(06/22/05)
I'm sick of hearing dumbasses with no math or statistical background quoting or criticizing various studies. I present to you the most common mistakes people bring to me in discussion. Because I have no credentials, this rambling is a bit more link-heavy than most.
- Confusing correlation with causation
- This is so common that it isn't even funny anymore. Someone recently remarked to me that "Old people say the most racist things; I wonder what kind of crazy racist stuff we'll be saying when we're 70." I tried, as politely as possible, to explain that they were saying the same racist stuff when they were 30 years old. Then there are people who claim that eating breakfast leads to better grades, and there are those who refute them.
- Falling for the classic clustering illusion
- People who talk of cancer clusters are often idiots. I hate hearing about people changing their gambling behavior based on streaks as well. Given a large enough sample, clusters are actually quite common.
- Relying on tiny sample sizes
- I know three nonsmokers who died of lung cancer within a 2 year period. All three had been schoolteachers at some point in their lives. It could be tempting to say "schoolteachers have a higher risk of dying of lung cancer," since I only know one other schoolteacher. But I know better - a sample size of 3 (or even the larger pool of everyone I know) proves nothing. Anecdotal evidence is worthless, and you should know this.
- Misleading percentages
- People like to selectively quote certain stats to promote a conclusion. I don't believe marijuana is a gateway drug, but it sure sounds like it when you hear that "12 to 17 year olds who smoke marijuana are 85% more likely to use cocaine." Despite the fact that this gateway theory has been debunked by many. After all, I could easily claim that water causes cancer, since 100% of all cancer patients have a history of drinking water. Even better are the people who claim that power lines cause leukemia, based upon a 69% increase in incidence. A closer look reveals that "a 70 percent increase in leukemia means that the 1 in 2,000 risk of leukemia becomes a 1 in 1,200 risk," which is so ridiculously small that nobody should care. Learn to distrust simple percentages by fear-mongers and anyone trying to promote an agenda.
- Confirmation bias
- If you actively looking for a statistical link, and you consider only evidence that supports your theory, your statistics are pretty worthless. Everyone in politics is guilty of this, as are many people in the scientific community. Yet, people still fall for this crap from unscientific sources. I don't understand it. [more]
