[ View menu ]

Are statistics books impossible to understand?

Filed in Gossip ,Profiles
Subscribe to Decision Science News by Email (one email per week, easy unsubscribe)


Decision Science News has long been influenced by Phil Greenspun’s observations, such as this one:

Any sufficiently complicated C or Fortran program contains an ad hoc, informally-specified, bug-ridden, slow implementation of half of Common Lisp.

which is known as Greenspun’s Tenth Rule. Trivia buffs will know that there weren’t 9 before it.  For those who aren’t programming nerds, this rule contains elements of: don’t reinvent the wheel, use the right tool for the job, and stand on the shoulders of giants, plus a hint of “my favourite programming language is better than yours”.

While doing some research on consumers’ (in)ability to detect differences between products, we came across a 1987 Greenspun & Klotz paper “Audio Analysis VI: Testing Audio Cables”. The authors are engineers who put participants to a single-blind test of sound quality transmitted through high and low quality audio cables.  (They found that people can tell two cables apart, but that the expensive cable was preferred only 15 times out of 28.) This seems to be straight-ahead experimental psych, so we were a little surprised when the authors, instead of reporting basic significance tests chose to rederive and compute them from scratch (in Common Lisp, of course) and document the story in a two-page appendix. Perhaps it was the Common Lisp that reminded us of it, but we couldn’t help drop Greenspun a line and ask:

From: Dan Goldstein
Sent: Monday, April 19, 2008

Dear Phil,
I wonder if you violated the spirit of your 10th law in your 1987 Computer Music Journal article by coding, and describing in a two-page appendix, what could have been replaced by reporting a simple significance test. Such tests are pre-coded in every stats package and in the back of every stats book. This is not a criticism, I’m just wondering why you went to all the trouble.

We were amazed and delighted to get a reply:

From: Philip Greenspun
Sent: Monday, April 21, 2008

Several issues…

1) most of the readers are not stats experts

2) stats books are virtually impossible to understand (I tried a bunch
and gave up, even though I was an undergrad math major!)

3) I was able to explain what I was doing using simple probability
laws that most of the readers would have understood, sharing some
insights from Al Drake, a great probability teacher in the MIT EECS

So many people use stats tests and don’t understand them.

We see where Phil is coming from, but wonder if he isn’t being a bit hard on statistics.