Sample Results 

Constraints

In a past version of CO (Pop91) the UK 1991 Household Sample of Anonymised Records was reweighted to fit 817 small-area Census counts drawn from 14 small-area Census tabulations.  The nature of these constraints is summarised here.

Goodness-of-fit

The fit of the synthetic microdata to known constraints is measured in two main ways: Absolute error and Z-score.  These measure respectively the absolute and relative (proportional) difference between Census and synthetic counts. 

The fit of the synthetic microdata estimated for the 1379 EDs in Leeds is summarised here.  For nearly 95% of all EDs, the fit to all 817 constraining counts is good.  For no ED is an entire constraining table poorly fitted.  Even when synthetic microdata are aggregated to ward level (33 wards in Leeds), the fit is good for an average of over 98% of all cell constraints.  

The fit of the synthetic microdata estimated for one Enumeration District (ED) in the city of Leeds to all 817 constraining Census count is presented here.  The goodness-of-fit for this ED is equal to the average achieved across all 1379 EDs in the city.