Predicting Performance Measurements

One of your refrains in the book is asking people to predict what they expect to find before measuring it. When did you start doing this yourself, or fall into the habit of trying to make predictions about performance measurements? I started making predictions when i took don canus fundamental algorithms glass and we counted cycles in the fake mix processor. So the point is to first raise a reader's awareness that you can actually estimate within factor ten how long things should take for almost anything.