Current Developments in Mobility Survey Methods 1 |
|
| Session Organisers | Professor Caroline Bayart (University Lyon 1) Dr Johannes Eggs (Infas - Institut für angewandte Sozialwissenschaft GmbH) Mrs Dana Gruschwitz (Infas - Institut für angewandte Sozialwissenschaft GmbH) |
| Time | Tuesday 16th July, 11:00 - 12:30 |
| Room | D31 |
Travel surveys collect data on the mobility of populations. Large-scale national household transport surveys are used to estimate key mobility figures on national or sub-national levels, like out of home rates, trips and transport mode rates, and to predict the demand of the population regarding the use of transportation modes.
Transport surveys face with a drop of response rates over the world. Even if weighting procedures allow to reduce the incidence of non-response, it is not always possible to postulate that people with some socio-demographic characteristics who do not respond to a survey have the same behaviour than people with the same socio-demographic characteristics who respond and survey non-response might produce bias.
Travel surveys have also some peculiarities, like the focus on trips’ collection for specific days only. They also need a higher number of respondents in comparison to “regular” social surveys, to reach an adequate precision to model and predict travel and transport demand on a regional level. Efforts are made to increase response rate for traditional transport surveys by improving the questionnaire, reducing respondent burden, increasing reminders… Even if results are generally positive, it is in most cases not sufficient. Moreover, implementation of travel surveys is relatively expensive.
New forms of mobility, like car - or bike - sharing that allow passive forms of data collection methods for this specific subgroup, mixed-modes surveys and incorporation of “Big data” are getting increased attention to lower transport survey costs. The potential of new interactive media and Big data seems to be high to improve transport surveys. But the question of data comparability remains. The danger when databases are merged is that a sample selection bias will be created and compromise the accuracy of explanatory models.
The aims of the session will be to discuss the potential of new technologies for mixed modes framework and the opportunity to combine transport survey results with other data sources. It will be possible to characterize bias generated by these methods and to give some perspectives for reduce them. Topics for this session may include (but are not restricted to):
• Survey mode effects in travel survey
• Passive data collection in travel surveys
• Integrating GPS data in trip and survey data
• Improving trip reporting in different modes
• Effects of proxy interviews
• Combining mobility data sources
• Improving household completion rates
• Methods to collect mobility data for specific sociodemographic groups (children)
Keywords: Mixed-modes surveys, combining survey data, mode effects, data comparability, big data
Professor Caroline Bayart (University Lyon 1) - Presenting Author
Professor Patrick Bonnel (ENTPE - University Lyon 2)
Dr Christelle Paulo (Ile-de-France Mobilités)
Mrs Anne-Eole Meret-Conti (Ile-de-France Mobilités)
Transport constitutes a strategic tool for urban policies and it is crucial to construct reliable survey protocols to obtain representative travel behavior data at reasonable cost. Unfortunately, household travel surveys response rates are decreasing. To reduce this bias of non-response, a web survey will be performed in parallel of the 2020 travel survey conducted in face-to-face in Paris area (France). The idea is to select two random samples and propose to the first one to answer the survey in face-to-face and to the second one to answer by web. To test this new protocol, a large test has been run. To date, around 650 persons have been interviewed at home and 560 individuals give complete and verified answers to the web survey.
The generalization of the web survey results to the whole population might be problematic due to possible bias in terms of representativeness and survey response in comparison with face-to-face survey. The sample is no longer random and the presence of respondents is determined by external factors which might be not identical for both media and which may also affect the variable of interest in the studied model. It is highly likely that the socioeconomic characteristics and the travel behaviors of the individuals who respond using the Internet are different from those of the individuals who respond to a face-to-face interview. The danger when databases are merged is that a sample selection bias will be created.
The paper initially discusses web potential for travel surveys, especially in a mixed modes framework. Then, it presents the results of the test conducted in Paris area. If a selection bias is characterized, it will be corrected by using endogenous switching regression models. These results will be compared to the previous travel survey conducted only in face-to-face in Paris conurbation in 2010.
Mrs Ecke Lisa (Karlsruhe Institute of Technology) - Presenting Author
Mr Hilgert Tim (Karlsruhe Institute of Technology)
Dr Chlond Bastian (Karlsruhe Institute of Technology)
Professor Vortisch Peter (Karlsruhe Institute of Technology)
The German Mobility Panel (MOP) is designed as a rotating panel over three consecutive years. People report their everyday-mobility by filling in a trip diary over a period of seven consecutive days. The multi-day and multi-period design has been nearly unaltered since 1994. The longitudinal panel approach allows the research of the dynamics and variations in travel behavior. However, the survey design results in a high respondent burden, thus endangering the quality and completeness of the collected data. Results are affected by attrition effects within one week (declining completeness of the individual report or discontinue the report completely) and reporting errors.
To ensure the necessary data quality the following approach is applied: We identify attrition effects within one week using regression models to check and evaluate whether a significant decrease in the mobility key figures over the course of the reporting week can be identified on a collective level. The characteristics of the MOP allow for the identification of those participants, who are causing these significant declines (e.g. trip-making against the background of the socio-economic characteristics). This allows for an active exclusion of these bad risks for the data quality. Moreover, the reported data of each participant is subject to an individual plausibility check. For this purpose, we developed a visualization tool, which graphically displays the mobility behavior of each survey participants at a glance for the plausibility check process. Thus irregularities and minor faults can easily be identified and corrected. Experienced and trained staff can easily detect inplausibilities, using this visualization tool. This allows for an immediate check of the outcome. Persons with blatant deficiencies in their report can be excluded: The procedures result in a high consistency and quality of the data for further use. All checks together ensure the necessary continuity and high data quality of the MOP-data.
Ms Janice Machado (Westat) - Presenting Author
Dr Marcelo Simas (Westat)
Mr Anthony Fucci (Westat)
Mr Alexander Cates (Westat)
Mr Shawn McCloskey (Westat)
Mr Jeremy Wilhelm (Westat)
Ms Laura Wilson (Westat)
The National Household Travel Survey (NHTS) provides an inventory of daily travel in the United States (US). It is the United States’ only source of national-level statistics on personal travel. The survey series (conducted since 1969) includes demographic data on households, persons, vehicles, and detailed information on daily travel by all modes of transportation for all purposes. Beginning in 1990, the NHTS was completed by telephone using a land-line random-digit-dialing sample. The US Department of Transportation contracted with Westat to conduct the 2017 NHTS. The 2017 Survey used an address-based sample and implemented a new methodology method where households were first recruited using a paper screener. Travel data was then collected on a second retrieval stage from all household members using both the web (70%) and telephone (30%) modes. This presentation presents an analysis of how the two travel reporting modes (phone and web) impacted survey response and travel patterns. The analysis examines the socio-demographic characteristics of households and persons based on the mode they used to complete travel retrieval as well as their travel patterns in terms of trip and tour rates.
Dr Marcelo Simas (Westat)
Dr Jill DeMatteis (Westat)
Ms Shelley Roth (Westat) - Presenting Author
Calibration methods such as poststratification, raking, and generalized regression estimation are widely used in complex sample surveys to improve the precision of estimates of totals, adjust for differential nonresponse and population coverage, and align survey estimates with external population parameters or estimates for face validity. Typically, household travel survey weights are developed for four types of analytic units: households, persons, trips, and vehicles. In the first and second stages, household and person weights are developed to represent the population of households and persons in the study area. At these stages, the weights are generally calibrated separately to external totals of households and persons. This approach of separately calibrating the household- and person-level weights to separate sets of totals is common practice and statistically valid; however, it creates unavoidable conflicts when using the household data to establish calibration targets for an Activity Based Model (ABM). The conflicts arise because the sub-models in an ABM can be household-based, person-based, and tour-based. A consistent set of totals is required to build targets for all sub-models, and this is not possible if the household weight applied to persons sum to different person-level totals. As a result of these inconsistencies, survey weights often are not used in ABMs used for the analysis of travel survey data. To overcome this difficulty, we compare two different alternatives—a dual-level approach based on iterative proportional fitting (IPF), the algorithm used in standard raking applications; and the iterative proportional updating (IPU) method developed by researchers at Arizona State University (Ye et al., 2009). The issues and methods presented in this research in the context of household travel surveys are more generally applicable to other situations in which it is necessary or desirable for estimated totals based on survey weights at two