IV - Accuracy
While the policy analyst is likely to
focus on the functionality of the model , the modeler ’ s attention may be directed to how accurate the model is . That is , how well does it capture the relevant features of the target system ? Accuracy is particularly important for models that are used to compare options or make specific forecasts about potential future events . However , even models that simply represent features and relationships benefit from greater accuracy because they capture and communicate knowledge instead of misconceptions .
There are two major elements to developing accurate models . The first is the quality of the abstraction of the relevant features and relationships while constructing the model design , already described . It is important that the design process effectively incorporates a broad range of expertise so as to understand and describe the important elements of the target system that should be represented in the model .
The second is the quality of the implementation of that design to create the final model . The Test phase corrects the more obvious errors . However , there is a more subtle issue that is implicit in the selection of the modeling technique . It is easy to select a familiar technique , but such selection can introduce substantial inaccuracy because each technique is designed to model systems with specific characteristics and frames the target system in a different way . Accuracy fundamentally depends on how well the assumptions of the technique chosen match the structure of the target system so that distortions are not introduced .
This section describes the assumptions imposed by different methods and identifies some of the methods that make those assumptions . A summary of the various methods mentioned with a brief description and references follows ( see section 7 ). However , the intention is not to prescribe a particular method for specific characteristics , but rather to demonstrate the variety of methods available and provide guidance on the issues to discuss with the modeler .
������������������������������
The distinction between quantitative and qualitative appears simple .
Quantitative information is expressed with some numerical measure , such as time , distance , or money . These values are also objective , different measurements taken at different times by different observers will generate the same number . In contrast , qualitative information does not have a natural numerical measure , and includes abstract concepts like opinions or moderators such as ‘ very ’. The scoring of qualitative information is subjective , with potentially different values assigned by different assessors . University course designers and textbook publishers routinely classify research methods as quantitative ( e . g ., linear regression ) or qualitative ( e . g ., interviews ).
In practice , there is a great deal of ambiguity when classifying modeling methods . This is primarily because there are two properties of methods , which can be assessed as either quantitative or qualitative ; the type of information that is represented within the model , and the way in which the relationships between model entities are expressed . Consider a simple price model for widgets that cost $ 4 to manufacture and 100 per week are sold for $ 5 , for a profit of $ 100 per week . This information is clearly quantitative . However , the relationship between price and amount sold could be expressed qualitatively , such as ‘ a large price increase would reduce sales ’. Thus , it is not necessary for the type of information to match the type of relationship .
76