Estimates and Precision

The following example is very interesting because it show how overall precision
increases as activites are broken up in tasks.

A project is made up of 4 task. We estimate the time for each of these tasks to be 100 hours, 200 hours, 400 hours and 300 hours respectively. We are pretty sure (99% certain) that our estimates have a precision of plus or minus 30 pourcent.



Total Duration = ______________ ?
Range + ou - = _________________________?

1) What is the estimate for the activities? (X plus or minus Z)
2) What is the precision for the sum of the activities?
3) Note how precision increases with the number of tasks.

Statistical Solution




To cover most cases, we can assume 6 stand.dev = Max - Min.

A 100 hour task varying by 30% represents a minimum of 70 and a maximum of 130.
6 stand.dev = 130-70 = 60 --> stand.dev = 10




Variance = (stand.dev)x(stand.dev)
Variance of a sum is the sum of the variances = 3000





Total Stand.Dev = Square Root of 3000 = 54.7726

We are 99% certain that the total duration will be a value between 1000 - (3*stand.dev) and 1000+(3*stand.dev).

We are 99% certain that the total duration will be a value between 1000 - 164.32 et 1000 + 164.32.


To anwer the initial questions!

1) What is the estimate for the activities? (X plus or minus Z)

1000 plus ou moins 164.32

2) What is the precision for the sum of the activities?
The precision now is 16.4 pourcent.


3) Note how precision increases with the number of tasks.
Starting with a 30% precision on the individual activities we end up with a 16.4 aggregate value.


Although statistics and the real world don't always agree the above method does represent the best unbiased way of determining the increase in precision.

The morale of the story is don't spend too much $+time on individual tasks estimates. The total precision is much higher than the precision of the parts.

----------------------------------

Fully Qualifying estimates is very important. A fully qualified estimate is a central value, a precision and a level of certainty.

I weigh 147 pour + or minus 5 lbs. How sure am I of my estimate. Well, not too sure,
because I haven't been on a scale for 15 years.

John: How long will it take to setup those machines
Jim: 40 hours
John: Exactly 40 hours
Jim: Between 20 and 60
John: Are you sure.
Jim: Hell no! I am just trying to get rid of you. Paul might not even be here tomorrow. So I guess it could take 80 hours or more.

John knows how to qualify estimates. He comes out of the discussion with more info than he would had he taken the 40 hour estimate as granted.


____________________________________

Estimates (basic principles)
 If we want to meet requirements, deliver on time within budgets, we need realistic estimates.
 GIGO (Garbage In Garbage Out)
 A estimate is a prediction that has as many chances of being to high as of being to low. It is a centered value.
 Each phase requires an estimate for the next phase and one for the total project.
 The people doing the work are not the best to consult for an estimate
 use a range rather than a fixed number
 Use the technique appropriate to the phase
 Use several techniques.
 Revise the estimates as the project unfolds
 Gather and analyse data
 Use common sense
 Add amounts for contingencies (Other costs > 20% is reasonable)
 Qualify your estimates fully by specifying the central value, the range of error and the level of certitude.
Items below Not required for Exam....
 You will often see formulas such as below:
New estimate = (Min + 4*PlusProbable + Max)/6
 Range = (Max-Min)
 3 standard deviations = (Max - Min)/2 certain at 99%


Back to work...
Gilbert