Robert CLAY, University of Leeds, United Kingdom
Dynamic microsimulation is increasingly used to validate government policy. A synthetic population of individuals are aged forwards in time under hypothetical scenarios searching for optimal strategy. This ageing process requires transition probabilities. The next state of an individual is drawn randomly from a probability distribution conditioned on current attributes and behaviours. Modelling of this distribution is highly flexible and widely implemented lending microsimulation to a wide number of data types, research fields, and policy questions. Inherently it is a method well suited to undertaking applied policy experiments.
Good calibration of transition probabilities is essential to the functionality of a dynamic microsimulation. Poor prediction of next state undermines the validity of any estimated policy outcome. It is then surprising that there is little literature on best practice in the development of transition probability models. There have been numerous reviews into the general state of microsimulation but none (to the author’s knowledge) focusing specifically on best practice for estimating transition probabilities.
This research provides a review into the state of transition probabilities. Previous literature is explored to determine the common types of transition probabilities and known pitfalls. Four key areas of presentation, missingness, extrapolation, and heterogeneity are covered discussing methodology to overcome these issues.
This review is supplemented with a case study. A microsimualtion ‘Icarus’ is developed to estimate the effect of counterfactual employment policies on the mental health of vulnerable sub-groups of the UK population. This microsimulation is built using the Understanding society dataset and the vivarium open source python framework. Results demonstrate how transition probabilities can be improved drawing conclusions on best practice and speculation of what future innovations may hold.
Mots clés : Microsimualtion|Mental Health|Employment|Transition Probabilities|Review
A105363RC