Stats Made Easy

Practical Tools for Effective Experimentation

Saturday, August 26, 2006

Weighty words against decade-long study

"I feel like the researchers were trying to manipulate their data to match their conclusion." Linda Bacon of the University of California at Davis said this about a 10-year study of more than half a million slightly overweight U.S. adults. The researchers, led by the National Cancer Institute,* concluded that, once these somewhat "gravity-challenged" Americans reached age 50, they were 20 to 40 percent more likely to die in their next decade than those in their cohort who maintained a healthy weight. A body mass index (BMI) between 25 to 29 is considered overweight -- above a BMI of 30 you fall into the obese category, for which there's seemingly no question about associated health problems. However, Glenn Gaesser of the University of Virginia at Charlottesville joins critic of this recent study by saying "They are presenting the data in a way that paints overweight and obesity in the worst possible light." Paul Ernsberger, associate professor of nutrition at Case Western University, piled on with this comment: "They're standing on their heads squinting at it backwards trying to make it fit." The criticisms stem from the reliance by researchers on participant's recollections, which can be very unreliable. Critics also note the high number of people excluded from the final analysis, thus introducing possible bias. Furthermore, previous studies, especially one by the Centers for Disease Control and Prevention, indicated a beneficial effect of being slightly overweight. In any case, this study provides some food for thought by the large numbers of us baby-boomers that fall into this category of being a bit bloated after all these years of good living in the USA.

*New Study Shows That Being Overweight at Middle Age Can Be Harmful

Sunday, August 20, 2006

Lake Wobegon Effect in ACTion

In 1987 a survey of educational departments resulted in all 50 states claiming their children to be above average in test scores for the USA. This is a common fallacy that is defined in Wikipedia as the Lake Wobegon Effect after the mythical town in Minnesota, where according to author Garrison Keillor, all women are strong, the men good looking, and their children above average. I see the Lake Wobegon Effect manifested in reports on this year's ACT college-assessment scores. My daily newspaper, the Saint Paul Pioneer Press, reported on August 16 that "Minnesota...[is] best at college test." They based this ranking on the percent passing all four benchmark scores.* Counteracting this, State Superintendent Elizabeth Burmaster reports "Overall, Wisconsin beat the nation on the ACT.” Then again the Deseret Morning News reports "Utah tops U.S. on ACT." I cannot see how any of these self-congratulatory reports can be inferred from the actual 2006 Average ACT Scores by State. However, I suppose someone might explain the necessary contortions for patting themselves on the back. The ACT statistics actually back up the Boston Globe pronouncement that "Massachusetts' class of 2006 scored the highest of any state on the ACT math exam and scored behind only Connecticut on the overall exam." However, it turns out only 13 percent of high school graduates took this test versus well over 50 percent in Minnesota, Wisconsin and other midwestern USA states that rely on the ACT for screening college applicants. (I believe that the eastern states use the SAT exam.) Sorting the ACT scores on the average composite score (a handy feature offered at their web site), I see that all seven states ahead of Minnesota had fewer that 20% of their graduates taking the test, so I have no qualms saying that my home state is the smartest of all!

*PS. I am very alarmed at the poor results for science (37%) and math (52%) relative to english (76%) and reading (62%). :(

Thursday, August 10, 2006

Stats that will be the death of you

The most recent issue of National Geographic magazine features a unique graphic that visualizes your odds of dying by various causes. You can see the chart and underlying statistics at a website maintained by the National Safety Council (NSC). The total odds of dying by any cause in the USA (and possibly in other countries too!) are 1 in 1 or 100 percent. I suggest you get over this without too much thought (maybe later!), and check out the less likely things that could kill you. For example, I see that the odds of dying by earthquake or flood are roughly the same -- less than 1 in 100,000. I'd be more worried living along a coastline in an area riddled with fault lines (take heed you Californians!) than here in Minnesota, although we've got more than our share of lakes and rivers. At about 50,000 to 1 odds, lightning and bee stings create more concern for survival -- ouch! The NSC reports a 1 in 5000 chance that an American will die in an air accident versus only about 1 in 2 million odds that a venomous reptile will put an end to things. However, they do not calculate the deadly combination of Snakes on a Plane. That would be very scary!

"More people die on the last day of their life than on any other day."
–– Statistic purportedly published by the Houston Post on 20 July 1989

Sunday, August 06, 2006

Randomization to rebel against deterministic forces

As a practitioner of statistically designed experiments I am a big believer in randomizing the order of runs specified by standard templates. This counteracts time-related lurking variables such as material degradation, machine warm-up and increasing ambient temperature. Thus it was a shock to read this passage in a book I'm reading that's titled The Traveler: "...anyone who used random numbers to guide his life should be hunted down and exterminated." It turns out that the good guys in this novel carry RNG's -- random number generators -- to confound the forces of evil. If you harbor paranoia about the ever-increasing intrusions on privacy by government and big business, you will like this book. Otherwise it comes across as "New Age Nonsense," as noted by the Washington Post. Oh, and by the way, is there any significance to the fact that the publisher is Random House?