Random utility models often involve terms which represent alternative-specific errors, and the main attractive feature of the multinomial probit (MNP) model is that it allows a rather general covariance structure for these errors. However, since observed choices only reveal information regarding utility differences, and since scale cannot be determined, not all parameters in an arbitrary MNP specification may be identified. This paper examines identification restrictions that arise in the linear-in-parameters multinomial probit framework, and provides discussion and recommendations for estimation and analysis of probit normalizations.

In this paper we revisit various important issues relating to practical estimation of the multinomial probit model, using an empirical analysis of car ownership as a test case. To provide context, a brief literature review of empirical probit studies is included. Estimates are obtained for a full range of model specifications, including models with random (uncorrelated and correlated) taste variation and/or a general random error structure, and issues of estimability, specification testing, and alternative normalizations for probit models are addressed. Three model trust region algorithms for finding maximum likelihood estimates are compared, and the superiority of a structured quasi-Newton method employing "model switching" over more traditional approaches (Broyden-Fletcher-Goldfarb-Shanno secant update, Berndt-Hall-Hall-Hausman) is demonstrated. The trust region algorithms have reliable convergence properties and provide useful diagnostic information. Finally, a comparison of some probit integral approximation schemes (Clark, and two variants of Mendell-Elston) versus numerical integration is included. There is additional evidence against using Clark's approximation, but a variant of the Mendell-Elston approach appears promising. Numerical problems with variable-ordering techniques (such as separated-split) are demonstrated.

Dynamic, disaggregate choice models which use longitudinal data are known to have clear advantages over cross-sectional models, but they also have their own unique estimation problems. The correlation among unobserved error components ("heterogeneity") that is likely to exist in such data sets can be the source of apparent state dependence, but true state dependence is also possible. A review of car ownership models reveals that the issue of heterogeneity versus true state dependence has not been adequately addressed in the transportation literature. This paper develops computationally convenient ordered-response probit models for panel data, estimates models of car ownership, and performs tests of heterogeneity versus true state dependence. Conclusions in (the more general) one-factor models are found to differ from those obtained from (the more restricted) components of variance models, and the issue of initial conditions is also found to affect the conclusions.

We compare multinomial logit and mixed logit models for data on California households' revealed and stated preferences for automobiles. The stated preference (SP) data elicited households' preferences among gasoline, electric, methanol, and compressed natural gas vehicles with various attributes. The mixed logit models provide improved fits over logit that are highly significant, and show large heterogeneity in respondents' preferences for alternative-fuel vehicles. The effects of including this heterogeneity are demonstrated in forecasting exercises. The alternative-fuel vehicle models presented here also highlight the advantages of merging SP and revealed preference (RP) data. RP data appear to be critical for obtaining realistic body-type choice and scaling information, but they are plagued by multicollinearity and difficulties with measuring vehicle attributes. SP data are critical for obtaining information about attributes not available in the marketplace, but pure SP models with these data give implausible forecasts.

We compare multinomial logit and mixed logit models for data on California households' revealed and stated preferences for automobiles. The stated preference (SP) data elicited households' preferences among gasoline, electric, methanol, and compressed natural gas vehicles with various attributes. The mixed logit models provide improved fits over logit that are highly significant, and show large heterogeneity in respondents' preferences for alternative-fuel vehicles. The effects of including this heterogeneity are demonstrated in forecasting exercises. The alternative-fuel vehicle models presented here also highlight the advantages of merging SP and revealed preference (RP) data. RP data appear to be critical for obtaining realistic body-type choice and scaling information, but they are plagued by multicollinearity and difficulties with measuring vehicle attributes. SP data are critical for obtaining information about attributes not available in the marketplace, but pure SP models with these data give implausible forecasts.

The increasing diversity of vehicle type holdings and growing demand for BEVs and PHEVs have serious policy implications for travel demand and air pollution. Consequently, it is important to accurately predict or estimate the preference for vehicle holdings of households as well as the vehicle miles traveled by vehicle body- and fuel-type to project future VMT changes and mobile source emission levels. Leveraging the 2019 California Vehicle Survey data, this report presents the application of a utility-based model for multiple discreteness that combines multiple vehicle types with usage in an integrated model, specifically the MDCEV model. The model results suggest the important effects of household demographics, residence location, and built environment factors on vehicle body type and powertrain choice and usage. Further the predictions associated with changes inbuilt environment factors like population density can inform the design of land-use and transportation policies to influence household vehicle holdings and usage that can in turn impact travel demand and air quality issues in California.

View the NCST Project Webpage

A discrete choice model has been developed in which the choice alternative consist of vehicle transactions rather than portfolios of vehicle holdings. The model is based on responses to customized stated preference questions involving both hypothetical future vehicles and the household's current vehicle holdings. The stated choices were collected from 4747 survey respondents located throughout most of the urbanized portions of the state of California. Respondents were asked what their next vehicle transaction would most likely be (replace a current vehicle, add another vehicle, or dispose of a current vehicle), and respondents who wanted to replace or add vehicles were asked to indicate their most preferred vehicle from a set of six hypothetical vehicles. The hypothetical vehicles were described in terms of fourteen attributes, manipulated according to an experimental design. 

The transactions model is a multinomial logit model of the choice of the hypothetical vehicles and whether or not the hypothetical vehicle will be a replacement or addition to the household fleet. The model is conditioned on the household's current vehicle stock. and the characteristics of the current vehicles are important explanators of the stated preference choices. In addition to the model estimates, forecasts are given for a base case scenario in 1998. 

This model is one component in a micro-simulation demand forecasting system being designed to produce annual forecasts of new and used vehicle demand by type of vehicle and geographic area. The system will also forecast annual vehicle miles traveled for all vehicles and recharging demand by time of day for electric vehicles. These results are potentially useful to utility companies in their demand-side management planning, to public agencies in their evaluation incentive schemes, and to manufacturers faced with designing and marketing alternative-fuel vehicles. 

To mitigate climate change and air pollution, multiple US states and other countries have beensetting and adjusting goals and policies aimed at shifting sales from conventional, fossil-fuel–powered vehicles to plug-in electric vehicles (PEVs), defined as plug-in hybrid and battery electric (all-electric) vehicles. For example, US policies have offered federal tax credits for the purchase of PEVs, with limits set on how many PEVs from a single manufacturer, which PEVs, and which consumers qualify. A key to developing or adjusting these policies is understanding how financial incentives affect consumers’ decisions to purchase or lease PEVs. To better understand the impact of financial incentives on PEV leasing and purchasing, researchers at the University of California, Davis, analyzed survey responses from approximately 2,800 California PEV owners. The survey asked: If the federal tax credit were not available would you: purchase or lease the same PEV, switch to a different PEV, switch to a conventional or hybrid (non-plug in) vehicle, or not acquire a vehicle at all? This policy brief discusses findings from those survey responses and provides policy implications.

View the NCST Project Webpage

This paper describes an ongoing project to develop a demand forecasting model for clean-fuel vehicles in California. Large-scale surveys of both households and commercial fleet operators have been carried out. These data are being used to calibrate a new micro-simulation based vehicle demand forecasting system. Based on pre-specified attributes of future vehicles (including specified clean-fueled vehicle incentives), the system will produce annual forecasts of new and used vehicle demand by type of vehicle and geographic region. The system will also forecast annual vehicle miles traveled for all vehicles and recharging demand by time of day for electric vehicles. These results are potentially useful to utility companies in their demand-side management planning, to public agencies in their evaluation incentive schemes, and to manufacturers faced with designing and marketing clean-fuel vehicles.

This research describes a new model of household vehicle use behavior by type of vehicle. Forecasts of future vehicle emissions, including potential gains that might be attributed to introductions of alternative-fuel (clean-fuel) vehicles, critically depend upon the ability to forecast vehicle-miles travelled by the fuel type, body style and size, and vintage of vehicle.