is visiting from India to do research, and he took some time out to
give a colloquium talk. We love big audiences, so the boss and I
bribed students to attend the talk; Dr Kannan
provided pizzas and I gave extra credit points for notes. The
talk concerned an improved model for stepped stress levels in
accelerated life testing; it was right at the limit of what my
students–especially the biologists–could understand. Still,
several students had good questions after the talk, indicating they had
learned something. The
notes turned in for extra credit told pretty much the same story; they
didn’t grasp the whole talk, but they did pick up a few good
ideas. So everyone was happy, and we declared the effort a
success.
Coincidentally, my biologists have been working on a simple diagnostic
model in their statistical methods class. They’re analysing data
from a Dutch study that explored a possible link between lung cancer
and pet birds in a large group of cigarette smokers. In addition
to some basic demographic data, there is data on each individual’s
smoking history, but only a yes/no indicator showing whether the
individual has a pet bird. When a logistic regression model is
fitted to the data, the pet bird indicator is highly significant, with
a large odds ratio, suggesting that owning a bird increases the risk of
lung cancer about as much as 25 years of smoking!
Naturally, these two threads came together yesterday when several of my brighter students thought about their results: how long had the smokers been exposed to the birds?
We have no information about that at all, and now our conclusions are
suspect. This was such a nice case study, but now it’s just
junk. BUT,
I’ve got some students who’ve done some good critical thinking–like
practicing statisticians–and that’s the whole point of teaching the
course. Dr Kannan and I have done the happy dance, and will get
more pizza for the fall semester.
The Random Texan 
Leave a comment