Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Misspecified Hybrid Choice Models: An empirical study of parameter bias and model selection.
KTH, School of Architecture and the Built Environment (ABE), Urban Planning and Environment, System Analysis and Economics. (CTS)ORCID iD: 0000-0003-4512-9054
2018 (English)Other (Other academic)
Abstract [en]

Model misspecification is likely to occur when working with real datasets. However, previous studies showing the advantages of hybrid choice models to account for measurement errors have mostly used models where structural and measurement equations match the functions employed in the data generating process, especially when parameter estimate biases were discussed.

 

The aim of this study is to investigate the extent of parameter bias in misspecified hybrid choice models, assess if different modelling assumptions required to make the hybrid choice models operative impact the parameter estimates of the choice model, and evaluate the prediction accuracy of misspecified hybrid choice models in comparison with a simpler, also misspecified, multinomial logit. For these tasks, a mode choice model is estimated on 100 synthetic datasets. The synthetic datasets were designed to mimic the conditions present in real datasets; hence the postulated structural and measurement equations of the hybrid choice models are less flexible than the functions used for the data generating process.

 

Results show that hybrid choice models, even if misspecified, manage to recover better parameter estimates than a multinomial logit. However, hybrid choice models are not unbeatable, as results indicate that misspecified hybrid choice models will still yield biased parameter estimates. Moreover, results suggest that all models, the multinomial logit and the hybrid choice models, successfully isolate the source of model bias, preventing its propagation to other parameter estimates. Furthermore, results indicate that parameter estimates from hybrid choice models are robust to modelling assumptions. Finally, results show that a simple multinomial logit provides higher out-of-sample prediction accuracy than the hybrid choice models, highlighting that better parameter estimates, do not always translate into better model predictions.

Place, publisher, year, pages
2018.
Keywords [en]
Hybrid Choice Models (HCM); Integrated Choice and Latent Variable models (ICLV); Mode choice; Latent variables; Model misspecification, Parameter bias, Synthetic dataset, Out-of-sample prediction
National Category
Transport Systems and Logistics Economics
Identifiers
URN: urn:nbn:se:kth:diva-240354OAI: oai:DiVA.org:kth-240354DiVA, id: diva2:1271358
Note

QC 20181217

Available from: 2018-12-17 Created: 2018-12-17 Last updated: 2018-12-18Bibliographically approved
In thesis
1. Learning about the unobservable: The role of attitudes, measurement errors, norms and perceptions in user behaviour.
Open this publication in new window or tab >>Learning about the unobservable: The role of attitudes, measurement errors, norms and perceptions in user behaviour.
2019 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Unobservable factors are important to understand user behaviour. Moreover, they contain information to help design services that willsolve today’s challenges. Yet, we have barely scratched the surface ofthe underlying mechanisms ruling user behaviour. For decades, userbehaviour analysis has focused on the capabilities of observable variables,as well as assumptions of regular preferences and rational behaviourto explain user choices; and amalgamated unobservable factorsinto ”black-box” variables. As a response, the field of behaviouraleconomics has produced an array of so-called choice anomalies, wherepeople seem not to be fully rational. Furthermore, as a consequence of the ”digital revolution”, nowwe harvest data on an unprecedented scale -both in quantity andresolution- that is nurturing the golden age of analytics. This explosionof analytics contributes to reveal fascinating patterns of humanbehaviour and shows that when users face difficult choices, predictionsbased only on observable variables result in wider gaps between observedand predicted behaviour, than predictions including observableand unobservable factors. Impacts of the ”digital revolution” are not limited to data and analyticsbut they have filtered through the whole tissue of society. Forinstance, telecommunications allow users to telework, and telework allowsusers to change their travel patterns, which in turn contributes toincrease the overall system complexity. In addition to the new worlddynamics facilitated by Information and Communications Technology,megatrends such as hyper-urbanization or increase demand of personalisedtransport services are imposing pressures on transport networksat a furious pace, which also contributes to increase the complexity ofthe choices needed in order to navigate the networks efficiently. In an effort to alleviate these pressures, new mobility services suchas electric and autonomous vehicles; bicycle and car sharing schemes;mobility as a service; vacuum rail systems or even flying cars are evolving. Each of these services entails a different set of observable variableslike travel time and cost, but also a completely different set of unobservableones such as expectations, normative beliefs or perceptionsthat will impact user behaviour. Hence, a good understanding of theimpact of underlying, unobservable, factors -especially when servicesare radically different from what users know and have experienced inthe past- will help us to predict user behaviour in uncharted scenarios. Unobservable factors are elusive by nature, hence to incorporatethem into our models is an arduous task. Furthermore, there is evidence showing that the importance of these factors might differ across time and space, as user preferences, perceptions, normative beliefs, etc.are influenced by local conditions and cultures. As a consequence, we have witnessed a surge of interest in behavioural economics over the past two decades, due to its ability to increase the explanatory and predictive power of models based on economic theory by adding a more psychologically plausible foundation. This thesis contributes to the existing body of literature in TransportScience in the areas of user perceptions, measurement errors, and the influence of attitudes and social norms in the adoption of new mobility solutions. The work builds on the behavioural economics theoretical framework, underpinned by economic theory, discrete choice analysis -rational behaviour and random utility maximization-, as well as social and cognitive psychology. Methodological contributions include a framework to systematically test differences in user preferences for a set of public transport modes, relating to observed and unobserved attributes; and a framework to assess the magnitude of unobservable measurement errors in the input variables of large-scale travel demand models. On an empirical dimension, findings support the existence of a ”rail factor”, the impact of modelling assumptions on parameter estimates of hybrid choice models, the presence of larger measurement errors in the cost variables than in the time variables, -which in turn translates into diluted parameters that under-estimate the response to pricing interventions-, and that the model with the best fit does not guarantee better parameter estimates. Therefore, I expect this thesis to be of interest not only to modellers, but also to decision makers; and that its findings will contribute to the design of the mobility solutions that users need and desire, but also that will benefit society as a whole.

Place, publisher, year, edition, pages
Stockholm: Kungliga Tekniska högskolan, 2019. p. 48
Series
TRITA-ABE-DLT ; 1837
Keywords
Attitudes, Measurement errors, Discrete choice analysis, Latent variables, Model misspecification, Normative beliefs, Rail factor, User perceptions, Social norms, Value of travel time savings
National Category
Transport Systems and Logistics Economics
Research subject
Transport Science
Identifiers
urn:nbn:se:kth:diva-240362 (URN)978-91-7873-022-3 (ISBN)
Public defence
2019-01-31, F3, Lindstedtsvägen 26, Stockholm, 14:00 (English)
Opponent
Supervisors
Note

QC 20181217

Available from: 2018-12-18 Created: 2018-12-17 Last updated: 2018-12-18Bibliographically approved

Open Access in DiVA

jmlv_MisspecifiedHybridChoiceModels(363 kB)15 downloads
File information
File name FULLTEXT01.pdfFile size 363 kBChecksum SHA-512
57fe5352af041af6a9c1e0b2fe73370796f6e4ff8e9204dbcd2a78cf0cc69e2fdee83cd5662546ae357df40ea2703cb6d8e19c05c4d0233eb61acfa911811fc0
Type fulltextMimetype application/pdf

Authority records BETA

Lorenzo Varela, Juan Manuel

Search in DiVA

By author/editor
Lorenzo Varela, Juan Manuel
By organisation
System Analysis and Economics
Transport Systems and LogisticsEconomics

Search outside of DiVA

GoogleGoogle Scholar
Total: 15 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 151 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf