Download Modeling 101. Different Types of Models Designed for Different Uses

Modeling 101.
Different Types of Models Designed for Different Uses
Computer modeling is the process by which a computer is used to develop a mathematical model of a
complex system or process. Computer modeling is an efficient way to take into account many different
factors and to simplify and organize real-world processes. Models can be designed to better explain
or understand historical data, to predict future behavior or perform virtual experiments, or to make
decisions about courses of action based on the likelihood of expected outcomes. There are several ways
to organize and classify different types of models; below is a discussion of some examples.
Compartmental
Models
Agent-Based
Models
Compartmental models take a top-down, or macro-level, approach. To build the model, the
modeler groups individuals with similar characteristics into compartments (or classes) and
specifies compartment-wide rules of how individuals flow from one compartment to another.
These rules are compartment wide, meaning that they usually focus on the average behavior
of a group and do not model detailed individual behaviors. A common compartmental
epidemic model is the SIR model, with compartments representing susceptible, infectious,
and recovered groups of people or animals and equations describing the flow of the
population from one compartment to the next. The equations describing the flow can be
deterministic, so that the same starting conditions will always give the same results, or they
can be stochastic (also referred to as probabilistic), so that random factors are involved in
the model and variability of the model outcomes can be measured. Stochastic models may
be useful for determining the range of possible outcomes that could occur in real life.
Agent-based models (ABMs) are also dynamic models that simulate the behavior of a system
over time. However, they take a bottom-up, or individual-level, approach, specifying the rules
that govern the behavior of individuals and allowing the overall behavior of the system to
emerge from the interactions of those individuals. Like compartmental models, the ABMs group
individuals with similar characteristics. However, the transmission or flow between states, such
as healthy and infectious, is determined by the behavior of the individual, not the group as a
whole. ABMs are often useful when the input data are focused on an individual rather than
a group level or when a small change in the behavior of a few individuals could have a large
impact on the system. For example, ABMs can show how a virtual person might behave
in a simulated community. During an epidemic, each individual has a chance of catching
or spreading an infection through encounters with others at home, work, school, and
elsewhere. How the disease spreads through the population depends on whether and when
particular individuals encounter each other, as well as what their particular characteristics
are at the time of the encounter.
Decision-Analytic
Models
Decision-analytic models are different from agent-based or compartmental models
because, rather than simulating the behavior of a system over time, they are used
to help decide on a course of action based on the expected payoffs or outcomes of
different scenarios.
Decision-analytic models often employ matrices of payoffs for different possible
outcomes, which can be displayed as decision trees showing the branching points and
paths to the outcomes and probabilities describing the likelihood that each branch
or outcome will occur. The objective of a decision-analytic model is often to choose
the path to the optimum outcome given some criteria for success, whether that is to
maximize gain, to minimize risk, or to meet some other criterion.
Discrete-Event and
Fixed-Time Models
To “run” a model, whether it is a compartmental, agent-based, or decision-analytic
model, the parameters that describe the transmission and flow of people or animals
are set and the initial status of the population is defined. The model then calculates
the population status at the next time period. Time periods can be defined either in
terms of events or in fixed periods.
In a discrete-event simulation, the model processes by stepping from one event to
the next and updating the simulation time by the discrete time-step between those
events. On the other hand, fixed-time simulation updates the simulation time in fixed
periods and determines which events have happened since the last time-step occurred
in order to update the state of the model.
MIDAS. The Models of Infectious Disease Agent Study (MIDAS) is leading the research on using computational and
mathematical models to prepare the nation for responding to outbreaks of infectious diseases, whether these occur
deliberately or through natural means. Its work is funded by the National Institute of General Medical Sciences of the
National Institutes of Health.
MIDAS is a network of research groups who are developing models of the spread of infectious diseases. Each research
group consists of scientists from many disciplines, including computer scientists, epidemiologists, infectious disease
specialists, statisticians, computation biologists, informaticists, social scientists, veterinarians, and economists. All of
these researchers contribute to the building and testing of the models that involve the spread of disease in people and
in animals.
For more information about MIDAS
MIDAS Web site: http://www.midasmodels.org
National Institute of General Medical Sciences (NIGMS): http://www.nigms.nih.gov/Research/FeaturedPrograms/
MIDAS/
MIDAS Scientific Director at NIGMS: Irene Eckstrand, [email protected]
About MIDAS
Funded by the National Institute of General Medical Sciences, NIH, MIDAS is a collaborative network
of research scientists who use computational, statistical and mathematical models to understand
infectious disease dynamics and thereby assist the nation to prepare for, detect and respond to
infectious disease threats.
http://www.midasmodels.org