Calibration Strategies to Reweight Travel Surveys: the Case of Households Urban Travel Surveys in France
Rendina, IFSTTAR, Rabaud, CEREMA, Hasiak, CEREMA
This paper aims to lead a discussion on the choice of auxiliary variables to be used during the weighting process of travel surveys.
All surveys are susceptible to a variety of different types of error that affect different parts of the survey process. These errors have different implications for data quality and are amenable to different forms of prevention or compensation (Groves, 1989; Richardson et al., 1995). The unit nonresponse refers to the failure of a unit in the sample frame to participate in the survey. While nonresponse results in a reduced sample size, a more important concern of researchers is the possible impact of nonresponse bias. Indeed, bias is introduced when those that do not respond to the survey are systematically different from those that do respond. As there is no justification for assuming that people who respond have the same characteristics as those who do not (Forsman et al., 2007). Thus, in computing estimates from the data collected, we may face biases, the size and direction of which are unknown.
Generally a weighing method is used to compensate these biases. For instance by the use of Calibration on margins, it is a weight-class method used when the total of each auxiliary information is known. Calibration on margins is an iterative process that adjusts certain sample totals or ratios, to make them match with certain corresponding totals or ratios that are known from the population. This stage is essential to ensure a representative sample and the comparison with some others statistics sources (for instance, census data or other surveys). The calibration on margins must be implemented both on variables which explain (or are correlated with) transport behavior, and also on the variable that explain the nonresponse mechanism, for which the total is accurately known (Deville, 1999).
The drawback of such technique is that if we add too much auxiliaries information in the calibration we may reduce the accuracy of estimates. This paper aims to lead a discussion on the choice of auxiliary variables to be used during the weighting process. To test different options, we will use the last 20 households urban travel surveys in France collected during the period 2009-2014 and spread over the country. This large database of about 100,000 households allows very fine analysis on response behavior, the differences with the census and therefore bias that one can do
Association for European Transport