalists to the highest standards of reporting accuracy, while providing them with concrete assistance to help them better understand the complexities and limitations of scientific and statistical material", annually presents Dubious Data awards to media stories with particularly egregious statistical or logical errors. One of the winners in 2000 was a report on a survey which claimed that 70% of people surveyed had tried to quit smoking, and precisely 0 had succeeded. What wasn't quite so clear was that only smokers were surveyed, and only current smokers. People who had successfully quit were automatically disqualified!

I got this gem from the STATS site: "It is a cardinal rule of social science research that the plural of anecdote is not data."