Stats Made Easy

Practical Tools for Effective Experimentation

Thursday, April 27, 2006

Repeated measure versus true replication

It was my pleasure yesterday to teach design of experiments (DOE) at the Ohio State University's Fisher College of Business for their Web-Leveraged Six Sigma Black Belt program. It turned out that one of the participants, Dennis Dubose -- a statistician with Sumco USA, teaches DOE also, so we compared notes. He shared a nice explanation of the difference between simply repeating measures versus performing a true replication of the system setup for purposes of estimating error. Imagine you flip a coin and it comes up heads. You ask a colleague to look at it and call out the result. Then another colleague is asked to observe the coin and state which side came up. This is a repeated measure. A true replication is accomplished only by re-flipping the coin.

Wednesday, April 19, 2006

If good news outweighs the bad, is it fair to say that on average one need not worry?

In response to studies that demonstrated chronic release of mercury vapor from amalgam fillings during chewing and brushing, the National Institute of Dental and Craniofacial Research (NIDCR) sponsored the Children's Amalgam Trial (CAT). Today's issue of Journal of American Medical Association (JAMA) unveiled the results of CAT, which is summarized by this intriguing quote from the lead author:

"The studies indicate that on average we probably don't have much to worry about..."
(David Bellinger, Harvard Medical School)

Saturday, April 15, 2006

New study reveals that bad news not only travels fast, but it gets amplified along the way

In my prior career as a quality professional I'd heard that positive word-of-mouth about a company's products and services may be passed along to three people, whereas bad news transmits much further -- to at least twice as many individuals. This is a powerful inducement for businesses to go all out to avoid bad will. A new study, done in part by the prestigious Wharton School, adds a great deal of insult to the injury caused by customer complaints: As reported in the April 17th Business Week magazine and previously by CBC News, the stories about bad products and/or service become magnified with each re-telling, so that people down the line are up to 5 times as likely to avoid the business in question as the original unhappy customer. Evidently Hollywood is well aware of this phenomenon, because when they know a major movie will be a bomb, they roll it out to as many theaters as possible to maximize revenue the first weekend before the bad buzz can kill it.

Sunday, April 09, 2006

How many firkins in a hogshead?

While teaching mixture design for optimal formulation at two chemical companies the last two weeks, I was surprised to see that they still specify ingredients as pounds per hundred (pph)-- bsed on their main component. This is really old-fashioned, not only for the use of English units of measures,* but because it does not account for the dilution effect. For example, a Stat-Ease client making a construction caulk ran a central composite design, a response surface method (RSM), that varied the filler from 100 to 250 pph and plasticizer from 50 to 100 pph based on 100 pounds of polymer. They kept the other 57 pounds in their recipe constant in terms of relative proportions for the ingredients. We did some simple calculations with a spreadsheet to make the bargraph shown on a 100 percent weight basis. Seeing how everything (depicted by colored segments) varied by concentration (not just the two ingredients they intended to change), it now became clear to the client that they would best re-design their experiment as a mixture.


*(Hired into the petroleum industry as chemical engineer in 1975, I quickly learned that there are 42 gallons in a barrel. Further research into English units of volume revealed that some number of firkins would fit into hogshead -- I forget now. Just this last week I discovered a new measure of volume at a brew pub -- they sold a "growler" of beer (64 fluid ounces) -- a grrrreat deal!)