What is modelling?
Modelling is an important part of decision science. Modellers make simplified representations of reality to produce judgements about reality and guide real decisions. Models can be abstract and theoretical, or data driven and empirical. Both categories can be prescriptive, predictive, or descriptive.
Modelling is a cost-effective way to test the successes and failures of potential projects without committing large-scale resources. Modelling is also useful as means of building support necessary to successfully launch projects. Modelling can inspire leadership toward more successful project, policy, or process design and it can warn leadership of potential losses, and thereby guide strategic decision making and planning. Modelling is also a successful means of supporting decisions with less evident risk vs reward metrics. This is especially important when teams are not similarly directionally aligned, and there are challenges getting everyone on board.
How to model?
Before modelling can begin, modellers need to engage in some level of feasibility study. This process is greatly improved with experience. The feasibility study allows modellers to assess the modelling data plan and engage in preliminary levels of model selection. After the feasibility study is completed, management should know what modelling will cost, how long it will take, and have a rough indication of what range of results can be expected. In the feasibility stage, there should be descriptive charting and analysis, but not yet mathematical analysis. After feasibility, data needs to be gathered and prepared. This can be a long process, but is made much shorter through experience. When data are finally ready, and in many cases before, model selection happens again. Is it still advantageous to proceed with the original model selection? Next, the model must be built. This is the heavy lifting. Where existing modelling frameworks are inadequate new maths must be programmed. Finally, the model will be run, debugged, and run again. In some cases, sensitivity analysis can be performed. This will yield the most probable range for results when the model is run repeatedly across varying parameters.
|Descriptive models||Decision models|
|Best Case/worse case analysis||Pay-off matrix modelling|
|What-if analyis||Decision tree analysis|
|Break-even modelling||Cost-benefit modelling|
|Exploratory Data Analysis||Probabilistic models|
|Computational models||Econometric Models|
|Linear and Non-linear Programming||Multivariate regression analysis|
|Network Models||Probit and Logit models|
|Goal Programming||Non-parametric and parametric modelling|
|Computational General Equilibrium||-|