Membuat Model Kompleks ANP

• Complex models have three levels of networks:
  • The main or top-level network is where the BOCR nodes, the merits of the decision, reside. The strategic criteria also appear in the top-level model. Their purpose is to establish the priorities for the benefits, opportunities, costs and risks in this decision using ratings.
  • The control hierarchy networks. There are 4 such hierarchies, one for each of the BOCR nodes. Thus there are four separate networks that we refer to as control hierarchy networks. The structure in this network is usually a hierarchy with two clusters: a goal cluster containing the goal and a criterion cluster containing the criteria. They must be pairwise compared.
  • The bottom level decision subnets. There are at least four and often more – up to 8 or 12 different networks. The decision subnets must each contain a cluster of alternatives and the names of the alternatives must be exactly the same in each one. You must also create other clusters that include the factors you think apply and connect the nodes for which you want to make judgments. Do NOT connect nodes and leave them uncompared. The effect is to give them equal priorities which may not be what you intended at all and that would tend to dilute informed priorities you may have obtained through other pairwise comparisons.
• What are the different kinds of “criteria” we are dealing with? There are three levels of networks in a strategic model and “criteria” appear in each level.

• Strategic criteria appear in the top-level main model: These are the invariant objectives the decision-making entity always seeks to satisfy. For an individual it might be Family, Health, Wealth. Strategic criteria must be entered in the model after the wizard creates the basic template for the model. However, they are used at the end of the process in a ratings spreadsheet to evaluate the BOCR node priorities AFTER the subnets and their judgments are finished and synthesized for each of the BOCR.
• Control criteria appear in the second level. These are higher level, more abstract criteria in subnets attached to each of the BOCR nodes. They are generally evaluated in a hierarchy. They include, for example, Environmental, Economic, Social, Political, Technological. After pairwise comparing them to establish priorities, select the high priority control criteria to have decision subnets built for them. You may leave out low priority ones. You need to select enough control criteria to have their priorities sum to at least 70%.
• Criteria in the decision subnets appear in the third level. These are factors in the networks in the bottom level decision subnets that are not alternatives. We use the generic word criteria to include all the nodes including stakeholders, time periods and other concepts.

• Use the Wizard to generate the framework for your complex decision model (File menu>Open>New).

• Enter your control criteria and your alternatives. Modify the control criteria hierarchies to fit your model. Disconnect the goal node from the BOCR nodes in the main model. Make a cluster of strategic criteria and attach those to the goal node (in preparation for the ratings spreadsheet where you will evaluate the BOCR priorities). You may pairwise compare them at any time and build the ratings spreadsheet, but you must not enter the ratings until you have finished the bottom part of the model and determined the highest priority alternatives for Benefits, Opportunities, Costs and Risks.

• Complete your decision subnets in the third level by adding clusters of relevant factors.

• You should probably not use the identical structure in each decision subnet – fine-tune it to the type of decision subnet that it is. For example, if it is economic benefits you should put in factors that relate to that idea.
• Do not have “Sink” clusters. Clusters which have arrows pointing to them, but none pointing out are a waste of your time. The pairwise comparisons you make on the nodes in that cluster have no effect on the model. Priority flows around the paths in the model like water. It ultimately needs to get to the alternatives.
• Making Judgments. When you make connections in 3rd level decision subnets make sure the comparisons make sense. Can you formulate a question for each pairwise comparison? You should usually connect a node to at least two other nodes (usually in another cluster).
  • The same parent or source node may be connected to children nodes in more than one cluster. It is usually not good practice to connect a node to just one other node. There is no way to enter judgments as there is nothing to pairwise comparison. The priority of the parent node just passes directly to the child node. The software will do it okay, but it does not add useful information to your model.
  • Make the cluster comparison judgments. When making judgments be sure to make the cluster comparison judgments. Use the shortcut A<B to launch the cluster comparisons. The software will walk you through the necessary comparisons. The comparative question to be answered is: Which of two clusters impacts the parent cluster (of the comparison) more with respect to Economic Benefits (or whatever decision subnet you are in).
  • Do not make connections and then fail to make judgments. This will result in the positive information being input that all the factors in the resulting comparisons are equal – not the same thing at all as making no links. Unintentionally saying two nodes are equal may dilute actual knowledge you have put in during other comparisons.
• Sanity Check. Use the Computations>Sanity check command to find out if you have forgotten to do comparisons of nodes or clusters. Click on the buttons to the right to get details about where the problem lies.
• What about repeating node names, for example, Economic? The software treats nodes in separate networks that have the same name as different entities – except for the nodes in the Alternatives cluster. The nodes are usually conceptually different. For example, Economic as a strategic criterion in the top level network means “economic well-being”. As a control criterion in a control hierarchy it may mean “economic issues”. Finally, you may also have “economic” or some variation of that as a factor in several decision subnets and it may have subtly different meanings in each one.
• Formulas. The additive negative is the default formula that the wizard will use. You can easily change the formula under the Design command. You should synthesize the model in two ways: the additive negative formula and the multiplicative formula. The additive negative formula may result in some alternatives having a negative value (red bars) in the synthesized answer. The multiplicative formula never gives negative results.
• Sensitivity. Make sure the additive negative formula is selected when you do sensitivity on any of the BOCR nodes. It is not possible to do sensitivity under the multiplicative formula – the distinguishing factors cancel each other out.