Population Synthesis in Activity-Based Models: Tabular Rounding in Iterative Proportional Fitting. Issue 1 (January 2015)

• Record Type: Journal Article
• Authors: Choupani, Abdoul-Ahad; Mamdoohi, Amir Reza
• Is Part Of: Transportation research record. Volume 2493:Issue 1(2015)
• Page Start: 1
• Page End: 10
• Publication Date: 2015-01
• Subjects: Transportation -- Periodicals; Roads; Transport -- Périodiques; Routes -- Périodiques; Routes -- Conception et construction -- Périodiques; Roads; Transportation; 388.05
• DOI: 10.3141/2493-01
• Languages: English
• ISSNs: 0361-1981

Abstract

Activity-based travel demand modeling requires socioeconomic micro-data of the population under study. Because the acquisition of such data for the entire population is infeasible or highly expensive, techniques such as iterative proportional fitting (IPF) have been applied extensively to estimate such data for the population, synthetically. Despite its many advantages, IPF results in noninteger values instead of integers: fractions of households or individuals are obtained for zones. Although methods have been proposed to produce integers for noninteger tables, seldom has this problem been viewed as tabular rounding. This paper proposes a binary linear programming model for tabular rounding in which the integer-converted table totals and marginal sums perfectly fit the input data obtained from the Census Bureau. Furthermore, the model minimizes distortion to the correlation structure of the household- and individual-level noninteger tables. The model does not bias the joint or marginal distributions of the socioeconomic attributes of population units or the sampling of rare demographic groups (at a significance level of a = .05). The empirical comparison of the proposed method with eight existing methods demonstrates that the proposed model outperforms the tested methods. Sensitivity analysis demonstrates that the integer conversion of small values is as significant as it is for large values. In this study, deterministic methods outperform stochastic methods in accuracy and perfect fit to census data. Finally, a scoring and ranking tool is used to reflect concisely the advantages and disadvantages of these methods.