Two recent articles worth reading:
Decoding your email personality, NYT: (link)
This describes the field called "forensic linguistics", the use of statistics to link (or delink) writings from their writers. Mark Zuckerberg is being sued again, this time by one Paul Ceglia who has produced emails (saved as Word documents) that show that he may own half of Facebook. Mark hired a linguist who argues that he can show that those emails are "unlikely" to have been written by Mark. Language Log has some additional insights here.
This is another fascinating example of statistics used in real life. I find it hard to believe that in this case, it would be possible to say with acceptable confidence whether Mark is the real author or not.
Counting calories? Your weight-loss plan may be outdated, NYT (link)
Some new conclusions from a long-term study that tracks people for decades to try to understand the link between food consumption and weight gain/loss. I'm going to need to find time to read the underlying study but certain aspects of this class of health studies have always bothered me.
First, the results are often expressed as: milk is good, cookies are bad, butter is ok, etc. This presentation gives the impression that all components of a diet are independent of each other. If milk will rise my weight by 0.2 percent and cookies by 0.3 percent, then if I take both, my weight would go up by 0.5 percent. Unfortunately, things are not so simple. In this example, cookies and milk are often consumed together; someone who chooses to eat fewer cookies is likely to also drink less milk.
Second, these studies report averages for the entire population but surely there are subgroups that differ from the average person.