Lies, darn lies, and statistics

[Jeffery]

Statistics are powerful tools, and they should be used responsibly. It is unfortunate that many political propaganda papers and talkshows are using them irresponsibly, and sometimes making conclusions that are downright preposterous, based on statistics and probability.

It's amazing how some radio talkshow guests will quote statistics and jump to incredible conclusions, when there is a much more obvious explanation to the phenomenon observed. They fail to apply the principles of logical reasoning (such as that correlation does not equal causation) to the data.

One example is the Advocates for Youth, who did a data study of teen fathers in 2003. The researchers noted that they were more likely to have "low self-esteem" than their peers were. Therefore, said the researchers, the teens "turn to as a way to boost their self-esteem."

Now I don't know what these "researchers" were smoking, but isn't it obvious that the study subjects (high-schoolers) did NOT become fathers on purpose? I mean, that would be like trying to find out what motivates people to get in car wrecks!

Apparently the researchers for policy analyst Dr. Michael Males in Washington, DC were smoking the same stuff as the Advocates, because in the 2004 policy-informing paper concerning teen parents, he wrote: "The best prevention strategy is to reduce the number of young women and men in circumstances from which escape through early parenthood is desirable."

I could give other examples, such as the oft-quoted assessment that shacking up before the official marriage is known to "cause" divorce after the couple marries later on. Well, they may have found a correlation, but that dos not mean that it is a cause-and-effect relationship. The type of people who shack up in the first place are the type who are more likely to get divorced later on.

Anyone can quote a statistic, but to abuse the numbers to show cause-and-effect relationships that aren't there is totally fallacious. It goes back to the old fallacy of correlation equalling causation, and this is the reason for the saying: "If you torture a statistic long enough, it will confess to anything."


Jeffery

[TheDoorbellRang]

1. I took a course on statistics in college

and one of our textbooks was "How to Lie With Statistics." It was quite eye-opening; too bad all kids aren't required to take courses on some common ways info is presented as disinfo. Correlation and causation are mixed up so often it's a wonder they're not listed as synonyms in the dictionary.

[Jeffery]

2. That's where the saying comes from...

"figures don't lie, but you can lie with figures."

What other statistical fallacies did you remember learning about from that textbook, other than correlation-causation?

[Toots]

3. When I was in fourth grade my teacher gave us an example of how to twist the facts

Russians and Americans in a car race. Americans won the race. Russian papers reported, Russians come in second while Americans come in next to last...Rush Limbaugh learned this lesson very well.

[TheDoorbellRang]

4. A really basic one we see every day

is distortion using charts. Use the results shown on the DFA poll right now. Altho I'm absolutely thrilled with the results (Al Gore is winning), the chart showing each candidate's percentages is visually deceiving. Note of caution, the DFA folks probably don't intend this distortion, but it's become so entrenched no one even notices it anymore.
http://democracyforamerica.com/

The chart, if depicted accurately, would show space equivalent to 100%. Instead, the chart after Gore's +31% is cut off at about where 35% would be, instead of where 100% would be. This has the effect of showing huger differences between each candidate than actually exist. If I knew how, I'd reproduce two charts so you could see what I mean, but hopefully I explained it alright. The effects aren't quite so nefarious here, but I've seen this effect used quite often thru the years to exaggerate differences that aren't as extreme as the chart would lead one to believe if you were just reading the numbers. Eg, if you set up a bar chart showing 50 people supported your POV vs 48 against it, the disparity is going to be visually twice as large if your chart only goes up to 51% as opposed to a more accurate 100%. And bar charts don't lie, do they? ;(

[Jeffery]

5. wrong units

There's also the use of inappropriate units, such as a can of soup advertising "only 1.6 grams of sodium" instead of 1600 milligrams, thre standard unit to measure sodium. The global warming deniers always say that "human activity could not have caused more than 2.5*C of warming so far", and notice they say 2.5*C because it sounds smaller than 4*F.

[TheDoorbellRang]

6. Remember the argument they used for the flag burning amendment?

Supposedly we needed it because there had been (gasp) a 150% increase in incidents from the previous year!1!1! They avoided using the actual data, which showed that the previous year there had been a total of 4 incidents, and the year they tried to pass it there had been 6.

A real "burning" issue.

[Jeffery]

7. That is called...

...the fallacy of small sample size.

I am working on an article about the abuse of statistics. I'll have to use that example!