Re: OT: The Geek defense

On Thu, 28 Feb 2008 11:10:59 -0000, tim <TimJ@xxxxxxxxxxxx> wrote:


There has been quite a lot of research on the question of programming
aptitude, because of the fact that a lot of people just don't ever "get
it". Originally programming courses had high failure rates. Academics
were tearing their hair out. Then they moved the goal posts so that they
pass, but they still can't program. The result is a significant number of
people who have a degree in CS but can't even do a simple programming
challenge.

Aptitude tests, though often sold for a high price, have proven to be
pretty useless. NO better than IQ tests, which are not that useful for
these purposes.

There was one interesting piece of research (I have the link on my other
computer) which showed fairly convincingly that most people do not
actually think in a logical way. They function by simple verbal pattern
matching, memorisation, intuition, and by copying other people. These
individuals, almost independently of how intelligent they are, cannot
program.

A minority of people think in a different way, constructing mental models
of things and people, and then operating on those models. These people can
program. As you pointed out, you don't need a high IQ to program, though
very complex programs are another matter.

My found the same when evaluating tests. I started with six programmers of known ability,
all pretty high but some higher than others. I assigned each a numeric score on ability. I
gave them a Programming Aptitude Test and ran a regression between the test and ability
scores. It was around .55, little better than random. I did the same with a dozen other
tests. One was on trivia, one I wrote, one was an IQ test, some were on non-computer
technical fields. In all cases, there was no correlation.

Then I tried one section of the military General Classification Test (GCT) called Spatial
Relationships. It was multiple choice with no words, just pictures of two dimensional
patterns you had to mentally fold and rotate. Bingo. The correlation was over .90.
Notably, the highest scorers finished fastest, in as little as half the allotted time
which was 12 minutes for 20 questions. The instructions said no one was expected to finish
in 20 minutes.

A friend gave it to his larger and more diverse team of 70 programmers, who were college
student stringers (part-time workers). Bingo again. The correlation was even better,
around .95. His full time systems programmers did it in half the time and got 100% right.

I started using it on applicants who were experienced programmers. Most couldn't finish in
12 minutes, and got quite a few wrong. A waitress who knew nothing about programming did
it in half the time and got 100% right. Three months later she was good; six months later
she was a star.

The traditional job interview is worthless for assessing logical ability. It measures
personality and social skill. I can tell more from glancing at one page of code than from
a half hour of conversation.

One can view Aspergers/Autism as an extreme manifestation of this
"systematising" approach to the world. Whole books have been written
advocating this view. If this is true, it would not be surprising if the
associated personality traits were more common in the programming world.

This is not necessarily a bad thing. The other extreme away from the
autistic person is the "social butterfly" for whom nothing exists but
social relationships, and who cannot deal in any way with things that
require the exercise of logic and reason. These are different ways of
looking at the world with strengths and weaknesses.

On the radio last week a relationships expert posed and answered the question: who is the
better mate, a geek or the most popular person in the class? The answer was the geek, but
not for the expected reason that the popular person was already taken. The reason she gave
is that the popular person is "self-monitoring", presenting an image the listener wants to
see. It changes for each listener. It's not who the popular person really is. The geek
isn't so artful; what you see is real.

.

Quantcast