11.

TUTORIAL FOR ANALYTIC HIERARCHY PROCESS

CASE: JENNYS GELATO, P. 226
The Analytic Hierarchy Process (AHP) is a general problem-solving method that is useful in making complex decisions (e.g., multi-criteria decisions) based on variables that do not have exact numerical consequences. We will explain the basics of creating and using AHP models. The software is supplied by Expert Choice, Inc., and this tutorial is based on Expert Choices online tutorial. We encourage you to work through the online tutorial to get a fuller understanding of all the features of the AHP model. Decision modeling using Expert Choice (in the Evaluation and Choice mode) typically consists of ve steps: Structuring the decision model. Entering alternatives. Establishing priorities among elements of the hierarchy. Synthesizing. Conducting sensitivity analysis.

STRUCTURING THE DECISION MODEL

You start by breaking down a complex decision problem into a hierarchical structure. You can use the following elements: Overall goal (subgoals) to be attained. Criteria and subcriteria. Scenarios. Alternatives. As an example, we will look at how a manager of an ice cream store chain decides where to open a new outlet. The goal is to select the better of two locations. The main criteria in this decision are rental space cost, trafc of potential customers, and the number of competitors. The store locations they are considering are a suburban mall (moderate rent, high population of teenagers and retirees who are known to be purchasers of ice cream) and Main Street downtown (high rent, main trafc is ofce workers who are not around in the evenings and weekends). An Expert Choice model consists of a minimum of three levels: the overall goal of the decision at the top level, subcriteria at the second level, and alternatives (the two locations for the store) at the third level. In models with many levels, place the more general factors of the decision in the upper levels of the hierarchy and the more specic criteria in the lower levels. To start a new model, go to the Expert Choice File menu, and choose New. Enter a name for the le in which to save the model you will build (e.g., location.ec1) and click OK.

* Tutorial 11, June 2005

11. Tutorial

To set up the model node by node, click Direct.

Type in a description of the goal of the decision (e.g., Select Site Location for new Ice Cream Outlet). Click OK. For this example, the decision criteria are extent of competition (COMPETN), customer t (CUST FIT), and cost (COST). To enter the criteria, set the cursor on the GOAL, go to the Edit menu and choose Insert. This adds a level to the hierarchy directly below the active GOAL node. When you enter each criterion (e.g., COMPETN) press Enter and the program will allow you to give a description of the criterion (e.g., the number of competitors in the area), then press OK. It will take you to a new criterion, after you have entered information about all the nodes at this level of the hierarchy, press <Esc>.

Analytic Hierarchy Process

ENTERING ALTERNATIVES
To enter the alternatives, rst select the COMPETN node. Next, on the Edit menu, choose Insert. Type an abbreviation (eight characters or less) for the rst alternative (e.g., Mall). Again you will be asked to enter a more detailed description for the alternative (e.g., Suburban mall location). Click OK to enter the next alternative. When you have entered all alternatives, press <Esc>.

11. Tutorial

Activate the subcriterion node under which you entered the alternatives, go to the Edit menu and choose Replicate children of current node.

Click to peers or to all Leaves to replicate the alternatives to all of the other nodes at the bottom of the model. Once the alternatives have been associated with all subnodes, the alternatives will be displayed at the bottom of the screen.

Analytic Hierarchy Process

To save a model, go to the File menu and choose Save.

ESTABLISHING PRIORITIES AMONG ELEMENTS OF THE HIERARCHY

When you have specied the decision model fully, you can evaluate the alternatives and criteria at each level to derive local priorities (weights) by making pairwise comparisons among elements at the same level in the hierarchy. First select the node for which you want to establish priorities. Go to the Assessment menu and choose the assessment method, for example, Pairwise or Data. Entering importance weights (Assessment, Data): You can enter absolute values for the weights of the nodes (instead of relative values such as those derived in the pairwise comparison mode), by clicking Assessment and then Data. For example, you can assign importance weights for the criteria competition, customer t, and cost, as shown below:

11. Tutorial

After you enter the importance weights, click Calculate. A new dialog box will appear with the graphical representation of the importance weights.

If you want want to keep the changes, click Record. The new values will be reected in the decision model.

Analytic Hierarchy Process

If the decision model does not show the weights, click the P/F icon at the top of the screen.

Entering pairwise comparisons (Assessment, Pairwise): You can make simple pairwise comparison judgments on each pair of items in the comparison set. To begin the pairwise assessment process, move to the rst criterion by double-clicking on its node (e.g., COMPTN). Select Assessment and then Pairwise (this is only available if there are nine alternatives or less) to compare the alternative sites on this criterion.

11. Tutorial

Choose the type of comparison (Importance, Preference, Likelihood) and the mode of comparison (Verbal, Graphical, Numerical). Choose a type and mode with which you feel comfortable. The type you select will not affect any calculations performed by Evaluation and Choice (the selected type will appear in the comparison statement). Generally, Importance is appropriate when comparing criteria, Preference is appropriate when comparing alternatives, and Likelihood is appropriate when comparing uncertain events.

Click OK. When only two nodes are to be compared, the Graphical comparison mode is the default.

Analytic Hierarchy Process

To indicate your preference in the graphical mode, click and drag the upper or lower horizontal judgment bars. Alternatively, you can enter your assessment in the Verbal and Numerical mode, in the Matrix mode, or in the Questionnaire mode. Choose the appropriate tab. When you have completed all of the comparisons, the program will automatically calculate and graph the priorities and give you a measure of the inconsistency of your judgments (Inconsistency Ratio). (In our example, there is only one pairwise comparison and therefore, no redundant information: the inconsistency measure is meaningless.)

Click Record to return to the main screen. Treat the other criteria in a similar fashion (e.g., double-click on the COST node) to continue the assessment process. After you nish making the pairwise comparisons and evaluating the alternatives (at the lowest level of the hierarchy), compare the criteria pairwise with respect to the goal. The program uses this information to derive the relative importance of the criteria (cost, customer t, extent of competition). To make pairwise comparisons between criteria (instead of directly assigning importance weights earlier in Entering Importance Weights), double-click the GOAL node. Select Assessment and then Pairwise to compare the criteria. Select Importance for the Type and Verbal for the comparison mode. Clicking Skip Preliminary Questions will bring you to the following screen:

10

11. Tutorial

Judge the relative importance of the two criteria and click Enter to move to the next comparison. Click Calculate to get a measure of the inconsistency of your judgments.

The Inconsistency Ratio is purely informational. The program does not enforce consistency. It is unlikely that you will be perfectly consistent in making comparative judgments, particularly when dealing with intangibles. As a rule of thumb, make the inconsistency ratio 0.10 or less. If you wish to improve the inconsistency ratio, go to the Assessment menu and choose the Matrix mode. Pull down the Inconsistency menu and select 1 most to see which is the most inconsistent judgment (highlighted cell). You can enter a new judgment value perhaps after looking at the Best Fit value that the system computes for reference purposes. You can see this value by clicking the words Best Fit in the upper left of the matrix or by going to the Inconsistency menu and selecting Best Fit.

Analytic Hierarchy Process

11

When you have nished entering your judgments, click Record to save the values and to return to the main screen.

SYNTHESIZING
After you record your preferences, the program synthesizes the priorities across the hierarchy to calculate the nal priorities of the alternatives. You can choose between two numerical methods of synthesizing the data: the Distributive and Ideal (default) Synthesis Modes. To synthesize, go to the Synthesis menu and select From Goal.

12

11. Tutorial

To see the list of the global priorities for all the hierarchys nodes, click the Details button. You can select the Distributive synthesis button, the Ideal synthesis button, or the Set button, which will lead you through a series of questions to determine the proper mode. The choice of the appropriate synthesis mode for an application depends on whether you view the decision situation as prioritizing among all the alternatives based on their relative worth (Distributive) or as picking a single best alternative (Ideal). The Ideal mode preserves the original ranks as alternatives are added. The distributive mode allows ranks to change. Allowing changes in rankings could mean, for example, that alternative A is preferred to alternative B before alternative C is introduced into the mix of alternatives being evaluated, but with the introduction of C, B becomes preferred to A. In the Distributive mode, criteria weights depend on the degree to which each criterion differentiates between the alternatives being evaluated. This favors (i.e., assigns higher weights) to alternatives that are both better than other alternatives on important criteria and unusual among the alternatives. We describe these two modes for synthesizing priorities below. For further information and examples, refer to the Expert Choice online help or tutorial.

DISTRIBUTIVE SYNTHESIS MODE

In the Distributive synthesis mode, the program normalizes the weights of the alternatives under each criterion. That is, it distributes the global weight of a criterion among the alternatives, thereby dividing up the full criteria weight into proportions that correspond to the relative priorities of the alternatives. (The normalization is applied at all levels of the hierarchy up to the goal node.) You should choose the Distributive mode if you want your preferences for alternatives to depend on the number and type of other alternatives being evaluated. Use the Distributive mode when all the alternatives are relevant. For example, use it when the decision task calls for prioritizing alternatives, allocating resources, and planning when the ranking of the alternatives is affected by the other alternatives. In general, if the alternatives are distinct (i.e., not duplicates of each other) and are well separated in their characteristics, you should use the Distributive mode.

IDEAL SYNTHESIS MODE

In the Ideal synthesis mode, the priorities of the alternatives are divided by the largest value among them and then multiplied by the global weight of the given parent node. Consequently, the most preferred alternative in a group receives the entire group priority of the criterion immediately above it. The other alternatives receive a proportion of the global weight. (Note that if the same alternative is best for all the criteria, that alternative obtains an overall value of one, while the other alternatives obtain proportionately less. The sum for all the alternatives will be more than one.) Unlike the distributive mode, the Ideal mode maintains the rank of the best alternative. The Ideal mode is useful when the alternatives are not distinct and you do not want the mere presence of copies or near-copies of alternatives to affect the decision outcome. For example, when comparing three computers, two low-priced computers are very similar to one another and the high-priced computer is better under most other criteria. The two low priced computers are both better alternatives with respect to cost, but they would cut into each others weight if the weight of cost is distributed among the alternatives (as is done in the distributive mode). The Ideal mode would give the entire weight for cost to the cheapest computer, thereby making it a stronger competitor to the high-priced computer. The Ideal mode divides the numerical ranks of the alternatives for each criterion by the largest value among them, instead of normalizing the entire set. The most-preferred alternative receives the

Analytic Hierarchy Process

13

value of one. In this case, the ranks of the alternatives do not depend on each other. Each new alternative added is compared only with the highest ranked alternative for that criterion. In general, use the Ideal mode when your sole concern is for the highest ranked alternative, and the others do not matter, or when several alternatives have equal or very similar values along most criteria. To see distributive or summary weights for alternatives, Click the Distributive or the Ideal button and then click the Summary button. To see distributive or ideal weights for alternatives and details, click the Distributive or the Ideal button and then click the Details button.

SENSITIVITY ANALYSIS
Use sensitivity analysis to investigate how sensitive the rankings of the alternatives are to changes in the importance of the criteria. Expert Choice offers ve modes for graphical sensitivity analysis: Performance Dynamic Gradient Two-dimensional Difference Sensitivity analysis from the Goal node will show the sensitivity of the alternatives with respect to the criteria below the goal. You can also perform sensitivity analysis from nodes under the goalprovided the model has more than three levelsto show the sensitivity of the alternatives with respect to lower-level criteria. To initiate the performance sensitivity analysis, go to the Sensitivity-Graphs menu and choose Performance.

14

11. Tutorial

The criteria are represented by vertical bars, and the alternatives are displayed as horizontal line graphs. The intersection of the alternative line graphs with the vertical criterion lines shows the priority of the alternative for the given criterion, as read from the right axis labeled Alt%. The criterions priority is represented by the height of its bar as read from the left axis labeled Crit%. The overall priority of each alternative is represented on the OVERALL line, as read from the right axis. To perform what-if analysis, click a rectangular criterion box and drag it up or down to change the priority of that criterion. You can observe the changes this makes in the ranking of the alternatives on the right. In the above example, The Mall has higher overall priority (61 percent) even though it performs poorly on the Cost criterion. To restore the original priorities, press the <Home> key. Returning to the Main menu also erases what-if changes and restores the priorities as originally calculated. You can use the Window menu command to change to other sensitivity graphs. (What-if changes will be reected in the open windows only if you have not pressed the <Home> key.) There are other methods of investigating sensitivity using Expert Choice: Dynamic sensitivity: The dynamic sensitivity analysis is a horizontal bar graph that you can use to increase or decrease the priority of any criterion and see the change in the priorities of the alternatives. For instance, as you increase the CUST FIT criterion by dragging that bar to the right, the priorities of the remaining criteria decrease in proportion to their original priorities. The program recalculates the priorities of the alternatives based on their new relationship. Gradient sensitivity: The gradient sensitivity analysis assigns each criterion a separate gradient graph. The vertical line represents the current priority of the selected criterion. The slanted lines represent the alternatives. The current priority of an alternative is where the alternative line intersects the vertical criterion line. 2D Plot sensitivity: The two-dimensional plot sensitivity shows how well the alternatives perform with respect to any two criteria. Differences graph: The differences graph shows the differences between the priorities of the alternatives taken two at a time for all of the criteria. You can go to the Options menu and choose Weighted or Unweighted to show the differences in either manner. When unweighted, the criteria are treated as though they have equal priorities. When weighted, the criteria show both priorities and differences.

LIMITATIONS OF THE EDUCATIONAL VERSION OF THE SOFTWARE

Maximum number of levels (including the Goal level) 4