2013/03/21 by Robin Lovelace, Lovelace, Robin, Dimitris Ballas +1 · 2 citations
Decision Sciences · Economics, Econometrics and Finance · Mathematics · Social Sciences · #FOS: Computer and information sciences #Insurance, Mortality, Demography, Risk Management #Methodology (stat.ME) #Migration, Aging, and Tourism Studies #Spatial and Panel Data Analysis #demographic modeling and climate adaptation #stat.ME
paper · pdf · doi:10.48550/arxiv.1303.5228
51 pages, 10 images (including supplementary information)
arxiv created 2013/03/21 · openalex publication_date 2013/03/21 · arxiv updated 2013/03/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Iterative proportional fitting (IPF) is a widely used method for spatial microsimulation. The technique results in non-integer weights for individual rows of data. This is problematic for certain applications and has led many researchers to favour combinatorial optimisation approaches such as simulated annealing. An alternative to this is `integerisation' of IPF weights: the translation of the continuous weight variable into a discrete number of unique or `cloned' individuals. We describe four existing methods of integerisation and present a new one. Our method --- `truncate, replicate, sample' (TRS) --- recognises that IPF weights consist of both `replication weights' and `conventional weights', the effects of which need to be separated. The procedure consists of three steps: 1) separate replication and conventional weights by truncation; 2) replication of individuals with positive integer weights; and 3) probabilistic sampling. The results, which are reproducible using supplementary code and data published alongside this paper, show that TRS is fast, and more accurate than alternative approaches to integerisation.