08 October 2007

This blog entry is proven to reduce your gullability by 75%!!!

I have had a lot of "almost posts" in the past few weeks. Most have had to do with how statistics are used to win arguments, scare people, and generally hide the reality. I know it's cliché to mention the book How to Lie with Statistics, so I won't. I just want to vent on 2 things I hate about how people use statistics and one thing I like about statistics that is often glossed over.

Hate #1: "They have proven that X causes/prevents/is better than Y." X can be smoking weed and Y can be lung cancer/Alzheimer's/cigarettes, or whatever. I hate it when people cite papers they have not read in order to claim outright victory in an argument that is by NO MEANS closed. I will cite but not discuss the proposition that most published research is wrong, or at least some is by definition.

Hate #2: "X reduces the risk of Y by 50%." X can be drinking red wine and Y can be dying of heart disease, or whatever. Whenever you hear something like that, your first thought should probably be "so what?!" The relevant statistic is what your risk was and what your risk would be if you do X. Overlooking the possibility that the statement is probably wrong (see Hate #1), if your current risk is 1 in 5,000, your new risk is 1 in 10,000. But if your current risk is 1 in 5, your new risk is 1 in 10. The later is much bigger news to you.

Hate #2 brings me to what I think is glossed over in statistics, especially as reported by the media. Statistics look at a population, but you and I are both individuals. A population statistic, say the risk of dying in a car crash this year, is about 0.013% (40,000 Americans/year/300 million Americans). If you live to 100, the lifetime risk is 1.3% or 1 in 75. But population statistics are meaningless if the population is diverse.

Do you think that your risk is 1.3% if you drink and drive? Nope, higher.

What if you always obey the speed limit? Nope, lower.

The Internet abounds with risk calculators for health problems, presumably because there is so much data in this field. They typically ask you to answer a number of questions and then use some multiplier for each answer to determine your overall risk. Often the calculator will tell you by what % you can reduce your risk by changing your behavior. (Life Expectancy Calculator)

I would like to see a compendium of risk calculators, but for everyday stuff, not just medical conditions. Like, by what percentage can I reduce my risk of choking if I don't eat hard candy or mow my own lawn?!

No comments: