Statistics

Let me tell you a story about my graduate thesis. It was entitled The Partitioning of Benzo(a)Pyrene in the Pregnant Laboratory Mouse. What I was working on was how this chemical and it's metabolites move through the placenta and so I took tissue samples from the mother, fetus, and placenta of lots of mice and then compared the concentration of this chemical in all of the tissues. It was not terribly important research and was not published anywhere.

When I was writing up my results I went back to my Biostatistical Research Analysis class notes and figured out which mathematical test I needed to use to find out if there was a significant difference between the groups of samples. I found that there was and took the results to my mentor, who looked at them and told me I had it backwards.

A few weeks later I presented my research to the department and the statistician stood up afterward to tell me I had it backward... which I now did, because my mentor had me switch things around.

My point? If you're going to suspect one piece of any research, suspect the statistics. Lots of times even the big scientific honchos don't know what they are doing and one little mathematical slip someplace can make a huge difference. If the work is important to you, look at the numbers that the statistics are derived from and see if the conclusion makes any sense.

The other piece of knowledge I derived about statistics from that period in my life was this: correlation does not equal causation. For instance, you might notice a strong correlation between school attendance and owning an MP3 player. You could reach the conclusion that people buy MP3 players because they go to school. That wouldn't be right, would it? It's more likely that people of school age are more likely to be interested in both music and technology... Next time you hear about a correlation between eating cheeseburgers and howling at the moon, you can take it with a grain of salt.

Links:

FedStats: Government statistics. You can probably trust them. Really.

A statistics textbook online.

A little shorter and easier explanation.

Word Of The Day

Margin of Error: Since the president elect used this phrase in his weekly address, I though it would be a good one to define. This represents how precise a number is. If the margin of error is small, that means the scientist is pretty sure she has it close. If it is large, there is a lot more "wiggle" in the number. This number is often used with the words, "plus or minus," in front of it. For instance, "the average grade of the class was 70%, plus or minus 20%." That's a huge margin of error. The actual average could be anywhere from an A to an F. You wouldn't want to have much faith in that number.