Shere Hite's "Women and Love" is often used to illustrate sampling problems (particularly with response rates). There's a pretty good overview here.

Emily, this is also historical/presidential election related, but here goes.

During the early 20th century, the Literary Digest regularly conducted a widely known and respected presidential poll based heavily on telephone books and auto registration. In 1936, this poll predicted that Republican Alf Landon would soundly defeat Democratic incumbent Franklin Roosevelt.

Roosevelt won in a landslide. In fact, Landon won only 2 states and 8 electoral votes. The problem? By emphasizing persons with phones and cars, the sample was badly skewed toward persons with higher incomes (i.e., more likely to vote Republican).

The Literary Digest went under not long after, and it's widely believed that the poll was responsible.

A little more on the subject

tommyrot, that would be good as another example of questions affecting conclusions -- I'll look for it. I'm also looking for examples of bad statistical methods -- surveys where they ask only union members in Massachusetts about their politics and then generalize to the United States, for example.

Strega, Fred Pete, those are both exactly the kind of thing I'm looking for. Thank you!

Some of Kinsey's work had problems there - asking questions about sexual preferences, fidelity, etc., in the presence of spouse and/or kids, for example.

Oh! Oh! Oh!

There was that very crappy article in Time (or maybe Newsweek) about how 80% (or somesuch) of the internet was devoted to porn. It had the scary cover of the child recoiling in horror from a computer screen.

But where they got that percentage was from BBSs. Then they generalized BBSs to the internet. Very stupid. They also refused to submit their study for peer review, so it looks like objectivity was the last thing they were after.

You should be able to google and find lots of stuff, as it was quite controversial amongst online folks.

When I was in college, the radio station I worked with did a survey to see who was listening (not many, it turned out). What was interesting was the degree of falsification of surveys that occurred because the cold calling was boring and frustrating when dealing with a small rural town who didn't care what the little college radio station was doing.

I admit it, I was one of the ones falsifying, but my fake answers were that most people didn't listen to us or never heard of us, a few "expected" people did listen (high schoolers and a couple of college students), plus at least one outside-the-curve person, a granny who dug rock n roll. The other person--who was caught cheating--had all her fake answers the same, making it obvious she'd just decided to fill out all the sheets and be done with it. I at least put some creativity into it. I never admitted the cheating, and I wonder if anyone else did the same.

Don't know if that's relevant, but I enjoy the story.

So the lesson is that not just anyone can make up data well - IOW it's harder that it looks.

The SMART cars have been cleared for U.S. import.

[link]

Don't know about the wisdom of buying them as Mercedes isn't the behind the effort.

So the lesson is that not just anyone can make up data well - IOW it's harder that it looks.

Also, have creative types as well as number crunchers on your Statistical Information Interpretation team. I would have been such a good advertising/PR person.