The presentation "Quantifying the Impact of Agile Practices," Larry MacCherone at the RallyOn 2013 Conference, presents some results on estimating impacts. The chart below shows 4 estimating types, including No Estimates, the sample sizes for each type and the components that make up the estimating types.
The Software Development Performance Index (SDPI) scale on the left ranges - by eyeball measurement - from 46 to 55.
The Higher the number the better the performance of the process. The presentation speaks to the components of the index further.
But first another piece of information ...
Teams doing Full Scrum have 250% better Quality than teams doing No Estimating
But are these differences meaningful statistically?
Let's start with several reading assignments, before answering
- How to Lie with Statistics, Darrell Huff - this is a must have book for anyone working in an environment where numbers are used to make decisions.
- Statistics: A Very Short Introduction, David J. Hand, Oxford University Press - this is a short summary of all the other books on statistical processes sitting on my office shelf.
- The Flaw of Averages: Why We Underestimate Risk in the Face of Uncertainty, Sam Savage - another must have book to learn that those tossing around numbers are likely unaware of the flaws in their logic.
Let's start with the numbers from the chart
Since the raw underlying data is not available, we can't do any p-Factor assessment from the population samples, but there is a simple question that can be asked.
Are there any statistical differences between the 4 SDPI's? If you look below at the quick and dirty assessment of the only data available, it looks like all 4 approaches are within a single digit variances of each other. Not that useful actually.
So the critical question still remains
How can you make a decision in the presence of uncertainty without estimating the impact of that decision?