Accounting for scale heterogeneity within and between pooled data sources

There is growing interest in incorporating both preference heterogeneity and scale heterogeneity in choice models, as a way of capturing an increasing number of sources of utility amongst a set of alternatives. The extension of mixed logit to incorporate scale heterogeneity in a generalised mixed logit (GMXL) model provides a way to accommodate these sources of influence, observed and unobserved. The small but growing number of applications of the GMXL model have parameterized scale heterogeneity as a single estimate; however it is often the case that analysts pool data from more than one source, be it revealed preference (RP) and stated preference (SP) sources, or multiple SP sources, inducing the potential for differences in the scale factor between the data sources. Existing practice has developed ways of accommodating scale differences between data sources by adopting a scale homogeneity assumption within each data source (e.g., the nested logit trick) that varies between data sources. This paper extends the state of the art by incorporating data-source specific scale differences in scale heterogeneity setting across pooled RP and SP data set. An example of choice amongst RP and SP transport modes (including two ‘new’ SP modes) is used to obtain values of travel time savings that vary significantly between a model that accounts for scale heterogeneity differences within pooled RP and SP data, and the other where differences in scale heterogeneity is also accommodated between RP and SP data.     

Mixed RP/SP models incorporating interaction effects

In order to analyse the impact of a new train service in Cagliari (Italy) a databank including information from a revealed preference (RP) and a stated preference (SP) survey was set up. The RP data concern choice between car, bus and train; the SP data consider the binary choice between a new train service (quicker, more frequent, with a lower fare and more stations than the current one) and the alternative currently chosen by car and bus users. Logit models allowing for correlation among RP alternatives were estimated for this mixed RP/SP data set using the artificial tree structure method. The analysis included level-of-service variables measured with an unusually high level of precision, latent or second order variables (such as comfort), inertia and interaction variables. Different specifications of the utility function were tested, including the expenditure rate model, and the effects of these specifications on modelling results are highlighted. Our results show that for a population mainly composed of fixed income workers, the expenditure rate model is superior to the traditional wage rate model, yielding lower and more significant subjective values of time. Moreover, we found that the non-linear specifications appear to be more suitable as not only better model results were obtained, but also the real distribution of the error terms was revealed (i.e. highlighting correlation among public transport options).     

Estimating the long-term effects of different passenger car technologies on energy/fuel consumption and emissions of greenhouse gases in Europe

Abstract

A stated preference (SP) experiment of car ownership was conducted in Mumbai Metropolitan Region (MMR) of Maharashtra in India. A full factorial experiment was designed to considering various attributes such as travel time, travel cost, projected household income, car loan payment and servicing cost. Data on 357 individuals were collected which resulted in 3213 observations for the calibration of the work trip and recreational trip car ownership models. The car ownership alternatives considered 0, 1 and 2 cars. A multinomial logit framework was used to develop the car ownership model taking the household as a decision unit. The specification and results of the SP car ownership model are discussed. The observed and predicted values matched reasonably when the validity of the SP car ownership model was tested against revealed preference (RP) data. The car ownership models developed in this study exhibit a satisfactory goodness of fit. It is concluded that the SP modelling approach can be successfully used for modelling car ownership decisions of households in developing countries.     

Correcting state dependence and serial correlation in the RP/SP combined estimation method

Revealed preference (RP) data and stated preference (SP) data have complementary characteristics for model estimation. To enhance the advantages of both data types, a combined estimation method is proposed. This paper discusses the method and practical considerations in applying it, and introduces a new method of considering serial correlation of RP and SP data. An empirical analysis is also presented.     

Door-to-door travel times in RP departure time choice models: An approximation method using GPS data

A common way to determine values of travel time and schedule delay is to estimate departure time choice models, using stated preference (SP) or revealed preference (RP) data. The latter are used less frequently, mainly because of the difficulties to collect the data required for the model estimation. One main requirement is knowledge of the (expected) travel times for both chosen and unchosen departure time alternatives. As the availability of such data is limited, most RP-based scheduling models only take into account travel times on trip segments rather than door-to-door travel times, or use very rough measures of door-to-door travel times. We show that ignoring the temporal and spatial variation of travel times, and, in particular, the correlation of travel times across links may lead to biased estimates of the value of time (VOT). To approximate door-to-door travel times for which no complete measurement is possible, we develop a method that relates travel times on links with continuous speed measurements to travel times on links where relatively infrequent GPS-based speed measurements are available. We use geographically weighted regression to estimate the location-specific relation between the speeds on these two types of links, which is then used for travel time prediction at different locations, days, and times of the day. This method is not only useful for the approximation of door-to-door travel times in departure time choice models, but is generally relevant for predicting travel times in situations where continuous speed measurements can be enriched with GPS data.     

An investigation on the performances of mode shift models in transit ridership forecasting

This paper aims at investigating the over-prediction of public transit ridership by traditional mode choice models estimated using revealed preference data. Five different types of models are estimated and analysed, namely a traditional Revealed Preference (RP) data-based mode choice model, a hybrid mode choice model with a latent variable, a Stated Preference (SP) data-based mode switching model, a joint RP/SP mode switching model, and a hybrid mode switching model with a latent variable. A comparison of the RP data-based mode choice model with the mode choice models including a latent variable showed that the inclusion of behavioural factors (especially habit formation) significantly improved the models. The SP data-based mode switching models elucidated the reasons why traditional models tend to over-predict transit ridership by revealing the role played by different transit level-of-service attributes and their relative importance to mode switching decisions. The results showed that traditional attributes (e.g. travel cost and time) are of lower importance to mode switching behaviour than behavioural factors (e.g. habit formation towards car driving) and other transit service design attributes (e.g. crowding level, number of transfers, and schedule delays). The findings of this study provide general guidelines for developing a variety of transit ridership forecasting models depending on the availability of data and the experience of the planner.     

Accounting for inertia in modal choices: some new evidence using a RP/SP dataset

Inertia is related with effect that experiences in previous periods may have on the current choice. In particular, it has to do with the tendency to stick with the past choice even when another alternative becomes more appealing. As new situations force individuals to rethink about their choices new preferences may be formed. Thus a learning process begins that relaxes the effect of inertia in the current choice. In this paper we use a mixed dataset of revealed preference (RP)-stated preference (SP) to study the effect of inertia between RP and SP observations and to study if the inertia effect is stable along the SP experiments. Inertia has been studied more extensively with panel datasets, but few investigations have used RP/SP datasets. In this paper we extend previous work in several ways. We test and compare several ways of measuring inertia, including measures that have been proposed for both short and long RP panel datasets. We also explore new measures of inertia to test for the effect of “learning” (in the sense of acquiring experience or getting more familiar with) along the SP experiment and we disentangle this effect from the pure inertia effect. A mixed logit model is used that allows us to account for both systematic and random taste variations in the inertia effect and for correlations among RP and SP observations. Finally we explore the relation between the utility specification (especially in the SP dataset) and the role of inertia in explaining current choices.     

Analysing Demand for Suburban Trips: A Mixed RP/SP Model with Latent Variables and Interaction Effects

We analyse mode choice behaviour for suburban trips in the Grand Canary island using mixed revealed preference (RP)/stated preference (SP) information. The SP choice experiment allowed for interactions among the main policy variables: travel cost, travel time and frequency, and also to test the influence of latent variables such as comfort. It also led to discuss additional requirements on the size and sign of the estimated model parameters, to assess model quality when interactions are present. The RP survey produced data on actual trip behaviour and was used to adapt the SP choice experiment. During the specification searches we detected the presence of income effect and were able to derive willingness-to-pay measures, such as the subjective value of time, which varied among individuals. We also studied the systematic heterogeneity in individual tastes through the specification of models allowing for interactions between level-of-service and socio-economic variables. We concluded examining the sensitivity of travellers’ behaviour to various policy scenarios. In particular, it seems that contrary to political opinion, in a crowded island policies penalising the use of the private car seem to have a far greater impact in terms of bus patronage than policies implying direct improvements to the public transport service.     

Estimation of switching models from revealed preferences and stated intentions

This paper proposes a methodology for modeling switching behavior using simultaneously cross-sectional revealed preference (RP) data and stated intentions, a type of stated preference (SP) data. With explicit consideration of biases and random errors potentially contained in SP data, combined estimation with RP and SP data can exploit the advantages of both data sources. An empirical analysis of commuters' mode choice shows that the stated intention data have predictive validity if their biases and errors are properly corrected.     

A sequential approach to exploiting the combined strengths of SP and RP data: Application to freight shipper choice

The possibility of and procedure for pooling RP and SP data have been discussed in recent research work. In that literature, the RP data has been viewed as the yardstick against which the SP data must be compared. In this paper we take a fresh look at the two data types. Based on the peculiar strengths and weaknesses of each we propose a new, sequential approach to exploiting the strengths and avoiding the weaknesses of each data source. This approach is based on the premise that SP data, characterized by a well-conditioned design matrix and a less constrained decision environment than the real world, is able to capture respondents' tradeoffs more robustly than is possible in RP data. (This, in turn, results in more robust estimates of share changes due to changes in independent variables.) The RP data, however, represent the current market situation better than the SP data, hence should be used to establish the aggregate equilibrium level represented by the final model. The approachfixes the RP parameters for independent variables at the estimated SP parameters but uses the RP data to establish alternative-specific constants. Simultaneously, the RP data are rescaled to correct for error-in-variables problems in the RP design matrixvis-à- vis the SP design matrix. All specifications tested are Multinomial Logit (MNL) models.The approach is tested with freight shippers' choice of carrier in three major North American cities. It is shown that the proposed sequential approach to using SP and RP data has the same or better predictive power as the model calibrated solely on the RP data (which is the best possible model for that data, in terms of goodness-of-fit figures of merit), when measured in terms of Pearson's Chi-squared ratio and the percent correctly predicted statistic. The sequential approach is also shown to produce predictions with lower error than produced by the more usual method of pooling the RP and SP data.     

Novel approaches to forecasting the demand for new local rail services

Revealed Preference(RP) studies based on actual behaviour suffer from a number of statistical problems. Furthermore, RP methods are of little use when the effects of a new or radically altered service need to be considered. As a result, for a case study of demand forecasting for new passenger rail services, a new approach has been developed. Our starting point is to seek what we call Stated Intentions (SI) responses as to the likely usage of a new rail service. However, due to a combination of systematic biases, these responses may be taken to be gross overestimates. A check on the biases of this SI data may be supplied by a Stated Preference (SP) survey. Respondents are asked to make hypothetical choices which are sufficiently complex for there to be little chance of policy bias. It is ensured that choices presented contain useful ‘boundary values,’ being the relative valuation for which respondents would be indifferent between two offered alternatives. It is, however, crucial to ensure that the SP survey is simple enough for respondents to manage, since excessive error variability in the responses will cause the calibrated coefficients to be rescaled, presenting problems for forecasting. From the SP surveys, it is estimated that SI data overstates usage of new rail services by around 50%, even if it is assumed that nonrespondents to the SI survey are nonusers. It is concluded that an SI/SP approach can potentially provide accurate forecasts, but there are a number of practical constraints that may prevent this.     

Modelling travellers’ risky choice in a revealed preference context: a comparison of EUT and non-EUT approaches

Recent work on risky choice modelling has sought to address the shortcomings of expected utility theory (EUT) by using non-expected utility theoretic (non-EUT) approaches. However, to date these approaches have been merely tested on stated choice data which is flexible and cheap. In this study, we empirically investigate the feasibility and validity of non-EUT approaches in a revealed preference (RP) context in which travel time distribution is extracted from observed historical travel time data, and subsequently present systematic comparisons between EUT, weighted utility theory, rank-dependent expected utility theory, and prospect theory (PT). The empirical evidence indicates that each non-EUT model has important behavioural insights to offer, moreover, EUT as well as non-EUT models can be applied to the RP context. However, the EUT and non-EUT model fits are generally similar with only PT providing a marginally improved model fit over EUT. The key findings presented in this study reinforce the importance of exploring non-EUT models within a revealed preference context before they can be applied reliably to modelling risky choices in the real world.     

Stated preference analysis of travel choices: the state of practice

Stated preference (SP) methods are widely used in travel behaviour research and practice to identify behavioural responses to choice situations which are not revealed in the market, and where the attribute levels offered by existing choices are modified to such an extent that the reliability of revealed preference models as predictors of response is brought into question. This paper reviews recent developments in the application of SP models which add to their growing relevance in demand modelling and prediction. The main themes addressed include a comparative assessment of choice models and preference models, the importance of scaling when pooling different types of data, especially the appeal of SP data as an enriching strategy in the context of revealed preference models, hierarchical designs when the number of attributes make single experiments too complex for the respondent, and ways of accommodating dynamics (i.e. serial correlation and state dependence) in SP modelling.An earlier modified version was presented as the keynote address to the 1993 National Conference on Tourism Research, held at the University of Sydney, 19 March 1993. The comments of Jordan Louviere, Lester Johnson, Paul Hooper, W.G. Waters II and Mark Bradley are appreciated.     

Commute responses to employment decentralization: Anticipated versus actual mode choice behaviors of new town employees in Kunming,China

This study examines workers’ mode-choice responses to a typical job decentralization policy implemented in China’s urban development – government job relocation (GJR) to new towns in the urban periphery. Broadly, the literature suggests that job decentralization tends to increase car commuting; however, little is known about the effects of China’s GJR initiatives on individuals’ commuting mode choices. Using Kunming as a case study, this study examines how workers’ commuting mode choices have shifted in response to the GJR policy. Our study analyzes two travel survey datasets that span the job relocation process: (1) stated preference (SP) data on workers’ anticipated mode choices after a move of workplace to a planned new town; and (2) revealed preference (RP) data on workers’ actual choices of commuting mode after their jobs were moved. The findings suggest that after job relocation, workers’ actual commuting modes shift from more sustainable modes towards cars. The determinants of workers’ mode choices differ substantially between the hypothetical and actual setting of job relocation. The anticipated mode choices are largely determined by socio-demographic characteristics whereas the actual mode choices are strongly influenced by travel time and housing locations. The evidence from this study offers two important implications for future planning practice of job decentralization. First, planners and policy-makers should be skeptical about the transportation benefits of job decentralization. Second, while SP surveys can assist planners to predict individuals’ mode-choice responses, the robustness of SP results should be carefully assessed before translating into the evidence base for informing job decentralization policy-makings.     

On the semantic scale problem in stated preference rating experiments

Stated preference (SP) rating experiments are easy to design and conduct. Furthermore, they constitute a generalisation of SP choice experiments in terms of the information about preferences that can be achieved. However, its results are sensitive to the numerical values assigned by the analyst to the semantic scale used by individuals to express their preferences. We consider this problem in depth using a variety of statistical techniques, including ordinal probit and a novel optimal scale linear regression approach.     

Assessing systematic sources of variation in public transport elasticities: Some comparative warnings

There is an extensive and continually growing body of empirical evidence on the sensitivity of potential and actual users of public transport to fare and service levels. The sources of the evidence are disparate in terms of methods, data collection strategy, data paradigms, trip purpose, location, time period, and attribute definition. In this paper, we draw on a data set we have been compiling since 2003 that contains over 1100 elasticity items associated with prices and services of public transport, and car modes. The focus herein is on direct elasticities associated with public transport choice and demand, and the systematic sources of influence on the variations in the mean estimates for fares, in-vehicle time, and headway obtained from 319 studies. The major influences on variations in mean estimates of public transport elasticities are the time of day (peak, all day vs. off-peak), the data paradigm (especially combined SP/RP vs. revealed preference (RP)), whether an average fare or class of tickets is included, the unit of analysis (trips vs. vkm), specific trip purposes, country, and specific-mode (i.e., bus and train) in contrast to the generic class of public transport.     

Estimating GEV models with censored data

We examine the problem of estimating parameters for Generalized Extreme Value (GEV) models when one or more alternatives are censored in the sample data, i.e., all decision makers who choose these censored alternatives are excluded from the sample; however, information about the censored alternatives is still available. This problem is common in marketing and revenue management applications, and is essentially an extreme form of choice-based sampling. We review estimators typically used with GEV models, describe why many of these estimators cannot be used for these censored samples, and present two approaches that can be used to estimate parameters associated with censored alternatives. We detail necessary conditions for the identification of parameters associated exclusively with the utility of censored alternatives. These conditions are derived for single-level nested logit, multi-level nested logit and cross-nested logit models. One of the more surprising results shows that alternative specific constants for multiple censored alternatives that belong to the same nest can still be separately identified in nested logit models. Empirical examples based on simulated datasets demonstrate the large-sample consistency of estimators and provide insights into data requirements needed to estimate these models for finite samples.     

Contrasts between utility maximisation and regret minimisation in the presence of opt out alternatives

An increasing number of studies of choice behaviour are looking at Random Regret Minimisation (RRM) as an alternative to the well established Random Utility Maximisation (RUM) framework. Empirical evidence tends to show small differences in performance between the two approaches, with the implied preference between the models being dataset specific. In the present paper, we discuss how in the context of choice tasks involving an opt out alternative, the differences are potentially more clear cut. Specifically, we hypothesise that when opt out alternatives are framed as a rejection of all the available alternatives, this is likely to have a detrimental impact on the performance of RRM, while the performance of RUM suffers more than RRM when the opt out is framed as a respondent being indifferent between the alternatives on offer. We provide empirical support for these hypotheses through two case studies, using the two different types of opt out alternatives. Our findings suggest that analysts need to carefully evaluate their choice of model structure in the presence of opt out alternatives, while any a priori preference for a given model structure should be taken into account in survey framing.     

Park-and-ride lot choice model using random utility maximization and random regret minimization

This research aims to understand the park-and-ride (PNR) lot choice behaviour of users i.e., why PNR user choose one PNR lot versus another. Multinomial logit models are developed, the first based on the random utility maximization (RUM) concept where users are assumed to choose alternatives that have maximum utility, and the second based on the random regret minimization (RRM) concept where users are assumed to make decisions such that they minimize the regret in comparison to other foregone alternatives. A PNR trip is completed in two networks, the auto network and the transit network. The travel time of users for both the auto network and the transit network are used to create variables in the model. For the auto network, travel time is obtained using information from the strategic transport network using EMME/4 software, whereas travel time for the transit network is calculated using Google’s general transit feed specification data using a backward time-dependent shortest path algorithm. The involvement of two different networks in a PNR trip causes a trade-off relation within the PNR lot choice mechanism, and it is anticipated that an RRM model that captures this compromise effect may outperform typical RUM models. We use two forms of RRM models; the classical RRM and µRRM. Our results not only confirm a decade-old understanding that the RRM model may be an alternative concept to model transport choices, but also strengthen this understanding by exploring differences between two models in terms of model fit and out-of-sample predictive abilities. Further, our work is one of the few that estimates an RRM model on revealed preference data.

     

An application of a rank ordered probit modeling approach to understanding level of interest in autonomous vehicles

Surveys of behavior could benefit from information about people’s relative ranking of choice alternatives. Rank ordered data are often collected in stated preference surveys where respondents are asked to rank hypothetical alternatives (rather than choose a single alternative) to better understand their relative preferences. Despite the widespread interest in collecting data on and modeling people’s preferences for choice alternatives, rank-ordered data are rarely collected in travel surveys and very little progress has been made in the ability to rigorously model such data and obtain reliable parameter estimates. This paper presents a rank ordered probit modeling approach that overcomes limitations associated with prior approaches in analyzing rank ordered data. The efficacy of the rank ordered probit modeling methodology is demonstrated through an application of the model to understand preferences for alternative configurations of autonomous vehicles (AV) using the 2015 Puget Sound Regional Travel Study survey data set. The methodology offers behaviorally intuitive model results with a variety of socio-economic and demographic characteristics, including age, gender, household income, education, employment and household structure, significantly influencing preference for alternative configurations of AV adoption, ownership, and shared usage. The ability to estimate rank ordered probit models offers a pathway for better utilizing rank ordered data to understand preferences and recognize that choices may not be absolute in many instances.