Distribution of software development skill in the entire population is probably normal. However, developers are not the entire population, they are a tiny slice of the right tail. So the distribution is more like an exponential distribution, which has the mean smaller than the median.

The performance of doctors follows the normal distribution. I think that extrapolating to programmers is not a huge stretch.

    It used to be assumed that differences among hospitals
    or doctors in a particular specialty were generally 
    insignificant. If you plotted a graph showing the 
    results of all the centers treating cystic fibrosis—or 
    any other disease, for that matter—people expected that 
    the curve would look something like a shark fin, with 
    most places clustered around the very best outcomes.
    But the evidence has begun to indicate otherwise. What
    you tend to find is a bell curve: a handful of team
    with disturbingly poor outcomes for their patients, a
    handful with remarkably good results, and a great
    undistinguished middle.

http://www.newyorker.com/archive/2004/12/06/041206fa_fact?cu...

The difference between doctors and programmers is that programmers get to build/leverage tools that are abstractions of other tools, hence you can have orders-of-magnitude differences in productivity between programmers who use the best tools and those who don't. I assume that there's not that much variation between the way two different doctors carry out the same task like there is with programmers.

Actually, doctors do have order-of-magnitude tools for improving certain metrics, such as recovery and infection rates: http://www.newyorker.com/reporting/2007/12/10/071210fa_fact_...

Actually, a lot of natural phenomena does NOT follow a normal distribution. The problem is that most people assume it does anyway, and that's where a lot of errors come from.

Statistics requires a lot of correct assumptions in order to be accurate; sadly, most people overlook or fail to check their assumptions.

The thing is, natural phenomena doesn't follow normal distribution, but the sum of non-normal variables, will approach a Gaussian distribution, as per the Central Limit Theorem[1][2].

The problem is people assuming a normal distribution on a single variable with unknown distribution.

[1]: http://en.wikipedia.org/wiki/Central_limit_theorem [2]: http://stats.stackexchange.com/questions/22387/why-it-is-oft...

Yes, that's correct. If anyone wants to know more, an elementary statistics book will teach you all about the central limit theorem and the law of large numbers. Very useful things to know.

Khan Academy has a good lesson on the law of large numbers http://www.khanacademy.org/math/probability/v/law-of-large-n...

mlvljr, for some reason your post is showing as "dead". I'm not sure why??

Do you have any reason to assume that programming competence is a "natural phenomenon" ?

We're not talking about the statistical central limit theorem here, we're talking about people who are passionate about programming and spend years working on their craft.

You can only invoke the normal distribution thing when you're talking about an outcome that is the average of many independent quantities.

Research into the performance of medical doctors shows that their skill follows the normal distribution: http://www.newyorker.com/archive/2004/12/06/041206fa_fact?cu...

A lot of genetic traits follow normal distribution due to the way genes are combined. It doesn't follow at all from this that human skill follows a normal distribution, especially looking at the whole population. It seems to be more like a power law distribution.

Well, I don't know how you got that assertion about the normal distribution. But, one reason you could think it's not a simple normal distribution is that computer science classes often have bimodal grade distributions.

Software development is not a natural phenomenon.

Software development is not a D&D stat block item. No reason to believe it is even quantifiable (although I like to think some developers approach writing code the same: rolling dice). If it were quantifiable, what would be the metric?

Given that there are N animals of size X, how many animals are there of size X/10?

The phenomena are normally distributed, but there are often steep threshold effects in performance. For example, no number of teacup poodles is sufficient to win a dogsled race. Most dogs have above average completion times on the Iditarod race.

Which is not to say there is no market for chasing frisbies.