Attribute nonattendance in discrete choice models: measurement of bias, and a model for the inference of both nonattendance and taste heterogeneity
Field | Value | Language |
dc.contributor.author | Collins, Andrew | |
dc.date.accessioned | 2013-03-08 | |
dc.date.available | 2013-03-08 | |
dc.date.issued | 2012-08-22 | |
dc.identifier.uri | http://hdl.handle.net/2123/8966 | |
dc.description.abstract | An extensive literature has recognised that when discrete choices are made, only a subset of the attributes of the choice alternatives may be considered or attended to by each decision maker. The wider literature suggests that attribute nonattendance (ANA) is important, and that failure to recognise ANA contributes to biased model outputs such as willingness to pay measures, masked sensitivities, implausibly signed random parameters coefficients, and exaggerated taste heterogeneity. It may also be of intrinsic interest to the analyst, and reveal problems with stated choice experimental designs. This research uses simulated data to gain a deeper understanding of the biasing influences of ANA. It is shown that random parameters logit models handle ANA poorly, with the extent of the bias in the model outputs driven by both taste heterogeneity and ANA. The literature has identified some shortcomings and limitations of the existing methodologies for handling ANA. The simulated data are employed to further critique these methodologies, and demonstrate that they are likely to introduce their own biases. This thesis proposes the random parameters attribute nonattendance model, and seeks to improve upon the existing methodologies. The model combines discrete and continuous random parameters, and can infer ANA and taste heterogeneity, without the need to collect supplementary data. The model is tested on simulated data with encouraging results. In addition, in an empirical setting of short haul flight choice, with real stated choice data, the model outperforms the RPL model and several existing ANA methodologies. Necessary conditions to ensure identification are discussed. The model allows for a balance between parsimony, and the handling of correlation of ANA, through a spectrum of possible model specifications; with this tension explored in detail. Further insights into ANA behaviour are gained in the empirical study. The thesis makes a useful methodological contribution, by developing a model with unique properties that can capture the important and prevalent behaviour of ANA. | en_AU |
dc.rights | The author retains copyright of this thesis. | |
dc.subject | attribute nonattendance | en_AU |
dc.subject | random parameters attribute nonattendance model | en_AU |
dc.subject | latent class model | en_AU |
dc.subject | random parameters logit | en_AU |
dc.title | Attribute nonattendance in discrete choice models: measurement of bias, and a model for the inference of both nonattendance and taste heterogeneity | en_AU |
dc.type | Thesis | en_AU |
dc.date.valid | 2012-01-01 | en_AU |
dc.type.thesis | Doctor of Philosophy | en_AU |
usyd.faculty | The University of Sydney Business School, Institute of Transport and Logistics Studies (ITLS) | en_AU |
usyd.degree | Doctor of Philosophy Ph.D. | en_AU |
usyd.awardinginst | The University of Sydney | en_AU |
Associated file/s
Associated collections