Monthly Archives: January 2010

Better Estimating is the Solution to Poor Estimating

<div class="posterous_bookmarklet_entry">
Glenn Alleman has a personal blog called Herding Cats on project management, where he posts prolifically. While he has a particular hot button on what is termed “Project Management 2.0”, and quite rightly, he is also respectful of agile methods and approaches, while balancing them out with his “big project” experience of US DoD contracts. He is eloquent, if not slightly repetitive, on his observations and recommendations on managing the delivery of non-trivial, real-world outcomes, and brings the experienced practioner’s pragmatism to the debate.
One of my own buttons is also his: estimation. Below is a recent post, included in its entirety, where he discusses it, in reference to an article by another practitioner. At its simplest is an axiom of physics or engineering: a measurement or an estimate is meaningless without a description of the uncertainty.
« Project Performance Measurment | Main | Definition of the Day »

January 19, 2010

Better Estimating is the Solution to Poor Estimating

Probability and Statistics
Johanna Rothman has got me thinking about why management and project managers want serial development and have such difficulty with estimating.
It starts with the missing understanding that all point estimates are in fact Wrong. This starts with the misunderstanding between statistics and probability.
Cost and schedule estimates are a forecast about a future event, the final project cost or schedule duration. Because future events are uncertain, they must be described probabilistically. But to do this, we need knowledge of the underlying statistical nature of the processes that drive the cost and the schedule. Without this statistical knowledge, we can not make credible estimates of the probability of the outcome.
First All Point Estimates are Wrong
But these types of estimates are desired by management. Wrongly desired of course, but still desired. So how can we “push back” on this request?
The first step is to define the range of values for any point estimate and the confidence that the point estimate falls within that range. Without this understanding all the participants in the project will be disappointed.
The first step is to define what the point estimate means. Is it the:

  • Most likely – the mode of all the possible costs and schedule durations?
  • The 50th percentile – the median of all possible cost and schedule durations?
  • The expected value – the mean of all possible cost and schedule durations?
  • The 4th percentile – the highest possible cost and schedule duration?

While these variables are being estimated, the “actual” project cost and schedule is also an uncertainty:

  • The “point” or “best” estimate is not the only possible estimate – this means other estimates are possible.
  • Use of the phrase “most likely” implicitly assumes that other value are “less likely.”
  • “Most Likely” (mode), 50th percentile (median), and “Expected Value” (mean) are statistical terms characteristic of probability distributions.

This terminology implies that cost and schedule are statistical in nature and are defined by their probability distributions.

The use of statistical cost and schedule estimating provides insight into the variance of these estimates in way naive “roll up” estimates cannot. The use of simple (naive) roll up estimates usually leads to disappointment in the end, because they are statistically unsound from the beginning and get worse as time goes on.
“Earned Value for Cost Estimators,” John Driessnack, Washington Area Chapter, SCEA, April 24, 2006.

There are Four Types of Uncertainty on Projects and the Behavior in Their Presence

  1. Normal Variation – that occurs in the completion of tasks arising from normal work processes. Deming has shown that these uncertainties are just part of the process. Attempts to control them, plan around them, or otherwise remove them is a waste of time.
  2. Foreseen Uncertainties – are identifies but have uncertain influences.
  3. Unforeseen Uncertainties – are events that can’t be identified in the planning process.
  4. Chaos – appears when the basic structure of the project becomes unstable, with no ability to forecast the outcome of cost, schedule, or technical performance.

See – “Uncertainty and Project Management: Beyond the Critical Path Method,” A. De Meyer, C. Loch, and M. Pich, INSEAD Work Paper 2001/04/TM and “Managing Project Uncertainty: From Variation to Chaos,” A. De Meyer, C. Loch, and M. Pich, MIT Sloan Management Review, Winter 2002.
With this guidance, you will come to see what Johanna is suggesting does not have to be, should not have to be, and certainly is not the approach by those “managing in the presence of uncertainty.”
So when someone (in this case Johanna) says:

… Of course, a serial life cycle provides a prediction that’s almost guaranteed to be wrong …

the simple solution is to stop this type of thinking and start thinking in terms of probability and statistics and how the single point estimates are both wrong and can be improved by thinking in statistical terms.
So Here’s Some Fundamental Principles of Cost and Schedule Estimating

  • Naturally occurring variance is part of the underlying statistical behaviour of any netwrok iof activities.
  • Attempting to control or Over Control these natural variances is a waste of time – this is the purpose of cost and schedule margin.
  • Mitigate unforeseen uncertainties with risk buy down activities – have a Plan B for everything that needs cost and schedule protection.
  • Have an alternative plan for Unforeseen Uncertainties.
  • When Chaos emerges, replan the project