Spatial microsimulation

Dr Andy Evans

[Fullscreen]

The big idea

Build a synthetic population of individuals where we only have aggregate statistics

Typical problems solved:

Area statistics need reaggregating to different spatial units

Need to estimate missing population statistics

Need an individual level population for modelling

Microsimulation

Spatial vs aspatial

Static vs dynamic:
Dynamic uses transitional probabilities to roll static forward in time as a model.

Integer vs probabilistic

We're going to look at spatial static integer based microsimulation

Idea

We have area statistics and wasn't to recreate the underlying population

e.g. census statistics

We need a sample of a population with the same characteristics/P>

e.g. census sample of anonymised records, British household panel survey

Method

We fill the areas with individuals randomly

We then subtract an individual and add in a new one

If the aggregate statistics improve, we keep the change, otherwise we revert

We repeat until our error is ok

Are other methods resulting in probabilistic/floating-point populations, for example, iterative proportional fitting

Ancillary variables

At this point we have a disaggregated population

However, if our microdata individuals have other variables attached to them that are correlated with the constraining variables, we now have an estimate of these in our area populations

Standard regression will give us a lower bound on the quality of this estimate

Then...

Reaggregate to new geography

Roll individuals forward in time using probabilities of life changes, based on scenarios

Use as basis for agent based models etc.

Software

Flexible Modelling Framework by Kirk Harland

Modular, also comes with cluster analysis and graphing plugins. Just delete jar files and reboot if you don't want them

Open source, so security can be checked