User:Louis F. Sander/Sandbox: Difference between revisions

Revision as of 14:39, 22 August 2007

Click this link to see the Main Article
Click this one to see the Wikipedia Article

Thomas Saaty Citizendium
Thomas Saaty Wikipedia

Sections

Lead/Introduction
Where it is used
How it works
Teaching it
Criticisms (keep it short and objective)

Notes

Aspects of AHP

Math
Psychological
Computerized now, with gadgets
Disciplined look at a decision (focus on objectives, alternatives, more than voting)
Handles rational, intuitive, irrational, all at the same time (RS: maybe there is a better word than disciplined to capture this idea – though I can’t think what it might be right now. Somehow disciplined sounds too straightlaced, too mechanistic.)
Permits explicit presentation of evaluation criteria (de Stiguer)
Lets you compare the alternatives and fiddle with them, vs. just giving the best one
A lot of people use it. Way more of them than you think.

Miscellaneous

Selecting a spouse is important and has long-term consequences.

This is like logarithms, the LaPlace transform, etc., in that it replaces something hard to deal with with something much easier to deal with.

Deleted Material for Reinsertion

and has been successfully applied to many complex planning, resource allocation, and priority setting problems

Aspects of AHP are sometimes used in designing highly specific procedures for particular situations, such as the rating of buildings by historic significance.^[1]

Good Stuff

Problems with high stakes, involving human perceptions and judgements, and whose resolutions have long-term repercussions, call for a rational approach to their solution. (Bhushan promo)

You might also want to read the paper "The Analytic Hierarchy Process - An Exposition," E.J. Forman and S. I. Gass, Operations Research, 49, 4, July-August, 2001, pp. 469-486. (Saul Gass, U of MD) (RS: Forman and Tom originally started the Expert Choice company in 1983 as partners, and Forman still runs Expert Choice. Would using this article be in conflict with the “no vested interest” Wikipedia requirement?)

Potential References

Older and/or less useful

Drafts of possible sections

Pairwise Comparisons

In AHP, we rank a group of items (RS: on some property they have -- the overall ranking is obtained after synthesizing the priority vector results for all the properties) by comparing them to each other in pairs. If the group has more than a very few items, this can be very much easier than trying to rank all the items at once.

As we consider the pairs of items, we express the importance of one vs. the other by assigning it a weight from -9 to +9. (RS: no, an important distinction between AHP and other processes that directly assign numbers from 1 to 5 to items on some property is that the assigned number or judgment is always about the relation between a pair of items being compared and expresses how many times more the larger one is than the smaller one . We use numbers from 1 to 9. The negative numbers of the questionnaire are a convenience for soliciting judgments – they are actually inverses: 1/9, 1/8 and so on. It is easiest to see this if you form the matrix of judgments that results from the complete set of judgments of n items on a given property. Also RS: Lou – people use minus numbers instead of fractions because it is easier to present questionnaires that way and it is even easier to fill in matrices using a minus sign to indicate when the number is actually a reciprocal, instead of having to enter something like 1/3, but we really need to make it clear hat the actual math of the AHP does not use negative numbers.) To facilitate the assignment of weights, we can use a verbal scale ranging from "Much Less Important" to "Equally Important," to "Much More Important."

When we have finished all the comparisons in a group, AHP's mathematical algorithm evaluates our work and derives the appropriate overall weight for each of the items in the group. The greater the weight, the more important to us is the item to which it is assigned.

The algorithm also gives a numeric indicator of the consistency of our rankings. (If we greatly prefer apples to oranges, and greatly prefer oranges to turnips, it would be inconsistent to say that we also slightly prefer turnips to apples. When we say such a thing, AHP notices it and prompts us to reconsider. We may react to the prompting or not; after all we, not AHP, are making the judgments here.)

(RS: And though we can detect and note inconsistency, the AHP must allow inconsistencies because the real world is sometimes inconsistent, the Steelers beat the Giants, the Giants beat the Redskins and the Redskins beat the Steelers. Many decision theories prevent modeling such actual inconsistent events. In general it is better to be fairly consistent and that is why it is desirable to be able to point out to the judge where his inconsistency is – maybe he should at least think it over and see if he means that or not.)

Example

To illustrate the AHP technique of pairwise comparisons, let's use it on something that AHP is very well suited for: a problem with high stakes, involving human perceptions and judgments, whose resolution has long-term effects. To keep things simple, we'll choose a problem that is universally familiar, and we'll limit our comparisons to only three criteria.

Imagine that you want to evaluate, on a rational, disciplined basis, how important each of three criteria is to you in selecting a prospective spouse: Brains, Looks, and Personality. If we compare the criteria two by two, AHP can use our comparisons to assign a numerical weight to each of the three.

You can make such an evaluation on THIS WEB SITE from the Canadian Conservation Institute. Here's what to do after opening the site:

On the first screen, enter the number 3, to specify how many criteria you will be comparing.
Click Continue, then enter the names of the criteria: Brains, Looks, and Personality. (To facilitate our discussion, please enter them in that order; normally, the order doesn't matter.)
Scroll down and select the Line-by-Line method to facilitate entering the data.
Click Continue and begin your pairwise comparisons. (Look at the bar chart and note that before you make any comparisons, all three criteria have equal importance.)
For each of the three pairs of items, you will compare the first to the second by entering a number between -9 and 9. You can use the verbal scale to help you choose each number. (Example: If Brains were a little less important to you than Looks, you might enter -3 for this pair. If Brains were very much more important, you might enter a 9.) To facilitate our discussion, imagine that Brains are moderately more important to you than either Looks or Personality, and that you have no preference for Looks vs. Personality. Indicate these preferences by entering 5, 5, and 1 for the pairwise comparisons.
Click Calculate to see the results of your work. The higher the number assigned to each criterion, the more important it is to you, based on your own pairwise comparisons. In our example, Brains accounts for 71.43% of your preference, and each of the others for 14.29%. (These numbers are a bit too precise to apply to the human situation involved, so we might just rely on the bar chart to compare them.)
Look at the Consistency Ratio below the bar chart. If you've entered the numbers we asked you to, your consistency ratio will be zero, indicating that your comparisons were perfectly consistent. The Consistency Ratio is a measurement of the consistency of the data you entered (remember the apples, oranges, and turnips above). The higher the ratio, the less consistent were your entries. Whenever it is greater than 0.1 or so, you may want to review your entries to see if they really reflect your thinking.

The web site allows for easy modification of your judgments, and it can be instructive to change them and see how that affects the outcome. (The modification feature "times out" after a few minutes, after which you must start your evaluation from scratch.)

By experimenting with this small number of criteria, you can develop a feel for what AHP is doing. Keeping your first two comparisons the same, express a slight preference for Looks or Personality, and notice what happens to the results. Change your preferences in other ways, and see if you can follow the changes in results. (Don't worry too much about inconsistencies unless the ratio is larger than 0.1 or so -- inconsistency is a part of human nature, and we are dealing fairly closely with human nature here.)

To experiment at a more challenging level, enter a new problem with four criteria: Brains, Looks, Personality, and Wealth. With the larger number of criteria, it is harder to "see" how they interact, but the technique is every bit as valid. Imagine how difficult it would be for you, without the aid of AHP, to apply systematically-derived numerical weights to the six criteria of Brains, Looks, Personality, Wealth, Age, and Religion. Try it with AHP and see how simple such a thing can be. Once you've finished, see if you're pleased with the weights, and if you are comfortable about how they were derived.

Diverse perceptions and judgments

VERY ROUGH DRAFT OF THIS PART...

AHP is a method of measuring perceptions and judgments in relative terms.
Its inputs depend on the judgments of the participants/decision makers.
It allows the consideration and combination of many kinds of judgments about the matter at hand, e.g., rational, intuitive, irrational, other(?).
If the judgments are irrational, won't we likely get a clue to that from the Consistency Ratio?

Example (much smoother than the stuff above)

Most decisions require the decision maker(s) to consider price or cost. Compared to many other considerations in decision making, this one is simple, straightforward, and relatively free from emotional or subjective complexity.

When considering a pair of prices, it is obvious which is the higher and which is the lower; a calculator makes it easy to analyze the absolute and relative differences between the two. Given a longer list of prices, anyone can identify the lowest price and the highest one, and with the help of a spreadsheet program, can compute the exact numerical relationships between all the prices on the list.

But how can this information be applied to making a specific decision? Different decision makers can have very different ways of dealing with it; AHP can accommodate them all.

Take the familiar case of buying an automobile. The prices of the alternatives can be obtained from formal price quotations, and can be known exactly to the penny. After considering multiple bids on each of several autos, we might get a list like this:

$18,285.00 - Chevrolet Cobalt
$28,240.72 - Audi A4
$34,885.50 - Chrysler 300
$35,001.67 - Saab 9-5
$43,548.31 - BMW 5 Series

As each decision maker makes pairwise comparisons of these numbers, he or she will apply their personal judgment to the task. Those judgments can be of various kinds, and whatever they are, AHP will capture them. Imagine different decision makers having these different thoughts as they consider various pairs of prices:

"We're comparing many other factors about these automobiles, but now that we're focusing on price,...

... prices are numbers, and numbers never lie. Two items should be preferred in exact proportion to their price, with the lower price being better. A $20,000 price is exactly twice as good as a $40,000 price, exactly 10% less good than a $22,000 price, and so forth. I will make my comparisons according to the irrefutable perfection of numbers."
... you should always go with the lowest responsible bidder. I'll always prefer the lower price as strongly as I can over the higher one."
... you get what you pay for. I'll always prefer the higher price as strongly as I can over the lower one."
... price is no object in this case. Since we have enough money to buy whatever car we want, it's the other considerations that should make the difference. I'll rank all prices as equally preferable."
... no automobile is worth $35,000. I'll do my best in comparing the less expensive ones, but I'll give anything over $35K as little weight as possible."
... I know that U.S. car companies are always flexible on price. They'll always come down when you press them, so I'll give extra preference to the Chevrolet and Chrysler whenever they're part of a comparison."
... I may not know cars, but I know my numerology. I'll add up the digits in each price, then keep adding the digits in the results until I'm down to a single digit. The closer it is to a seven, the more will I prefer it."
... I'll do the best I can at comparing the prices, but any car under $25K is a piece of junk, and I'll give it as little weight as possible."
... I've never been comfortable with numbers. I'll just guess."
... comparing these prices isn't as easy as it looks. But as an experienced purchasing manager who buys 500 autos a year, I'll apply my knowledge and judgment, and I'm sure it will work out well."

NOTE TO LOU: Do something like the above with the wealth aspect of choosing a spouse. Maybe consider current salary, personal net wealth, his/her (estimated?) share of family net wealth, etc.

NEW SUBJECT

EXERCISE FOR MBA CANDIDATES:

Given the above list of prices, use Excel to calculate and display all the relationships between them.
Now that you know all that stuff, how do you use it in deciding which car you want to buy?

↑ Lippert, Barbara C.; Stephen F. Weber (October, 1995). "HIST 1.0; Decision Support Software for Rating Buildings by Historic Significance". National Institute of Standards and Technology, NISTIR 5683. Retrieved on 2007-08-20.

[1] Lippert, Barbara C.; Stephen F. Weber (October, 1995). "HIST 1.0; Decision Support Software for Rating Buildings by Historic Significance". National Institute of Standards and Technology, NISTIR 5683. Retrieved on 2007-08-20.

[1]

@@ Line 21: / Line 21: @@
 *Permits explicit presentation of evaluation criteria (de Stiguer)
 *Lets you compare the alternatives and fiddle with them, vs. just giving the best one
+*A lot of people use it. Way more of them than you think.
 ===Miscellaneous===