Stepping Back for a Broader View
We’ve looked at three basic approaches to modeling the time it will take to accomplish some planned work, as well as a hybrid approach that mixes elements from these three. All conform to some approximation of the amount of work and the rate of progress.
-
Linear:
-
amount of expected work
-
historical rate of accomplishing work
-
-
Parametric:
-
various parameters expected to correlate with or affect the amount of expected work
-
various parameters expected to correlate with or affect the rate of accomplishing work
-
-
Stochastic:
-
nominal amount of expected work plus range of work growth
-
nominal rate of accomplishing work plus range of variation (including delays)
-
Models are a way of simplifying your view of the world around you. They give you a map for navigating that world. This lets you emphasize the points that are salient and ignore the noise. What is salient and what is noise, of course, is a judgment call on your part.
Using models to calculate estimates is a great time and cognitive energy-saver. They allow you to repeatedly apply the same criteria to different situations. This is both a blessing and a curse. You need to keep an eye out for situations where your usual criteria doesn’t apply. If you question your model frequently, you should be able to stay out of trouble. If you double-check periodically with a different model, that’s even better. And, as always, when the map and the territory don’t agree, trust the territory. This tells you that there’s something your map doesn’t know.
As we’ve seen, no matter how sophisticated the model you’re using, it needs to be calibrated to your context in order to give answers appropriate to that context. The advice to have lots of accurate historical data is repeated in every tome about accurate software development estimation. The assumption is that things have remained static enough that this historical data will be representative of your current context and useful to calibrate your estimation process.