���������������������������������������������������������������
ing within the system can be represented by cohorts of identical ( or homogeneous ) individuals , or each individual can be separately represented in the model .
Many modeling techniques describe entities at the cohort or group level . A cohort captures all the individuals in the same state : for example , females aged 20 – 29 . For each cohort , the model tracks the number of individuals represented by the state . In cohort models , the individuals within the cohort are simply counted and hence treated identically . That is , any differences in behavior or relationships of individuals within the cohort are averaged out because individuals are indistinguishable from the others in the same state . This means that only differences captured by the definition of states ( gender and age in the example ) are assumed to have an effect . For those properties where states can change , a change in state nominally moves an individual from one cohort to another by changing the count for each affected cohort . Even where individuals do not have exactly the same behavior , a cohort model may be useful where the average behavior is a good approximation .
Alternatively , where there are significant relevant differences in behavior , additional states can be used to create smaller cohorts that are internally more homogeneous . In practice , the number of states can get out of hand very quickly because each combination of states must be considered . For example , for a disease progression model with two genders , four age groups , three risk factor status categories , and three disease history categories , there are 72 cohorts if fully modeled . Further , the model must also include the ways in which individuals move between all the cohorts ; so 72 cohorts will require over 5,000 transition arrangements ( though many are likely to be ‘ no transition possible ’). Therefore , it is important to limit the number of states wherever possible to avoid very complicated models that require many parameters for each relationship .
In contrast to cohort models that are simply counting individuals in particular state combinations , individual-based models keep track of each ( simulated ) individual . That is , if there are 25 individuals with the same set of states , a cohort model will group these individuals together and register that the cohort has 25 entities . An individual model , however , will have 25 instances of that entity in the model . Individual modeling is therefore resource intensive , but there are several reasons why it may be more appropriate .
Where there is substantial relevant heterogeneity , modeling individuals avoids the need to create a very large number of cohorts and relationships for transitions from one to another ( as already mentioned ). One important source of such heterogeneity is personal history or experience . For example , success in obtaining employment may depend on the pattern of previous employment rather than simply the proportion of time employed and the number of jobs — four years unemployed then three jobs of two years each is very different than one job of five years then two jobs of a few months each with two years unemployed between each . It would not be feasible to construct states for each pattern .
A different type of heterogeneity arises in models of systems where some concept of location or neighborhood is important . For some systems , this may be spatial or geographic . For example , in the land use model already described ( D ’ Aquino et al ., 2003 ; Lynam et al ., 2002 ), to water and other features such as soil type are inseparable from geography . For other systems , the neighborhood is defined by personal network rather than physical location . For
80