International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
Feature Trade-offs in Choices among Alternative Vehicle Powertrains
Thomas J Adler, Jeff Keller, William C Neafsey, Daniel Weinstein

Last modified: 11 May 2015

Abstract


Consumers currently have a choice among several different types of vehicle powertrains and fuels: vehicles with conventional gasoline or diesel engines, hybrid electric powertrains, plug-in electric vehicles and battery electric vehicles. Choices among these alternatives involve trade-offs among a number of features that vary significantly across them.

A U.S.-based auto manufacturer undertook a project to determine how consumers make these trade-offs currently and under different future fuel prices scenarios. The specific focus of the work was on the trade-offs that consumers make among features that inherently vary among powertrain alternatives, including the obvious trade-off between fuel costs and purchase price but also trade-offs involving driving range, acceleration rates, re-fueling time, passenger capacity and luggage capacity. There are corresponding vehicles design trade-offs among all of these features; for example, in a plug-in or battery electric vehicle, greater range can be achieved by increasing battery capacity which increases the vehicle cost/price and requires more space thus reducing luggage and/or passenger capacity. Ideally, vehicles are designed in ways that provide the best overall consumer value across the features, as estimated by the choice models, given these design realities.

The work described in this presentation included development of a set of stated choice exercises, collection of stated choice survey data in seven countries, estimation of vehicle choice models and detailed simulation of vehicle markets in each of the countries. The study was focused on choices among compact sized (class C and C/D) vehicles with conventional gasoline, diesel, hybrid electric, plug-in hybrid and battery electric powertrain/fuel alternatives.

The stated choice exercises were designed to mimic the information conventionally displayed on new car stickers and widely available to consumers through the manufacturers/dealers and various print and online sources. The vehicle features that were included in the design were manufacturer/brand, price, powertrain, fuel efficiency rating, charge time (for battery electric vehicles), annual fuel cost, vehicle range on a single charge or fuel tank, acceleration performance, passenger seating capacity and luggage capacity. Fuel prices were also varied among scenarios but not among alternatives within a scenario. The total number of miles traveled in a year was assumed fixed so the design resulted in a linear dependence among the fuel price, fuel efficiency and annual fuel cost variables and so effects for only two of the three variables could be estimated. However, since fuel efficiency, annual fuel costs and current gasoline prices are commonly displayed on new car stickers, all three of these attributes were correspondingly shown in the experiments.

A significant effort was put into providing consistently-worded survey text across six languages so that differences in estimated model parameters across countries could be properly attributed to differences in preferences. The data collection effort had a target of approximately 5,000 completed surveys, with respondents screened to include current C or C/D class vehicle owners who either purchased their vehicle within the past two years or who intended to purchase with the coming two years. Each respondent saw a block of 8 stated choice experiments from an efficient design of 256 generated using nGene.

Data from the surveys were used initially with simple aggregate multinomial logit models to identify any outliers, test alternative basic model specifications and identify sources of systematic heterogeneity. Nonlinear-in-the-parameters specifications were tested and found to be necessary to adequately represent the effects of several variables including purchase price, vehicle range and charge time. Latent choice models were developed to identify needs-based segments. And, finally, individual-level models were estimated using RSGHB, an open source R package.

The individual-level vehicle choice models were embedded in a market simulator. The simulator included all of the vehicles currently available in the C and C/D classes in each of the regional markets. The market shares resulting from the choice models were calibrated to represent actual current market shares for each vehicle model. The model was then used to estimate changes in market shares that would result from changes in the feature sets for different vehicles.

In addition, a key objective of the work was to estimate the feature trade-offs that consumers make when considering vehicle purchases. Marginal rates of substitution, as expressed in the consumers’ utility functions, were calculated but these provide only a partial picture of how the trade-offs are reflected in the full market. Market indifference curves, representing the trade-offs that are reflected at the market level, were also calculated using the simulator. The market indifference points are the levels for each pair of features at which an increased value of one (e.g., an increased range) is offset in by a decline in value in the other (e.g., an increase in price or a reduction in luggage capacity) so that market shares remain constant. For several reasons, these market indifference points can be significantly different than the marginal rates of substitution.

This presentation will describe the overall study approach but emphasize the practical application of choice modeling to applications involving market-optimal design of product features. 


Conference registration is required in order to view papers.