Modelling Flashcards

1
Q

What are the six principles of modelling?

A

Principle 1: Model Simple; Think Complicated

Principle 2: Be Parsimonious; Start Small and Add

Principle 3: Divide and Conquer; Avoid Mega models

Principle 4: Use Metaphors, Analogies and Similarities

Principle 5: Do Not Fall in Love with Data

Principle 6: Modeling May Feel Like Muddling Through

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

What is Principle 1 of modelling according to Pidd?

A

Principle 1: Model Simple; Think Complicated

  • Models are a simple representation of complex things
  • A model cannot do everything
  • Models need to be transparent and easy to manipulate and control
  • A model cannot replace thinking → it should help people to think things through (part of the process of reflection before action)
  • Metaphor to a chain-saw
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

What is Principle 2 of modelling according to Pidd?

A

Principle 2: Be Parsimonious; Start Small and Add

  • Start simple and add complications, but only if necessary.
  • The model should be refined over time until it’s good enough. (iterative, validation)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What is Principle 3 of modelling according to Pidd?

A

Principle 3: Divide and Conquer; Avoid Megamodels

  • When large model is needed it can be useful to develop a set of simple models
  • Large models are difficult to validate, interpret and to explain
  • Each of the component models can be tested separately.
  • Risk of making more simple models: the component models embody subtly different assumptions, which may lead to very strange behavior when they are linked together
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What is Principle 4 of modelling according to Pidd?

A

Principle 4: Use Metaphors, Analogies and Similarities

• Metaphors, analogies, and similarities are used to help understand a new concept by relating it to something familiar.
• This principle is particularly useful for modeling complex systems where it can be difficult to understand the relationships between different components.
• A model can be developed by combining multiple analogies and similarities in order to build a complete representation of the system.
• Examples of metaphors and analogies used in modeling include:
o Using a river as an analogy to describe the flow of goods in a supply chain.
o Using a traffic flow model to describe the flow of information in a computer network.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What is Principle 5 of modelling according to Pidd?

A

Principle 5: Do Not Fall in Love with Data

  • Modeling should drive data collection, not the other way around.
  • The decision about which type of model is needed should be decided before data collection begins.
  • Process of building a model:
    o The modeler uses preliminary data to get an idea of what type of model is needed. This is quantitative AND qualitative data.
    o Then the model built from the preliminary data drives the new data collection.
    o Avoid using the same data to build and test a model.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What is Principle 6 of modelling according to Pidd?

A

Principle 6: Modeling May Feel Like Muddling Through
- Modelling is NOT a linear process → you need to go back and forth
- Usually 60% of the time is spend on model structure

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Pidd mentions three important features of models. What are those?

A

1) The model is external and explicit – this means that the model can be examined, challenged, and written in a logistical language (mathematic or in computer programming)

2) The model is a representation of the real world. (The problem is that people see things in different ways)

3) No model will be a complete representation of reality

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Black box model:

A

A black box is one with unknown contents (the variables and relationships), where the analysts tries to understand the model by analyzing its outputs under defined inputs

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Transparent box modelling:

A

Everything is visible, we know the data, and we know the behavior and the outcome that the model will come up with.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Grey box Model

A

However, it is usually true that the behavior of a model of any complexity is not fully understood without some investigation on the part of the analyst. The model is grey and not transparent because there might be some behaviors that surprises the modeler and the model user. This is because a model is a system of objects and their relationships, within the defined boundaries.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Compass modelling:

A

A model used as a compass will not provide the accurate estimates or prediction in its outputs, but nevertheless it can still be extremely useful in skilled hands. Often when modelling in the way of the compass the people are very skilled and they are aware of the options available but seek some confirmation, and they have no intention of basing their entire decision making on the model, but they use it to get a direction

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

GPS modelling

A

• Users need to place great trust in the model and the results it produces.
• Such models are used as the sole basis of decisions and plans, hence require great attention to detail, validation, and verification.
• Such models are built with the expectation of significant payback, as they require investment in construction and data.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

What are the 4 archetypes of model use according to pidd 2010 and what are the two measures?

A

1) Decision automation,
2) Routine decision support,
3) System investigation & improvement
and 4) Providing insights for debate.

Routine use and human interaction

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Decision automation

A

A user operating in this mode supplies data to the models. The model should of course be monitored occasionally and updated. They replace the human decision making, but they are made such that a single wrong decision will not bring the business to its knees. The model should be fast and linked to the other systems of the company.

Examples: Regression, consumer credit scoring (allowance for a loan or credit card), dynamic pricing via websites (hotels, airline tickets), and automatic reordering of routine stocks.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Routine decision support:

A

The model is used to assist, but not replace people making routine related decisions. It is common for real-life to be more complex or rather less certain than can be wholly encoded in a model.
The model is used to reduce the set of options to be considered by human decision makers, that has information available not included in the model

Example: Scheduling ressources such as people and equipment in application areas such as operating theaters or airlines and trains another is forecasting.

17
Q

Modelling for investigation and improvement:

A

Models are used to support investigations that are relatively unique, which may involve system design, system improvement or just an attempt to gain understanding of a very complex situation. This model use typically relies on a model that is purpose-built for a particular study. Since models are used in this way, they are likely to be much more approximate than those used for decision automation and for routine decision support.

Examples: Location of a distribution depot, investigation of ways to improve performance in an accident and emergency department, design of business processes to support automated decision making etc.

18
Q

Modelling to provide insights

A

Multiple stakeholders –> different agendas

Often disagreement about the ends and means.

Used to explore options and reduce uncertainty, rather than taking a decision

It is often larger decisions that needs to be made.

Often simulation models are used in this mode by helping people to gain insights so they can move to a form of accommodation that leads to action.

19
Q

Importance of model validation for decision automation:

A

The validity is an important concern for model builders and users, since the models will be allowed to actually take decisions and to commit ressources. There is great effort made to ensure that the relationship between the factors of the model are correct, and that the output perform as expected given defined inputs (black box validation)

Black box and white box validation is very much used.

20
Q

Importance of model validation for Routine decision support:

A

Validity is also an important issue here, though it is slightly less crucial since humans are part of the decision process. There will be put a lot effort into black- and white box validation so as to build confidence that using a decision model will lead to better decisions.

21
Q

Importance of model validation for Modelling for investigation and improvement:

A

Validation here is less straightforward. This is because this mode of model use is often employed to explore options for system configuration that do not exist. It is usual to fall back on white box validation, where assumption of the model and parameters are critically examined

22
Q

Importance of model validation for Modelling to provide insights:

A

Validation is particularly problematic here, as they are not would-be representations of the real-world, but rather attempts to understand and represent how different stakeholders and interest groups sees the world.

23
Q

Data requirements for decision automation:

A

Require very detailed, accurate and up-to date data, as it forms the core of a routine decision-making process. The parameters of the models must be finely tuned and, likely as not, must be frequently revised to further improve.

24
Q

Data requirements for Routine decision support:

A

They are like those for decision automation, data hungry. Modelers usually go to great lengths to ensure that they are parametrized from high quality data. However, unlike decision automation, there is a recognition that the models can never include all the factors likely to be relevant.

25
Q

Data requirements for Modelling for investigation and improvement:

A

Much effort typically focuses on understanding how the system does or might operation, and in mapping the interactions that from the behavior of the system. Data are much less demanding.

26
Q

Data requirements for Modelling to provide insights:

A

The data requirements are much broader here. Indeed, much of the data may be qualitatively, being explications of people’s beliefs and assumptions about causal relationships or desirable outcomes.

27
Q

How does Pidd 1999 define a model in OR?

A

“A model is an external and explicit representation of part of reality as seen by the people who wish to use that model to understand, to change, to manage, and to control that part of reality in some way or other.”

28
Q

Value added with decision automation:

A

Intended for use on a day-to-day basis to take decisions that could have been made by humans but are made by models because performance is better. Better performance might relate to speed, their relative accuracy, their costs, or the audit trail they generate. Models replace human decision makers because of their superior performance, and because they ensure uniformity and consistency.

Drawbacks: 1) Loss of expertise in the organization, once model developers have left the scene there is a risk of insufficient expertise, 2) model-groupthink, in which competing organizations base their decisions on very similar models.

29
Q

Value added with Routine decision support:

A

They are used because there is evidence or belief that better decisions result from their use in conjunction with human judgement, but here the model alone cannot match the variety of the system, whereas the combination of model and human insights can provide the requisite variety.

Drawbacks: 1) adding human intervention can actually make things worse in some situations. 2) the risk that people who do know better may find it hard to marshal the evidence to modify or contradict the advice given by the model

30
Q

Value added with Modelling for investigation and improvement:

A

Enables people to reflect before implementation. The model is used as an experimental vehicle in which options may be explored. Models used in this way are often a part of risk reduction strategies. Because they are explicit and external with a tangible representation, their use enables users to follow through and realize the effects of different policies and plans.

Drawbacks: 1) they may not be valid at all

31
Q

Value added with Modelling to provide insights:

A

It is difficult to say what value it brings. In some cases, the model can provide a neutral language in which participants and stakeholders can start to appreciate other viewpoints. They may help see how events play out:
Drawbacks: bias and manipulation re the main risks

32
Q

GPS vs Compass

A

With NatMilk we have created a compass that tells us that we should probably build our own network. Had all the data been accurate we have built more of a GPS.