2.2 KiB
id, title, tags, date-created, daily
| id | title | tags | date-created | daily | |
|---|---|---|---|---|---|
| 2026-04-02T20:41:59-0400 | 2026-04-02 20:41:59 |
|
2026-04-02T20:41:59-04:00 | 2026-04-02 |
2026-04-02 20:41:59
Decomposition Fallacy
I have expressed in other notes1 my frustration with how estimate decomposition is most frequently explained: as an intuitive process which can not be performed incorrectly and by definition creates a more accurate estimate.
The argument that an estimate based on the aggregation of intermediate estimates is necessarily more reliable than one made in one go is intuitive, but a non sequitur.
Frequently, decomposition is explicitly unhelpful due to its neglect of the portfolio effect. If an estimator can give a high confidence estimate of total dollars spent on rental equipment over the life of a project, it does not follow that they could give one for dollars spent on scissor lifts alone with greater or even equal confidence.2
My criticism is not entirely fair, however. Fermi and MacGregor's estimation problems have correct answers and are thus a different class entirely from those I'm most familiar with. I choose to criticize decomposition generally despite this because the distinction is rarely acknowledged by its advocates so "decomposition = good" pervades the estimating zeitgeist completely, and I get baffled, horrified, or patronizing looks when I suggest that complete quantity takeoff may not be more accurate than a parametric budget that could be completed in a day.
- ↩︎
-
I chose this example specifically because it has been an issue at both EC's I've estimated for. We set a target based on some well founded model ($0.25 per direct labor hour or 1% of total sell price), then 3--4 decently paid estimators spend up to an hour discussing how to make that target from quantities and durations of individual pieces of equipment. Ultimately none are totally satisfied with it, nor should they be, since the actual quantities, durations, and equipment types will be quite different, and the entire exercise was pointless. ↩︎