Linking Everything Together

Altogether, a lightweight measure is not impossible. It can even be easy to achieve.

The first step is to ask yourself what you want to get from a measurement. Do not consider anything that you do not really need.

Then, determine the data you need to accomplish your measurement goal. Often, enough data is already available for simple models, which are usually the most useful, robust, and likely to be current and not prone to error.

Third, look for such data. There are lost treasures in your company: accounting departments, defect databases, customer service centers, configuration management systems, and others.

Fourth, try to automate the data collection as much as possible. Human-collected data is error prone and unreliable because metrics data collection is not a directly productive activity.

Fifth, collect the artifact measures for your software data. There are a few metrics extraction tools on the market. Some development tools come with metrics capabilities.

Sixth, build the models. I recommend using simple statistics. You can build more advanced models on top of such simple models. However, simple models are the place to start.

Seventh, act on the measures you have collected. The models are there to be used, not to showcase how good you are. If you do not use them to improve, say, the prediction of the development speed of your use cases or the defect rate, you are wasting your time. Soon, your boss, your customer, or even your subordinates will find some more useful way for you to spend your time.

If I managed to reinforce your interest in metrics, I suggest that you read the mythical "Fenton and Pfleeger," a kind of cult book for software metricians [Fenton+1997]. It is a fairly short description of a formal and yet not too difficult process to collect and analyze software measures.