SPATIAL MICROSIMULATION: A METHOD FOR SMALL AREA LEVEL ESTIMATION Dr Karyn Morrissey Department of Geography and Planning University of Liverpool Research Methods Festival, 2014 Rationale for Microdata Much modelling in the social sciences takes an aggregate or mesolevel approach. However, all government policy and investment has a spatial impact, regardless of the initial motivating factor. As such, policy level analyses call for individual or household level analysis at a disaggregated/local spatial scale. Particularly Health Policy Health is a produce of individual and social factors that vary geographically Why Simulate? Data Issues Census data: Available at the small area level does not offer any information on household income Survey data often contains detailed micro data, for example income, pensions and health data that is not included in the census - aspatial in nature Spatial Microsimulation offers a means of synthetically creating large-scale micro-datasets at different geographical scales. Aspatial Microdata Census Outputs at the small area level Matching Process Combinational Optimisation Methods, Reweighting, IPF Open source algorithm for each of these are increasingly available Synthetic Population Data Validation of unmatched variables Satisfactory Unsatisfactory Calibration through alignment Objective: Sum of MSIM Outputs are equal exogenous data target Estimate variable of interest using regression Create Alignment Co-efficient E.g.: SMILE’s Market Income Variables are each adjusted by multiplying the appropriate estimated individual earnings by the alignment coefficient E.g.: Fully calibrated micro-level earnings for Ireland SMILE SMILE is a Spatial Microsimulation Model My lovechild and sometimes referred to as SLIME depending on how it is behaving Using a statistical matching algorithm, simulated annealing, SMILE merges data from the SAPS and the Living in Ireland survey (income & health data) SMILE creates a geo-referenced, attribute rich dataset containing: The socio-economic, income distribution & health profile of individuals at the small area level Model Components & Analysis to Date Components: Agricultural/Farm Level Model; Family Farm Income Analysis (Hynes et al., 2009) Environmental Model; Conservation & Agri-Environmental Analysis (Hynes et al., 2009) Recreation Model; Walkers Preferences (Cullinan et al, 2008) Health Model; Access to GP Services (Morrissey et al., 2008) & the Spatial Distribution of Depression (Morrissey et al., 2010), Determinants of LTI (Morrissey et al., 2013) Income Model Labour Force Participation & it’s impact on Income (Morrissey and O’Donoghue, 2011) Marine Sector analysis Impact of the marine sector on incomes at the small area level (Morrissey et al., 2014); Impact of marine energy on the small area level in Ireland (Farrell et al., forthcoming) RGS-IBG Edinburgh, 3-5th of July, 2012 Health Application The spatial distribution of demand for acute hospital services (AHS) (Morrissey et al., 2009) It was found that demand for AHS was highest in the West & NW of Ireland Why? National Level Logit found that main-drivers of AHU are: Medical Card Possession Age LTI Is there a Spatial Pattern to theses Drivers which explains AHU at the ED Level? Drivers of AHU at the ED Level Exogenous Models Spatial Microsimulation models may be linked with other exogenous models Models may be either spatial or aspatial Linking to these models to a spatial microsimulation models allows their macro level results to be spatially disaggregated Supplementary Models Tax-Benefit Model Spatial Interaction Model Income Analysis Application • • • • • Incorporating a TBS into SMILE – Average Disposable Income was generated East of the country - higher levels disposable income 4 urban centres - higher than average disposable income CSO - provides county level estimates of disposable income Real value added by SMILE’s Examine the distribution of income within counties • Disposable income - low along the coastal regions of the West • Counties with urban centres, income higher in the in these counties than in the rural areas Accessibility Analysis: Health Service Application RHS: Access to a GP facility Spatial Interaction Model LHS: Probability of Using a GP service given one’s SocioEconomic Profile Logistic Model A Spatial Microsimulation Model of Comorbidity New UK work ESRC SDAI Funded Develop a spatial microsimulation model for comorbidity Whilst small area register data on single morbidities exist and may be accessible to researchers These only report 1 morbidity Comorbidity is an increasingly important health issue With both demand and supply side implication Comorbidity at the small area level Develop a model of comorbidity between CVD, diabetes & obesity at a small area level for England East Kent Hospital Trust our case partner The ESRC Secondary Data Analysis Initiative for funding this research. Post-Doc: Dr Ferran Espuny Conclusion Spatial microsimulation – computationally and data intense However, there are now open source software for microsimulation that offer the shelf models – all you need is to prepare the data Harland et al., (2012) Comorbidity model presented will be open source Always necessary to look at the spatial implication of policy and investment Spatial microsimulation model offers one way to do this Validation (and calibration) is key if the data is to be used to inform policy