6 Conclusions
This paper has examined a range of minimum distance methods used to compute new weights for large cross-sectional surveys used in microsimulation modelling. The methods involve the use of extraneous information about a range of population variables, for calibration purposes. The distance functions were restricted to those for which the first derivative can be expressed as a function of the ratio of the new weights to the survey weights, and for which that function can be inverted explicitly. In general, an iterative solution procedure is required. An approach based on Newton’s method was described and numerical examples were given for several distance functions. Finally, the performance of the method was examined using the New Zealand Household Economic Survey. Rapid convergence of the iterations was obtained, although care needs to be taken when imposing limits on the proportional adjustments to sample weights. Since the same basic approach (and computer program) can easily examine a range of distance functions, and Newton’s method converges extremely quickly, it is relatively costless to consider the full range of distance measures. However, in practice, convergence cannot be expected using all measures.
Finally it is worth remembering that reweighting may cause the distribution of important variables, in particular alternative sources of income, to change.[19] Checks on changes in a range of distributions are therefore recommended.
Notes
- [19]This point is also made by Klevmarken (1998).
