Following on from last week’s post by James, I thought I would continue the estimation discussion, but from a different angle.
Douglas Hubbard (author of How to Measure Anything: Finding the Value of Intangibles in Business) believes that we can calibrate our estimation skills so that we are less likely to under- or over-estimate. Although he warns against giving estimates (as most are provided for the wrong reason and are measuring the wrong things), if you have to give an estimate then Hubbard believes your are better off giving one after a bit of calibration: by honing your skills on unrelated topics you will be better placed to offer estimates in your field of work. For example, by learning how to provide valuable estimates on questions such as ‘What is the surface temperature of the sun”, you will be better at estimating how long a software project might take.
These tests are about facts (Alexander the Great’s birth year does not change), so they require participants to answer the questions without conducting research. This often receives criticism because “life isn’t like that; we can run spikes if we have no idea, or have at least done something similar”. Hubbard and McConnell say that doesn’t matter, but I wanted to know how people would fare if they were allowed to do research. That meant asking questions which were yet to be answered, but which people could still estimate.
So, I have created a couple of tests (one 90% Confidence Interval and one Binary) to see. Please complete the following tests, telling me how much research you did. Ideally, you will complete both without doing any research, then complete both again having done some research. We will then be able to calculate the effects that a little bit of knowledge has. As you will see, it is impossible to ‘know’ the answers to these questions as of today (although the further into June we go, the better your estimates might be – ref ‘Cone of Uncertainty).
Don’t forget to tell me how much research you did at the end otherwise your data has much less value.