Cannot Measure Productivity

29 August 2003

We see so much emotional discussion about software process, design practices and the like. Many of these arguments are impossible to resolve because the software industry lacks the ability to measure some of the basic elements of the effectiveness of software development. In particular we have no way of reasonably measuring productivity.

Productivity, of course, is something you determine by looking at the input of an activity and its output. So to measure software productivity you have to measure the output of software development - the reason we can't measure productivity is because we can't measure output.

This doesn't mean people don't try. One of my biggest irritations are studies of productivity based on lines of code. For a start there's all the stuff about differences between languages, different counting styles, and differences due to formatting conventions. But even if you use a consistent counting standard on programs in the same language, all auto-formatted to a single style - lines of code still doesn't measure output properly.

Any good developer knows that they can code the same stuff with huge variations in lines of code, furthermore code that's well designed and factored will be shorter because it eliminates the duplication. Copy and paste programming leads to high LOC counts and poor design because it breeds duplication. You can prove this to yourself if you go at a program with a refactoring tool that supports Inline Method. Just using that on common routines should allow you to easy double the LOC count.

You would think that lines of code are dead, but it seems that every month I see productivity studies based on lines of code - even in such respected journals as IEEE Software that should know better.

Now this doesn't mean that LOC is a completely useless measure, it's pretty good at suggesting the size of a system. I can be pretty confident that a 100 KLOC system is bigger than a 10KLOC system. But if I've written the 100KLOC system in a year, and Joe writes the same system in 10KLOC during the same time, that doesn't make me more productive. Indeed I would conclude that our productivities are about the same but my system is much more poorly designed.

Another approach that's often talked about for measuring output is Function Points. I have a little more sympathy for them, but am still unconvinced. This hasn't been helped by stories I've heard of that talk about a single system getting counts that varied by a factor of three from different function point counters using the same system.

Even if we did find an accurate way for function points to determine functionality, I still think we are missing the point of productivity. I might say that measuring functionality is a way to look at the direct output of software development, but true output is something else. Assuming an accurate FP counting system, if I spend a year delivering a 100FP system and Joe spends the same year delivering a 50FP system can we assume that I'm more productive? I would say not. It may be that of my 100FP only 30 is actually functionality that's useful to my customer, but Joe's is all useful. I would thus argue that while my direct productivity is higher, Joe's true productivity is higher.

Jeff Grigg pointed out to me that there's internal factors that affect delivering function points. "My 100 function points are remarkably similar functions, and it took me a year to do them because I failed to properly leverage reuse. Joe's 50 functions are (bad news for him) all remarkably different. Almost no reuse is possible. But in spite of having to implement 50 remarkably different function points, for which almost no reuse leverage is possible, Joe is an amazing guy, so he did it all in only a year."

But all of this ignores the point that even useful functionality isn't the true measure. As I get better I produce 30 useful FP of functionality, and Joe only does 15. But someone figures out that Joe's 15 leads to $10 million extra profit for our customer and my work only leads to $5 million. I would again argue that Joe's true productivity is higher because he has delivered more business value - and I assert that any true measure of software development productivity must be based on delivered business value.

This thinking also feeds into success rates. Common statements about software success are bogus because people don't understand WhatIsFailure. I might argue that a successful project is one that delivers more business value than the cost of the project. So if Joe and I run five projects each, and I succeed on four and Joe on one - do I finally do a better job than Joe? Not necessarily. If my four successes yield $1 million profit each, but Joe's one success yields $10 million more than the cost of all his projects combined - then he's the one who should get the promotion.

Some people say "if you can't measure it, you can't manage it". That's a cop out. Businesses manage things they can't really measure the value of all the time. How do you measure the productivity of a company's lawyers, its marketing department, an educational institution? You can't - but you still need to manage them (see Robert Austin for more).

If team productivity is hard to figure out, it's even harder to measure the contribution of individuals on that team. You can get a rough sense of a team's output by looking at how many features they deliver per iteration. It's a crude sense, but you can get a sense of whether a team's speeding up, or a rough sense if one team is more productive than another. But individual contributions are much harder to assess. While some people may be responsible for implementing features, others may play a supporting role - helping others to implement their features. Their contribution is that they are raising the whole team's productivity - but it's very hard to get a sense of their individual output unless you are a developer on that team.

If all this isn't complicated enough the Economist (sep 13-19, 2003) had an article on productivity trends. It seems that economists are now seeing productivity increases in business due to the computer investments in the nineties. The point is that the improvements lag the investments: "Investing in computers does not automatically boost productivity growth; firms need to reorganize their business practices as well". The same lag occurred with the invention of electricity.

So not just is business value hard to measure, there's a time lag too. So maybe you can't measure the productivity of a team until a few years after a release of the software they were building.

I can see why measuring productivity is so seductive. If we could do it we could assess software much more easily and objectively than we can now. But false measures only make things worse. This is somewhere I think we have to admit to our ignorance.