Multilevel simulation models
In the 1980s and early 1990s, the first simulation environment – MIMOSE – was developed for simulating interacting populations. Figure 6.1 shows the template for this kind of simulation model. The typical case is that there is a population with its attributes (for example, its size, its birth and death rates and its gender distribution), homogeneously consisting of a possibly great number of individuals with their own attributes (such as sex, age, political attitudes or annual income). Population attributes depend on aggregated individual attributes, and these in turn depend on the population attributes. For example, the gender distribution of the population depends on how many individuals are male and how many female, and whether an individual is born or dies depends on the population’s birth and death rates, which in turn may depend on the population size and sex ratio (see p. 47).
For computational reasons, a cyclic dependence within the same time step is forbidden, hence typically in each simulation step individual attributes are evaluated as depending on the values that the population attributes had in the previous simulation step, and population attributes are evaluated after all individual attributes in the same simulation step. From the outset, MIMOSE allowed an unrestricted number of object types (which may, but need not, be seen as levels in the sense of Bunge 1979: 13).
Hence, MIMOSE is also capable of performing classical microsimulation (see Chapter 4) where we often have a large number of persons, each belonging to a household, all of them making up a (sample of a) population. Earlier microanalytical simulation models did not include a complete feedback loop between persons or households and population in both directions.