Heterogeneous consumer preferences for alternative fuel cars in China

Transportation Research Part D 16 (2011) 607–613

Contents lists available at SciVerse ScienceDirect

Transportation Research Part D

journal homepage: www.elsevier .com/ locate/ t rd

Heterogeneous consumer preferences for alternative fuel cars in China

Lixian Qian, Didier Soopramanien ⇑Department of Management Science, Lancaster University Management School, Lancaster LA1 4YX, UK

a r t i c l e i n f o

Keywords:Alternative fuel carsCar ownership in ChinaTransportation incentive policies

1361-9209/$ - see front matter � 2011 Elsevier Ltddoi:10.1016/j.trd.2011.08.005

⇑ Corresponding author.E-mail address: [email protected] (D.

a b s t r a c t

This paper models consumer preferences for alternative fuel cars and conventional, petrolfueled cars in China. It compares discrete choice model specifications where each specifi-cation reflects how consumers choose between alternative fuel cars and conventional typesof cars. We find that, in China, consumers differentiate between the types of alternativefuel cars and are more likely to consider switching from petrol fuel vehicles to hybrid thanto electric cars. Segmentation analysis is used to explore groups of consumers and theirlikelihood of adopting alternative fuel cars.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Car use in many countries is a significant contributor of air pollution. In 2009, China became the largest new car market inthe world far from reaching the saturation level; in 2009 there were still only 10.89 cars per 100 households. The Chinesegovernment has initiated a set of policies, including tax incentives, to encourage consumers to buy greener cars and in 2009,after introducing purchasing tax incentives for small engine cars.

Much of the modeling and estimation of preference towards green cars has tended to impose the choice structure (Caul-field et al., 2010; Potoglou and Kanaroglou, 2007). Here in looking at the Chinese market, we do not impose any a prioriassumption about how consumers perceive the different types of car. Cars that are considered by consumers are likely toshare some common attributes compared to those not considered. This effectively means that if consumers ignore alterna-tive fuel cars when they are deciding to buy, then any policy action to encourage them to buy such vehicles will not have thedesired impact. So, it is important to take into consideration how consumers perceive types of cars when modeling consumerpreference for green versus conventional vehicles.

We also examine if different segments of the market behave differently when choosing between green and conventionalcars. In the context of China, as most potential buyers of green cars will be non-car owners, we see if these consumers aredifferent to existing car owners in terms of their preferences towards car type. Additionally, we examine whether consumersdifferentiate between electric and hybrid cars; most previous work tended to consider the market for green cars as a wholenot taking into account the possibility that some types of green cars may be perceived by consumers as closer substitutes toconventional types of cars.

2. Modeling

We apply discrete choice probability models where utility maximizing behavior is assumed. The utility function of con-sumer (i) choosing car alternative (n) depends on car attributes (X), possible governmental policies (Y) and choice-invariantindividual characteristics (Z):

. All rights reserved.

Soopramanien).

http://dx.doi.org/10.1016/j.trd.2011.08.005

mailto:[email protected]

http://dx.doi.org/10.1016/j.trd.2011.08.005

http://www.sciencedirect.com/science/journal/13619209

http://www.elsevier.com/locate/trd

Fig. 2. Sample choice scenario.

Fig. 1. Tree structures of MNL model and three nested logit models.

608 L. Qian, D. Soopramanien / Transportation Research Part D 16 (2011) 607–613

Uin ¼ Vin þ ein ¼ a0Xin þ b0Yin þ c0Zi þ ein ð1Þ

where utility (Uin) consists of the deterministic portion (Vin) and the error unexplained component is (ein).The assumption on the error component dictates the specification of the discrete choice model. We first consider the mul-

tinomial logit model (MNL) (McFadden, 1974), which assumes that the error term follows the type I extreme value distri-bution where independent of irrelevant alternatives (IIA) is assumed meaning consumers perceive all alternatives to becompletely different to each other. If consumers do perceive that some alternatives in the choice set are more similar to oth-ers, then the IIA property does not hold and we have to allow for the possibility that consumer preferences for some alter-natives are correlated. Here we consider the nested logit model (NL) (Ben-Akiva and Lerman, 1985) and assume that theerror term in Eq. (1) follows a Gumbel distribution. The NL model allows for the grouping of similar alternatives into a nestthereby relaxing the IIA assumption.

We use the tree structures depicted in Fig. 2 based on the different preference assumption towards alternative fuel andconventional types of cars. In the first structure, we assume that both hybrid and electric cars are perceived to be environ-mentally friendly and thus are grouped together to form a ‘‘Green’’ nest/group. In the second, preferences for hybrid and pet-rol cars are assumed to be positively correlated because both of types of vehicle consume oil, although at different efficiencylevels, while the electric vehicle is completely oil-free. We also explore the structure illustrated in tree 3 that assumes petroland electric cars are perceived similar because both use a single fuel. Because, the MNL model is a restricted specification ofthe NL model we can obtain the former from a NL model by making the assumption that all alternatives are independent;tree structure 4.

Table 1Descriptive statistics for the survey sample.

Variables Our sample National sample

Average family size 3.24 2.89Average no. of children less than 18 years old 0.46 /Average no. of working members 2.09 1.49Average household disposable income in 2009 (RMB) 124,900 49,635No. of cars owned per 100 households 66.60 10.89Average age of the household head (years) 37.76 /Proportion of male household head (%) 77.99 /Average distance from home to work place for household head (km) 9.15 /Sample size 527 65,506

Note: Data for the National sample is from China Statistical Yearbook 2010.

Table 2Attributes and levels in the stated choice experiment.

Attributes Vehicle 1 Vehicle 2 Vehicle 3

Fuel type Petrol Hybrid ElectricPurchase price (RMB) Specified by the respondent (1) 30% higher than the similar-sized

petrol car(1) 30% higher than the similar-sizedpetrol car

(2) 50% higher than the similar-sizedpetrol car




Annual running cost(RMB)

Market average level based onvehicle price

(1) 20% less than the similar-sizedpetrol car






Availability of chargingfacility

NA NA (1) 10% of parking spaces

(2) 40% of parking spaces(3) 70% of parking spaces

Vehicle range with fullcharging

NA NA (1) 80 km

(2) 120 km(3) 160 km

Incentives NA (1) 20,000 RMB allowance (1) 30,000 RMB allowance(2) Eligible for priority lane (2) Eligible for priority lane(3) Free Parking for 5 years (3) Free Parking for 5 years

L. Qian, D. Soopramanien / Transportation Research Part D 16 (2011) 607–613 609

3. Data

We use stated choice conjoint data collected as part of an online survey. Students from North China Electric Power Uni-versity and China University of Mining and Technology universities helped collect the data. When the students went backhome, they visited households with an invitation that described our research objectives and how to fill in the survey online.We recognize that an online survey could potentially bias the sample towards those who have computers and the Internet.So the students did also collect data from those households who did not have access to a computer/the Internet. We alsocollected other information on such things as demographic characteristics and car ownership. After deleting cases whereinformation on key variables was missing, we are left with 527 usable responses.

The demographic characteristics of our sample are summarized in Table 1. When the survey is compared with the 2009national urban household survey data, there is no significant difference in family size but high-income groups and those whoown cars are over represented. We therefore reweight our data based on household income and car ownership informationfrom the national sample.

For the stated choice data, consumers were presented scenarios involving one hypothetic conventional petrol, one hybridand on electric car. We constructed these three choice offerings of cars based on a list of attributes with three levels (Louviereet al., 2000) (Table 2). There are three incentives for potential buyers of hybrid or electric cars: a cash subsidy and two non-monetary incentives, fast lane access and free parking. The cash subsidy for the electric car is slightly higher than for thehybrid car because it consumes no oil.1 Priority lane and free parking policies are not yet available in China but we investigateconsumers’ preference towards such policies to test if and how consumers would respond to such incentives. Because the

1 This difference in cash subsidy is similar to that the government is planning to offer.

Table 3Parameters from the MNL and NL models.

Variable MNL NL (1) NL (2)

Coeff. t-stat. Coeff. t-stat. Coeff. t-test

Vehicle attribute (a)Purchase price �0.014 �16.018 �0.017 �13.253 �0.014 �15.433Running cost �0.059 �7.263 �0.057 �6.835 �0.064 �7.596Charging facility 0.933 5.23 0.971 5.333 0.926 5.183Vehicle range 0.005 3.982 0.005 4.02 0.005 4.014

Incentive (b)Cash subsidy 0.019 1.739 0.020 1.8 0.017 1.518Free parking 0.412 1.534 0.429 1.548 0.364 1.331Priority lane 0.165 0.618 0.176 0.641 0.114 0.419

Individual characteristics interacted with electric car (c1)Electric car, constant �1.307 �3.113 �1.392 �3.203 �0.986 �2.33No. of young children �0.583 �5.464 �0.638 �5.638 �0.502 �4.682No. of drivers �0.434 �6.227 �0.474 �6.375 �0.364 �5.248Family size 0.296 5.873 0.314 5.915 0.272 5.488Household Income 0.004 2.142 0.004 2.027 0.004 2.649No. of owned cars (0, 1, 2 or 3 cars) 0.296 1.869 0.294 1.776 0.349 2.238Age of family head �0.009 �1.759 �0.010 �1.746 �0.009 �1.821Sex of family head �0.821 �7.627 �0.855 �7.526 �0.821 �7.876Working distance of family head �0.032 �4.879 �0.034 �4.984 �0.029 �4.457

Individual characteristics interacted with hybrid car (c2)Hybrid car, constant 0.866 2.771 0.827 2.525 0.869 2.752No. of children �0.500 �6.492 �0.559 �6.508 �0.516 �6.676No. of drivers �0.315 �6.158 �0.349 �6.166 �0.324 �6.302Family size 0.049 1.284 0.064 1.546 0.055 1.436Household Income 0.003 2.643 0.003 2.407 0.003 2.564No. of owned cars (0, 1, 2 or 3 cars) 0.371 3.323 0.373 3.113 0.362 3.231Age of family head �0.007 �1.834 �0.007 �1.792 �0.007 �1.777Sex of family head �0.212 �2.473 �0.236 �2.547 �0.228 �2.648Working distance of family head �0.025 �5.528 �0.028 �5.506 �0.025 �5.464

Inclusive value (IV) parameters (k)Green nest in NL(1) 0.920 32.196Oil-consuming nest in NL(2) 0.772 15.331Log-likelihood at convergence �3952.366 �3949.104 �3940.251Log-likelihood at zero �4631.749 �4631.749 �4631.749Rho-square w.r.t. zero: 0.1467 0.1474 0.1493v2 against zero 1358.766 (df = 25) 1365.290 (df = 26) 1382.996 (df = 26)v2 against MNL 6.524 (df = 1) 24.230 (df = 1)


complete experiment design has 6561 scenarios, it is unrealistic to implement and we thus derive the orthogonal fractional de-sign with 32 scenarios which reduces the complexity of the tasks of the stated choice experiment for consumers.

We first introduce our experiment with detailed explanations about the three alternatives followed by their attributesand the incentives. Then eight choice scenarios are randomly selected from a total of 32 scenarios and presented to eachrespondent, where the purchase price and running costs of the base, petrol car are customized based on the respondent’spreference stated beforehand. The respondents are then asked to select one vehicle from each scenario. A sample choice sce-nario is shown in Fig. 2.

4. Analysis

The alternative specific attributes include purchase price, annual running cost, charging facility and vehicle range for elec-tric cars. The incentive variables consist of three types of incentives. We also account for the demographic characteristics ofthe households. We estimate the MNL model and the three different specifications of the NL models based on 4216 obser-vations from 527 respondents. The results are presented in Table 3.

The log-likelihood ratio index (p2) value of our MNL model and falls within the range found in other studies includingEwing and Sarigollu (1998) and in Golob et al. (1993). The log-likelihood ratio test implies that this MNL model is signifi-cantly different from the model without any variables. A Hausman test is used to explore the IIA assumption: The hypothesisthat consumers perceive the alternative car scenarios to be completely independent at the 5% significance level indicatingthat we need to explore models that account for correlated preferences of alternatives.

The NL models are estimated based on the three tree structures as shown in Fig. 1. A key parameter of the NL models isthe inclusive value (IV) parameter that has to lie between zero and one for the model to be consistent with the utilitymaximization described by the tree structures. Amongst the three choice structures, the third that groups electric and petrol

Table 4Cross-elasticity comparison of the MNL and NL models.

Cross elasticity in MNL model Cross elasticity in NL model (2)

Changing attribute Hybrid Petrol Hybrid Petrol

Electric carPurchase price 0.415 0.415 0.368 0.368Running cost 0.106 0.106 0.108 0.108Cash subsidy �0.044 �0.044 �0.039 �0.039Free parking �0.016 �0.016 �0.014 �0.014Priority lane �0.005 �0.005 �0.004 �0.004Charging facility �0.049 �0.049 �0.049 �0.049Vehicle range �0.083 �0.083 �0.083 �0.083

Electric Petrol Electric Petrol

Hybrid carPurchase price 1.206 1.206 0.92 1.232Running cost 0.312 0.312 0.263 0.354Cash subsidy �0.078 �0.078 �0.053 �0.072Free parking �0.046 �0.046 �0.031 �0.042Priority lane �0.016 �0.016 �0.009 �0.012

Electric Hybrid Electric Hybrid

Petrol carPurchase price 1.014 1.014 0.772 1.025Running cost 0.600 0.600 0.509 0.677


cars together yields an IV parameter that is significantly greater than one indicating its is inconsistency with random unitymaximization criterion. Therefore, we only present the parameter estimation results for the models based on structures 1and 2. The IV parameters for these are 0.920 for the ‘‘Green’’ nest and 0.772 for the ‘‘Oil-consuming’’ nest and are also sig-nificantly different from zero at 1% significance level. Moreover, Wald-test statistics indicate that they are also significantlyless than one, thus concurring with the results of the Hausman test where we reject the IIA assumption

The performance of the NL models is further investigated by performing log-likelihood tests (�2(LLbase � LLNL)). Com-pared with the MNL model, both NL models have much higher test values than the critical X2 value at the 5% significancelevel. If we compare the two NL models, we find that the second tree structure, which groups hybrid and petrol cars in‘‘oil-consuming’’ nest, is better, because its log-likelihood value is higher than under the first tree structure and the IV param-eter is also significantly more different from one according to the Wald-test. This would indicate that consumers perceivepetrol cars and hybrid cars to be more similar to each other compared to electric cars.

In Table 3 the parameters of the vehicle attributes are significant and the effects are similar to those in Potoglou andKanaroglou (2008). More specifically, the parameters of monetary attributes such as purchase price and running cost havenegative signs implying consumers prefer low-priced cars and low running costs. The positive parameters of charging facilityand driving range indicate the importance of providing more charging facilities and designing electric cars with longer rangebatteries. With regards to the various incentives, the parameters of all three incentives are insignificant in contrasts to pre-vious research conducted in more mature markets.2

With regards to the impact of the individual characteristics, households with young children and where there is morethan one driver prefer conventional petrol cars. This may reflect consumers’ concerns about the technological immaturityof AFCs in the short-term and a believe that it would be safer to buy a conventional vehicle, particularly when they haveyoung children and/or where there is a less experienced second driver. Larger households also tend to prefer alternative fuelcars, particularly electric cars. As expected, households with higher incomes are more willing to adopt alternative fuel cars ascan be seen by the positive and significant effect of household income variable when interacted with either electric or hybridcars in both NL models.

In addition, households with a male head are less likely to adopt green cars and young household heads are found to havestronger preferences towards AFCs, although these effects are only significant at the 10% level. Household heads who arelong-distance commuters are less likely to choose AFCs, possibly due to their concerns about the lack of fueling facilitiesand limited vehicle range which is similar to the findings of Ewing and Sarigollu (1998, 2000) and Potoglou and Kanaroglou(2007) amongst others. The results also show that car owners are more likely to switch to alternative fuel vehicles comparedto non-car owners.

To demonstrate why it is important to choose a model that appropriately fits the market context, we compare the elas-ticity effects between MNL and NL models in Table 4. Since we have shown that the second tree structure is the best NL mod-el, we only compare the resulting elasticity effects with those from the MNL model. In the first instance, the second tree

2 For example, both Ewing and Sarigollu (2000) and Potoglou and Kanaroglou (2007) find that a price subsidy such as waiving purchasing tax is a significantincentive to buy green cars in Canada.

Table 5Comparison of IV parameters of segmental NL models by car ownership.

NL model (1) NL model (2)

IV parameter (k) SE of (k) IV parameter (k) SE of (k)

Car owners 0.578 0.058 0.985 0.185Non-car households 0.948 0.037 0.725 0.063

Fig. 3. Direct-elasticities comparison based on car ownership status.


structure (Fig. 1) allows the cross-elasticities in the NL to vary for the hybrid or petrol cars across alternatives compared tothe MNL. The magnitude of the elasticity effects further shows that hybrid and petrol cars are perceived by consumers to besimilar. The cross elasticity effects of the attributes for hybrid cars have a bigger impact on the probability to buy a petrol carthan on the probability to buy an electric car.

We look at different types of consumers to see if they have heterogeneous preferences towards AFCs and conventionaltypes of cars. We look at two segments: car owners and non-car owners and estimate separate NL models for the segmentsbased on tree structures 1 and 2. We do not present the full set of parameters but we compare the key IV parameters and theelasticity effects for these segments.

For each segment, the IV parameters for the two tree structures and the corresponding log-likelihood value at conver-gence are summarized in Table 5. For car owners, NL model 1 is better than model 2 because its IV parameter is significantlydifferent from both zero and one while for model 2 it is insignificantly different from one, implying that car owners doconsider hybrid and electric cars to be similar4 and perceive petrol car to be a separate alternative. However, this is not


the case for those households who do not own cars. The IV parameter of NL(1) in this segment is insignificantly differentfrom one while for NL(2) for non-car owners’ it is significantly different from both one and zero, which implies stronger pref-erence correlation between hybrid and petrol cars for non-car owners. Basically, car owners and non-car owners do not thinkabout the AFCs and conventional types of cars in the same way.

Fig. 3 gives the elasticity effects for car and non-car owners for both NL models. We find that non-car owners are moresensitive than car owners to monetary attributes such as purchase price and annual running cost suggesting that existing carowners are less concerned by the high purchase price of AFCs compared to non-car households. Further, car owners are lesssensitive to purchase price but more responsive to the cash subsidy than non-car households. Households that are yet to owna car are more concerned about their ability to afford a car and they are probably focusing on the purchase price and not onthe price differences between the types of cars. Existing car owners positively respond to incentives compared to householdsthat do not own cars. Car owners are also less concerned about charging facility and vehicle range of electric cars which maybe due to their better understanding from prior driving patterns of how they would use an electric vehicle.

5. Conclusions

In our investigation of the effects of price and fuel/running cost and of government incentives on Chinese consumer pref-erences to buy alternative fuel cars we find that consumers do consider green and conventional types of cars when they thinkabout buying cars. However, they are more likely to consider hybrid and conventional types of cars but are less likely to con-sider electric cars suggesting that greening of the vehicle fleet will be through the greater use of hybrid rather than electriccars.

References

Ben-Akiva, M., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge, MA.Caulfield, B., Farrell, S., McMahon, B., 2010. Examining individuals preferences for hybrid electric and alternatively fuelled vehicles. Transport Policy 17,

381–387.Ewing, G., Sarigollu, E., 1998. Car fuel-type choice under travel demand management and economic incentives. Transportation Research Part D 3, 429–444.Ewing, G., Sarigollu, E., 2000. Assessing consumer preferences for clean-fuel vehicles: a discrete choice experiment. Journal of Public Policy and Marketing

19, 106–118.Golob, T.F., Kitamura, R., Bradley, M., Bunch, D.S., 1993. Predicting the market penetration of electric and clean-fuel vehicles. Science of the Total

Environment 134, 371–381.Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods: Analysis and Application. Cambridge University Press, Cambridge, MA.McFadden, D., 1974. Conditional logit analysis of qualitative choice behavior. In: Zarmebka, P. (Ed.), Frontiers in Econometrics. Academic Press, New York.Potoglou, D., Kanaroglou, P.S., 2007. Household demand and willingness to pay for clean vehicles. Transportation Research Part D 12, 264–274.Potoglou, D., Kanaroglou, P.S., 2008. Disaggregate demand analyses for conventional and alternative fueled automobiles: a review. International Journal of

Sustainable Transportation 2, 234–259.

Documents

Heterogeneous consumer preferences for alternative fuel cars in China