Author: Matt Tuttle
Since the last posting, we now know that an Order of Magnitude estimate can be derived by using this basic formula:
Estimate = (a + 4m+ b) / 6
a= most optimistic estimate
b= most pessimistic estimate
m= most likely estimate
We also know that in order to generate such an estimate we need some general knowledge on the subject because there is such a heavy emphasis on Value M in the formula. Without any knowledge on the subject at hand, providing a most likely estimate would be like coming up with a Silly Wild-Assed Guess—or a SWAG. However, even with such general knowledge, our Order of Magnitude estimate will still yield us a 35% accuracy level from the onset.
Not good enough? Next we introduced the concept of standard deviation, a measure of variability or dispersion of data points in comparison to the mean.
Standard Deviation = (b – a) / 6
We said that by calculating one standard deviation of the Order of Magnitude estimate, we could get the estimate from 35% accuracy to 68.7% accuracy. Why is that? Let’s look at a normal bell curve.
The red area represents one standard deviation (as know as one sigma) and how much of a normal curve it covers. If the exact average of a normal curve is right on the 10-mark, we see that one standard deviation ranges from the 9-mark to the 11-mark. If this were a piece of pie, you’d cover basically 69% of the pie leaving 31% not covered. For most project managers, that leaves too much uncertainty and error; they like answers that cover most of the curve.
We see in the green area that if we want to cover 95.4% of a normal curve then we need to provide for two standard deviations, between the 8-mark and 12-mark. This will only leave 5% uncovered and is a percentage that most business managers are ultimately comfortable in working in. 95% is pretty good bet and provides managers with statistical comfort; getting to 99.7%, or three standard deviations, between the 7-mark and 13-mark, certainly only leaves 0.3% of the curve uncovered, but estimating purposes, such coverage is not needed as 95% gets us where we want to be without much error left to factor in.
In Part III, we derived the following estimate and standard deviation for the length of the Golden Gate Bridge (in kilometers):
Estimate = 2.42 kilometers; standard deviation of 0.583 kilometers. In short, we could say this:
“I am now 68.7% certain that the Golden Gate Bridge is 2.42 kilometers in length plus or minus 0.583 kilometers—or the range of the distance of the bridge is between 1.837 kilometers and 3.003 kilometers.” With one standard deviation applied to our estimate, we are basically covering 68.7% of a normal curve—but we are still leaving out 31.3%. That’s a lot of room for error.
If we want to cover the 95.4% range and reduce our chance of error to only 5%, all we simply need to do is add more one more standard deviation.
One Standard Deviation = (b –a) / 6
Two Standard Deviations = 2 [ (b-a)] /6
Three Standard Deviations = 3 [ (b-a)] /6
So, from our previous example, to provide a 95.4% accurate, two standard deviation estimate, we take our standard deviation and simply multiply by two. Estimate now equals 2.42 k +/- 1.168k or “I am now 95.4% certain that the Golden Gate Bridge is 2.42 kilometers; that the range of the distance of the bridge is between 1.252 kilometers and 3.588 kilometers.”
Turns out the Golden Gate Bridge is exactly 1.9 kilometers in length and that our one standard deviation estimate actually covered that, but without looking it up or knowing for sure, we could only be 68.7% statistically sure of that fact. By applying one more standard deviation, by going to two sigma, we moved up to 95.4% sure and that is a much more comfortable area to be in when making important decisions. Yes, the range of the estimate increased dramatically, but in return, your chance for error reduced dramatically as well.
In Part V of the High-Level Estimating series, we are going take our knowledge of deriving an estimate and applying standard deviations and we are going to calculate a 95.4% accurate estimate for the length of a 4-task project schedule.











Author: Kristen Backstrom