Model Validity

1
Model Validity
2
Validating a Model

• Please note
• 1) A model is never perfect.
• 2) An absolute measure of validity does not
  exist.
• On what basis can we compare the validity of
  models?
• Four phases are suggested
• 1. An evaluation of model structure.
• 2. An evaluation of model logic.
• 3. An evaluation of design and/or input data.
• 4. An evaluation of model response.

3
An Evaluation of Model Structure

• Keep track of the model structure at all times.
• Best way is to start with a preliminary simple
  model and slowly build on it in a systematic and
  logical way.
• KISS (Keep It Simple, Stupid!)

4
An Evaluation of Model Logic

• If the model logic truly reflects the system
  model, then the model will react to a stimulus
  (or change) in the same way as the actual system
  would.
• Often, the relative difference in the outputs are
  all that we need to be concerned with (trends),
  rather than whether the model response and the
  actual system values are identical.

5
An Evaluation of Design and/or Input Data

• Two types of data in model development
• 1) the design data or information used to
  construct the model
• 2) the input data of data used to stimulate the
  system
• Data collection and verification may well be the
  most overlooked portion of model construction.
• Textbooks contribute to the problem because the
  data is often presented directly.
• The process of data collection often consumes the
  major amount of time and resources when dealing
  with actual problems.
• Good practice First decide on the basic form of
  the model, then identify its specific data needs.

6
An Evaluation of Model Response

• True validation is often said to be reflected
  solely in its ability to predict the behavior of
  the system that has been modeled.
• For example
• Consider a model which predicts the national
  economy based on the height of the hemlines on
  women's dresses (some obsrvers have noted a
  correlation).
• The fact that this model is accurate does not
  mean it is valid!
• Look at the response, the trends, and compare
  with actual data or theoretical information from
  other sources at hand to validate model response.