Developed by Thomas L. Saaty through the 1970s (The Analytic Hierarchy Process, 1980).
Turns pairwise comparisons on a one to nine scale into weights for criteria and scores for options, and flags contradictory judgments with a consistency ratio.
Three pairwise judgments become weights, and the method checks whether they agree
Judge each pair of criteria on the one to nine scale, toward whichever side matters more (1 equal, 3 moderately more important, 5 strongly, 7 very strongly, 9 extremely). The weights follow from the three judgments together. The consistency ratio measures how far the judgments contradict one another: quality over speed 3 to 1 and speed over cost 2 to 1 imply quality over cost near 6 to 1. Move quality against cost to Equal and the point crosses the 0.10 line. Illustrative.
VerdictRevisit the judgments: they contradict each other.Consistent enough to use the weights.
An example
A maker of handheld diagnostic devices had to pick one of three contract molders for its plastic housings, and the engineering, supply, and finance leads each favored a different one. The program manager ran the analytic hierarchy process in a two-hour session: the goal at the top, three criteria under it (quality, delivery speed, and cost), and the three molders at the bottom.
The group compared the criteria in pairs. Quality was moderately more important than speed, a 3; speed was a little more important than cost, a 2. Then finance argued that cost should rank equal with quality, a 1, and the consistency ratio jumped to 0.32, far past the 0.10 limit. The group had said that quality beat speed, speed beat cost, and cost tied quality, all at once. They talked it through and settled on 5. The weights came out at 65 percent quality, 23 percent speed, and 12 percent cost, and under them the second molder led clearly. The record showed exactly where finance's view had landed.
Why it works
People are poor at weighing five criteria at once and good at judging one pair at a time. The method asks only the easy question, which of these two matters more and by how much, and asks it for every pair. That is more judgments than the weights strictly need, and the extra ones are what let the method check itself.
If quality beats speed by 3 and speed beats cost by 2, quality should beat cost by about 6. When the group says 1, the judgments cannot all be true, and the consistency ratio measures by how much. A contradiction that would have hidden inside a weight is surfaced as a number, and the discussion goes to the pair that caused it. The judgments are also written down, so a later reviewer can see who held which view and how the weights followed.
The thinking behind it
Thomas L. Saaty Thomas L. Saaty, a mathematician and operations researcher, developed the analytic hierarchy process through the 1970s and set it out in The Analytic Hierarchy Process: Planning, Priority Setting, Resource Allocation (1980). A decision is broken into a hierarchy of goal, criteria, and alternatives. The people deciding compare the elements two at a time on a nine-point scale, and the priority weights come from the principal eigenvector of the matrix of comparisons. His consistency ratio measures how far the judgments contradict one another, so a group can see when its own answers do not hold together.
How to run it
- Build the hierarchy. Write the goal at the top, three to seven criteria under it, and the options at the bottom; more criteria than that call for subcriteria.
- Compare the criteria in pairs. For each pair, ask which matters more and by how much: 1 equal, 3 moderately, 5 strongly, 7 very strongly, 9 extremely.
- Derive the weights. Let a spreadsheet or the software compute them; with three criteria the normalized geometric mean of each row gives the same answer, and with more it comes close.
- Check the consistency ratio. Above 0.10, find the pair that strays most from what the others imply and discuss it before changing any number.
- Compare the options under each criterion. Repeat the pairwise comparisons for the options one criterion at a time, or score them on a common scale when there are many.
- Combine and test the ranking. Multiply the option scores through the weights, then move the largest weights to see whether the leader changes.
Where it goes wrong
Inconsistency gets fixed by fiddling. A group that nudges numbers until the ratio falls below 0.10 has hidden the disagreement the ratio found. Go back to the pair that strays most and ask the people who judged it what they meant.
The second is averaging people who disagree. Folding a finance lead's 1 and an engineer's 9 into a 3 produces a weight nobody holds. Take each stakeholder's judgments separately first, and combine them, with a geometric mean, only after the gaps have been discussed.
The third is a hierarchy grown too large. Ten criteria mean forty-five comparisons, fatigue sets in, and the later judgments turn careless. Group the criteria under a few headings and compare within each group.
When to use something else
- Weighted Decision Matrix. When the group can agree on the weights in one conversation and needs a ranking within the hour.
- Kepner-Tregoe Method. When some requirements are absolute, and an option that misses one should be out before any scoring starts.
- Binary Criteria Screen. When the options only have to clear a set of yes-or-no tests and nothing needs weighing.
Where it sits in SOLVE eX
Go to the source
- The Analytic Hierarchy Process: Planning, Priority Setting, Resource Allocation, McGraw-Hill, 1980