David Brownstone

Hybrid Choice Models: Progress and Challenges

We discuss the development of predictive choice models that go beyond the random utility model in its narrowest formulation. Such approaches incorporate several elements of cognitive process that have been identified as important to the choice process, including strong dependence on history and context, perception formation, and latent constraints. A flexible and practical hybrid choice model is presented that integrates many types of discrete choice modeling methods, draws on different types of data, and allows for flexible disturbances and explicit modeling of latent psychological explanatory variables, heterogeneity, and latent segmentation. Both progress and challenges related to the development of the hybrid choice model are presented.

The impact of residential density on vehicle usage and energy consumption

The debate concerning the impacts of urban land use density on travel in general, and on residential vehicle use and fuel consumption in particular, lacks reliable quantitative evidence. The 2001 U.S. National Household Transportation Survey (NHTS) provides information on vehicle miles of travel (VMT) based on odometer data, as well as annual fuel usage computations based on information about the make, model and vintage of all household vehicles. In addition, the 2001 NHTS has been augmented with land use variables in the form of densities of population and residences at the census tract and block level for each of the more than 69,000 households in the dataset. In order to obtain unbiased estimates of the effects of any of these land use variables on annual VMT and fuel consumption the authors present a model system that accounts for both self selection effects and missing data that are related to the endogenous variables. Results for the State of California show that the residential density effects are substantial and precisely estimated. Comparing two households that are similar in all respects except residential density, a lower density of 1,000 housing units per square mile implies a positive difference of almost 1,200 miles per year and about 65 more gallons of fuel per household. This total effect of residential density on fuel usage is decomposed into to two paths of influence. Increased mileage leads to a difference of 45 gallons, but there is an additional direct effect of density through lower fleet fuel economy of 20 gallons per year, a result of vehicle type choice.

A transactions choice model for forecasting demand for alternative-fuel vehicles

The vehicle choice model developed here is one component in a micro-simulation demand forecasting system being designed to produce annual forecasts of new and used vehicle demand by vehicle type and geographic area in California. The system will also forecast annual vehicle miles traveled for all vehicles and recharging demand by time of day for electric vehicles. The choice model specification differs from past studies by directly modeling vehicle transactions rather than vehicle holdings. The model is calibrated using stated preference data from a new study of 4,747 urban California households. These results are potentially useful to public transportation and energy agencies in their evaluation of alternatives to current gasoline-powered vehicles. The findings are also useful to manufacturers faced with designing and marketing alternative-fuel vehicles as well as to utility companies who need to develop long-run demand-side management planning strategies.

Efficient Estimation of Nested Logit models

This paper examines the Sequential, Full Information Maximum Likelihood (FIML), and Linearized Maximum Likelihood (LML) estimators for a Nested Logit model of time-of-day choice for work trips. These estimators are compared using a Monte Carlo study based on specification and data from a previously published empirical study. The sequential estimator is found to be much less efficient than either LML or FIML; and its uncorrected second-stage standard-error estimates are strongly downward biased. LML is only slightly less efficient than FIML, but is often easier to compute. However there are cases where the sequential and LML estimators do not exist.

Zoning, Returns to Scale, and the Value of Undeveloped Land

When land markets are incomplete, parcels can be scaled to make control compatible with use and to internalize externalities. The authors show that an arbitrage-proof equilibrium implies an increasing and strictly concave relationship between the value and size of land parcels. Undeveloped land sales in southern California strongly confirm the theoretical relationship. The authors find that zoning primarily restricts the conversion of land from agriculture to residential and industrial uses relative to the competitive equilibrium. The scale of land units is an effective private instrument for providing compatible land use even in the presence of strong zoning. Copyright 1991 by MIT Press.

Modeling Earnings Measurement Error: A Multiple Imputation Approach

Recent survey validation studies suggest that measurement error in earnings data is pervasive and violates classical measurement error assumptions, and therefore may bias estimation of cross-section and longitudinal earnings models. We model the structure of earnings measurement error using data from the Panel Study of Income Dynamics Validation Study (PSDIVS). We then use Rubins (1987) multiple imputation techniques to estimate consistent earnings equations under non-classical earnings measurement error in the PSID. Our technique is readily generalized, and the empirical results demonstrate the potential importance of correcting for measurement error in earnings and related data, particularly during recessions.

Bootstrapping improved estimators for linear regression models

Abstract Since Stein (1955) proved that the least-squares ( LS ) estimator for the linear-regression model is inadmissible and James and Stein (1961) demonstrated a superior estimator under squared-error loss, a number of better estimators have been proposed. One reason these estimators have not been used in empirical work is that it is difficult to derive their sampling distributions. This study shows that nonparametric bootstrapping provides good estimates of the sampling distribution of Mundlaks (1981) restricted principal-components estimator and a new Stein-rule estimator which shrinks Mundlaks towards LS . A Monte Carlo study of the bootstrap is performed for a range of increasingly collinear designs with 100 observations and 10 regressors. Using a squared-error loss function, the Stein-rule estimator performed much better than LS over most of the designs. The nonparametric bootstrap generated reasonable estimates of the estimators risk as well as standard errors of the individual components.

The demand for housing in Sweden: Equilibrium choice of tenure and type of dwelling

Abstract Most studies of housing demand and tenure choice only identify two modes of tenure: owner-occupied one-family houses and rental apartments. Furthermore they are typically based on a cross section across all households. In this study we use recent Swedish data to overcome these weaknesses. We identify owner-occupied apartments (coop shares) as a third mode of tenure, and show that this should be treated separately. We also use information about the households own assessment of its probability of moving during the next year. We demonstrate that it makes a large difference if likely movers are eliminated from the sample.

Bootstrap confidence bands for shrinkage estimators

Empirical application of shrinkage estimators has been limited by the inability to estimate confidence intervals. We investigate confidence band estimators based on Edgeworth expansions and bootstrapping. Our Monte Carlo results suggest that both Efrons bias correction method with acceleration and the simple percentile bootstrap methods generate reasonable confidence bands. Approximations based on Edgeworth expansions performed poorly. We then use the percentile method with a single bootstrap to generate bands for predictions of GNP growth rates from a leading-indicators model. Our study shows that simple percentile bootstrap confidence bands perform well enough to support empirical applications of shrinkage estimators.

A microsimulation model of Swedish housing demand

Abstract This paper presents a microsimulation model of housing demand aimed at analyzing effects of changes in tax schedules. Like many other countries, Swedens income tax laws treat homeowners and renters asymmetrically. Therefore most changes in tax laws will affect housing demand and tenure choice. We develop a microsimulation model which generates consistent estimates of the mean and variance of total housing demand and show that commonly used measures like analytic elasticities evaluated at mean values of the exogenous variables can yield very misleading predictions. Our model can be used to simulate almost any change in the tax system; here we report results of three simple experiments, one of which is a change in the rate at which imputed income from owner-occupied housing is taxed.