Presentation at 'Moving beyond tax-benefit and demographic modelling', University of Leeds on July 2nd 2009. One of a series co-funded by the ESRC and the BSPS on ‘Microsimulation modelling in the UK: bridging the gaps’.
Estimating income, expenditure and time-use within small areas
1. Estimating income, expenditure and time-use within small areas Ben Anderson, Paola De Agostini & Tony Lawson ESRC/BSPS Microsimulation Seminar Series University of Leeds July 2nd 2009
There are a range of statistical methods Multilevel and hierarchical modelling etc But we’re not using them We’re creating a synthetic ‘Income Census’ We fill each ‘area’ (LSOA)… with ALL households from the relevant region Then give them fractional weights so that key constraint variables in each area match known Census distributions
(and must be in the Census or Census-like data) Relatively few candidates in the Census
There are a range of statistical methods Multilevel and hierarchical modelling etc But we’re not using them We’re creating a synthetic ‘Income Census’ We fill each ‘area’ (LSOA)… with ALL households from the relevant region Then give them fractional weights so that key constraint variables in each area match known Census distributions
OECD scale : 1 (first adult) + 0.5 for extra adults & children except 0.3 for child < 14
No external validation available (yet) for expenditure
Uses Stephen Jenkins’ stata ineqdec0 command to include 0 values
No external validation available (yet) for expenditure
,408 = Ghana, Turkmenistan & United States .268 = Bosnia, Hungary, Finland http://en.wikipedia.org/wiki/List_of_countries_by_income_equality
R sq = 21%
Total minutes per day used as sums total behaviour - e.g. what if per minute charging for service? Also indicator of broadband bandwidth demand.
R sq = only 10.9% Total minutes per day used as sums total behaviour - e.g. what if per minute charging for service? Also indicator of broadband bandwidth demand.
Can’t show demand system modelling results (yet) but here a fictitious example of what we’ve been doing
Validation against the census is excellent!
Total minutes per day used as sums total behaviour - e.g. what if per minute charging for service? Also indicator of broadband bandwidth demand.
Traditional static microsimulation
Would be very interesting to look again at this as broadband changes the way TV is delivered
Demand system: ML method, as gets more complex (e.g. more demographics) gets even slower