Create a model
Below are general guidelines to help you build models that provide the greatest value with the least effort.
Identify the objectives: What are the objectives of the decision maker? Sometimes the objective is simply to maximize expected monetary profit. More often there is a variety of other objectives, such as maximizing safety, convenience, reliability, social welfare, or environmental health, depending on the domain and the decision maker. Utility theory and multi-attribute decision analysis provide an array of methods to help structure and quantify objectives in the form of utility. Whatever approach you take, it is important to represent the objectives in an explicit and quantifiable form if the objectives are to be the basis for recommending one decision option over another.
It is a useful convention to put the objective variable or variables (hexagonal nodes) on the right of the diagram window, leaving space on the left side for the rest of the diagram.
The most common mistake in specifying objectives is to select some that are too narrow, by concentrating on the most easily quantifiable objective — typically, near-term monetary costs — and to forget about the other, less tangible objectives. For example:
- When buying software you might want to consider the usability and reliability of different software packages, as well as long-term maintenance, not just cost and performance.
- In pricing a product, you might want to consider the long-term effects of increased market share in developing new customers and markets and not just short-term revenues.
- In selecting a medical treatment, you might want to consider the quality of life if you survive the treatment, and not just the probability of survival.
Identify the decisions: The purpose of modeling is usually to help you (or your colleagues, organization, or clients) discover which decision options best meet your (or their) objectives. You should aim, therefore, to include the decisions and objectives explicitly in your model.
A decision variable is one that the decision maker can affect directly — which computer to buy, how much to bid on the contract, which medical treatment to choose, when to start construction, and so on. Occasionally, people want to build a model just for the sake of furthering understanding, without explicitly considering any decisions. Most often, however, the ultimate purpose is to make a better decision. In these cases, the decision variables are where you should start your model.
When starting a new influence diagram, put the decision variables — as rectangular nodes — on the left of the diagram window, leaving space for the rest of the influence diagram to the right.
Link the decisions to the objectives: The decisions and objectives are the starting and ending points of your model. When you have identified them, you have reduced the diagram construction to the process of creating the links between the decisions and objectives, via intermediate variables. You might wish to work forward from the decisions, or backward from the objectives. Some people find it easiest to alternate, working inward from the left and the right until they can link everything up in the middle.
It helps to identify the decisions and objectives early in model construction, to keep the focus on what matters. There can be a bewildering variety of variables in the situation that might seem to be of potential relevance, but, you only need to worry about variables that influence how the decisions might affect the objectives. You can ignore any variable that has no effect on the objectives.
Focus on identifying the variables that make clear distinctions — variables whose interpretations won’t change with time or viewer. Extra effort here will be repaid in model accuracy and cogency.
Move from the qualitative to the quantitative: An influence diagram is a purely qualitative representation of a model. It shows the variables and their dependencies. It is usually best to create most or all of the first version of your model just as an influence diagram, or hierarchy of diagrams, before trying to quantify the values and relationships between the variables. In this way, you can concentrate on the essential qualitative issues of what variables to include, before having to worry about the details of how to quantify the relationships.
When the model is intended to reflect the views and knowledge of a group of people, it is especially valuable to start by drawing up influence diagrams as a group. A small group can sit around the computer screen; for a larger group, it is best if you have the means to project the image onto a large screen, so that the entire group can see and comment on the diagram as they create it. The ability to focus initially on the qualitative structure lets you involve early in the process participants who might not have the time or interest to be involved in the detailed quantitative analysis.
With this approach, you can often obtain valuable insights and early buy-in to the modeling process from key people who would not otherwise be available.
Keep it simple: Perhaps the most common mistake in modeling is to try to build a model that is too complicated or that is complicated in the wrong ways. Just because the situation you are modeling is complicated doesn’t necessarily mean your model should be complicated. Every model is unavoidably a simplification of reality; otherwise it would not be a model. The question is not whether your model should be a simplification, but rather how simple it should be. A large model requires more effort to build, takes longer to execute, is harder to test, and is more difficult to understand than a smaller model. And it might not be more accurate.
“A theory should be as simple as possible, but no simpler.” Albert Einstein
Reuse and adapt existing models: Building a new model from scratch can be a challenge. If you can find an existing model for a problem similar to the one you are now facing, it is usually much easier to start with the existing model and adapt it to the new application. In some cases, you might find parts or modules of existing models that you can extract and combine to address a new problem.
To find a suitable model to adapt, you can start by looking through the example models distributed with Analytica. If there is an Analytica users’ group in your own organization, it might collect a model library of classes of problems of interest to your organization.
“If I have seen further than [others] it is by standing upon the shoulders of Giants.” Sir Isaac Newton
Aim for clarity and insight: The goal of building a model is to obtain clarity about the situation, about which decision options will best further your objectives, and why. If you are already clear about what decision to make, you don’t need to build a model, unless, perhaps, you are trying to clarify the situation and explain the recommended decisions for others. Either way, your goal is greater clarity. This goal is another reason to aim for simplicity. Large and complicated models are harder to understand and explain.
- Tutorial: Create a model
- To open or exit a model
- Combining models into an integrated model
- Test and debug your model
- Tutorial: Sharing a model with ACP
- Example Models
- Example Models and Libraries
- Guidelines for creating lucid models
- Model file formats
- Model Licensing