Here’s the thing about statistics in this day and age: anyone can enter data into Microsoft Excel and, with a few clicks of the mouse, execute any number of statistical procedures and end up with any number of statistical outputs. Even a kindergartener could do it, right?
Indeed, as long as that kindergartner was skilled at detecting and correcting errors in the original data set, knew which specific statistical procedure was most relevant to a particular experimental objective, and could deliver a meaningful interpretation of the resulting output. That means people like Milton Loyer ’67 are still awfully handy to have around at places like the Penn State Fruit Research and Extension Center in Biglerville, Pennsylvania, where he evaluates research on fertilizers, pesticides and crop management strategies.
Loyer spends much of his time combing through enormous sets of data, sometimes up to 10,000 lines at a time, in which there are bound to be keystroke errors (researchers, especially grad students, aren’t always as meticulous in this regard as a statistician would like). Sometimes the errors are obvious, like when a misplaced decimal point throws a figure off by one or more orders of magnitude. Other times, the errors are much more sneaky and subtle, revealed only in patterns that rouse Loyer’s suspicions and send him on the hunt for an explanation.
One example was a set of growth measurements from the orchard, in which the shoots appeared to be longer and shorter on alternating trees in a constant pattern – first longer, then shorter, longer, shorter. Loyer wandered out to the field to ask the workers how they’d collected the data. The orchard is big, they told him, and the job is boring. To keep things interesting, two of them had alternated roles at every tree. One would measure the shoots, and one would write down the numbers. They had used different measuring sticks, and on closer inspection, Loyer found that one of the sticks had an extra 2 millimeters on the end below the true zero mark. The mystery was solved and the data were adjusted accordingly. Excel can’t do that kind of thing.
After graduating with a math degree from EMU, Loyer went on to earn his PhD at Montana State University. He later taught at Messiah College in Grantham, Pennsylvania (EMU math professor Owen Byer was a student of Loyer’s there in the mid-’80s). In addition to his work as a statistician, he runs the archives of the United Methodist Church’s Susquehanna Conference at Lycoming College in Williamsport, Pennsylvania, where he is also an adjunct math professor. He also finds time to manage his own affordable housing ministry in Williamsport, do some occasional statistical consulting, and, recently, “after much thought, several false starts, and helpful insight from colleagues” explain a mind-bending, paradoxical result that turned up in the statistical analysis of an apple defect study.1
The real breakthrough came after a summary of this paradox – which he’s trying to popularize as “Loyer’s paradox” – was booted from Wikipedia on grounds of being impossible. The insight that led to Loyer’s explanation of the paradox was sparked during subsequent discussion of the matter with Wikipedia’s statistics editor (although the solution has been published, the entry has yet to be reinstated on Wikipedia). That’s something Excel won’t do either. — Andrew Jenner ’04
1. Loyer, Milton W. and Gene D. Sprechini. “Can the Probability of an Event Be Larger or Smaller Than Each of Its Component Conditional Probabilities?” Chance: A Magazine for People Interested in the Analysis of Data 24.1 (2011): 44-53. Print.