Today's Mini-Lecture: Statistics

There are lies, damn lies, and statistics.

99% of statisticians surveyed agree with the above statement.

A New York Times/CBS poll this week shows that 45 percent of Americans believe Mr. Hussein was "personally involved" in Sept. 11, about the same figure as a month ago.

-- The impact of Bush linking 9/11 and Iraq

Statistics, my friends, as anyone with some rigorous math behind them will tell you, are mostly bullshit. Why?

It's very easy to make statistics say what you want them to say. Without even trying.

For example, suppose you conduct a poll of 100 people from the same region of a Bible Belt Midwestern town, in a consistently-Republican-voting county. And of those 100 people, 45 of them answer your question, "Do you agree with US government reports that Iraqi dictator Saddam Hussein was directly involved in the 9/11 attacks?" with a 'Yes'. 55 do not.

You then can state, in writing, that 45 percent of Americans believe Mr. Hussein was "personally involved" in Sept. 11. You asked 100 people, and 45 said 'Yes'. That's 45%, right?

Yes and no. You only asked 100 people. Was that 100 people a representative sampling of all Americans? No. Is 100 people enough to get a representative sampling? 100 people is nothing, when they're all from the same region, and you're talking about America, with population groups whose subdivisions (where people frequently line up along ideological lines) are larger (substantially larger) then the set of people you surveyed.

What you've just done is conducted a completely non-scientific poll. No prep work to determine how representative your sample is. No large population surveyed several times, with the results then combined. No standard deviation calculations. No sharing of your data with third-party analysts to validate your results. You've just collected 100 useless responses. But all the article in your local paper says is "45% of Americans believe Mr. Hussein was "personally involved" in Sept. 11."

And the average reader might even believe it. Because numbers don't lie, do they?

No, they don't lie, but they don't always tell the whole truth, either.

So, the next time you find yourself face to face with a statistic, ask the following questions:

1. How many people were surveyed?
2. What demographics of race, religion, economic position, location, country of origin, and political party were surveyed?
3. Could this be just a bunch of hooey?

And pay close attention to the answers.