Measuring Precisely

Sep 26th, 2002 by Tony in QSM2

Since Capers Jones got us started, we in the software industry have devoted significant effort to measurements of software quality and productivity. Much of this effort has been misspent in a search for illusory precision. It does no good to measure precisely if you don’t have a personal observation model that matches that level of precision.

Do you care if there are 3.7296 or 2.7297 rat hairs per sausage? Will it make any difference in your or my actions if I can measure the humor quotient, or even lines of code per programmer day, to four significant digits?

Measurement without models only allows extrapolation. If the previous five projects have produced 25, 26, 27, 28, and 29 function points per month of labor, then I can extrapolate that the next will produce 30. Such extrapolation is actually based on a model too – a model that says progress is linear. But this model is too crude and simplistic to reflect real software quality dynamics. Without a model, measuring with greater precision may enable us to make more precise extrapolations (more significant digits), but it doesn’t allow us to make more accurate ones (better correspondence with reality).

— Jerry Weinberg, Quality Software Management Vol 2, Chapter 2

No Comments