Socio-Economic Differentials and Stated Housing Preferences in Guangzhou, China

ARTICLE IN PRESS

Habitat International 30 (2006) 305–326 www.elsevier.com/locate/habitatint

Socio-economic differentials and stated housing preferences in Guangzhou, China

Donggen Wang, Si-ming Li

Department of Geography, Hong Kong Baptist University, Kowloon Tong, Hong Kong, China

Abstract

Households in Chinese cities today have to increasingly rely on the market to satisfy their housing needs. The growing freedom in choosing one’s own residence implies increased variations in all aspects of housing consumption. Examination of individuals’ housing preferences is crucial in understanding these variations. This paper studies the housing preference of Guangzhou people through choice experiments framed in state-of-the-art experimental design methods. Joint logit models comprising both neighbourhood and dwelling attributes are estimated for all subjects and for various sub-samples classiﬁed by family income, age, education, nature of employment organization, district of current residence, etc. The models are then used to compute utilities for different attribute levels, the impacts of these attributes on choice probabilities, and the relative prices that the subjects are willing to pay for buying a home in different districts, with different accessibilities, of different types, etc. Neighbourhood and location-related attributes are found to be more important than dwelling-related attributes in home purchase decisions. Further, factors such as family income, age, education, nature of employment organization, etc. are found, to various degrees, have affected housing preference. Based on the preference structures revealed, we envision a new urban morphology to take shape in Chinese cities which is not too dissimilar from the ones in cities in the West, with the inner core dominated by the aged and the urban poor and the outskirts occupied by younger people and the rich and well-educated class. r 2004 Elsevier Ltd. All rights reserved.

Keywords: Housing studies; Stated preference approach; Housing market; China

E-mail addresses: [email protected] (D. Wang), [email protected] (S.-m. Li).

306 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

Introduction

Two and half decades have passed since China first launched its housing reform. For many years under China’s planned economy, construction and provision of urban housing rested mainly upon the state work units or simply work units or danwei1and were subsumed under capital construction investment allocated to the work units under the annual budgetary exercise (Wu, 1996). Efforts were made during the early reform periods to disengage the work units from being directly involved in housing construction. Development companies were established to build ‘‘commodity housing’’ for sale, presumably according to market principles. Initially, though, the great majority of such commodity housing was sold to the work units for subsequent allocation to their workers (Li, 2000). Individual households purchasing homes directly in the market were rare. Also there were restrictions as to who could buy this ‘‘commodity housing’’. In general, foreigners (including Hong Kong and Taiwan ‘‘compatriots’’) were precluded from doing so. In cities with sizeable numbers of expatriates such as Beijing, Shanghai and Guangzhou, commodity housing for foreigners or waixiao shangpifang, which was usually of higher quality, was also built to accommodate the needs of and perhaps also to better monitor the foreign population. Urban development tied to specific capital construction investment projects tended to be associated with a high degree of haphazardness and uncertainty, especially when funding was subject to the outcome of the annual budgetary exercise. A major objective of the supply side reform was to bring in more orderly urban growth. Under the reform, the municipal government rather than the work units became the most important player orchestrating urban development. Often the municipal government would allocate a large tract of land to a development company, usually at a price. The development company would then undertake housing and other real estate development projects. It would also be responsible for roads, sewage, landscaping and other infrastructure provision. By the early 1990s the bulk of new housing in urban China was commodity housing built by the development companies. Invariably the new housing estates were located outside the former work unit compounds, resulting not only in increased commuting but also in new dimensions of differentiation of the urban residential space. Comparatively speaking, demand-side reforms were carried out with much greater caution. In China the term ‘‘public housing’’ refers to the housing provided by the state work units and also the municipal housing bureau. Until mid-1990s the reform was restricted to gradually raising public housing rents and selectively selling public housing to workers of state work units at discounted prices (Li, 2000). Housing allocation remained largely the prerogative of the work units. In fact it has been revealed that state work units had played an even greater role in housing provision under the reform, despite all the rhetoric of marketization and commodification (Wu, 1996; Li, 2003). In recent years, though, the pace of reform has quickened considerably. The pronouncement of cessation of welfare allocation of housing by the former premier, Zhu Rongji, in 1998 perhaps marked a watershed in China’s urban housing reform history. Since then, there have been massive disposals of the stock of public housing. In Shanghai and a few other cities full property rights have been given to owners of ‘‘reform housing’’, i.e., former public housing that

1This include state-owned enterprises as well as government departments and party and quasi-government organizations. ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 307 had been sold to workers of state work units and others at highly subsidized prices. The domestic housing market and the market for foreigners have also been merged. A major obstacle hindering home purchase by individual households in the market was the problem of affordability. Under the traditional work unit system, in-kind payment including virtually free housing constituted a major part of the reward the work unit paid to its workers. Monetary income was of relatively minor importance. This resulted in exceedingly high price-to- income ratios for commodity housing. In Beijing, for example, in 1992 the average price of commodity housing stood at RMB21613 per m2. A 60 m2 apartment would cost RMB96780. The average household income, on the other hand, stood at RMB 8300. The price-to-income ratio was therefore 11.65 (see Lau, 2003, for details of these and the computation on price-to-income ratios reported below). That is, the average household had to save all its income for more than 11 years in order to purchase a flat in the open market! Obviously, commodity housing was beyond the means of most except the very rich. Five years later, in 1997, the average household income increased markedly to RMB24056. Yet the average price for a 60 m2 apartment rose by an even larger margin, to RMB320000. The price-to-income ratio increased further to 13.31. In more recent years, there has been an apparent revamp of the remuneration system. There have been further and quite drastic increases in wage, and many work units now grant cash subsidies to their workers for home purchase in the open market. Also, home price appears to have stabilized. In fact, the average price for a 60 m2 apartment in Beijing dropped slightly to RMB283000 in 2001 while the average household income continued to increase to RMB34980. Hence, the price-to- income ratio declined to 8.09. Commodity housing is now more affordable. The rate of urban homeownership has showed corresponding sharp increases. For the first time in the history of the People’s Republic of China, a large portion of urban households can now exercise choice in housing consumption. The market finally begins to reign. The increasing role played by the market and the growing freedom in choosing one’s own residence imply increased variations in all aspects of housing consumption in China: where the residence is located; what kind of neighbourhood and location attributes the dwelling is associated with; in what tenure mode housing is consumed and how much; etc. Probably because of this, recently more attention has been given to the individuals and individual households in China housing research. One major area of concern is housing tenure. Li (2000a), employing data derived from a sample survey in Guangzhou, studies how different types of households are channelled to different types of housing under a semi-marketized regime. Huang and Clark (2002) and Ho and Kwong (2002) study tenure composition. They conclude that both market mechanisms and institutional forces are of importance in structuring the mode of housing tenure. A related area of research is residential mobility. Li and Siu (2001a, b) examine mobility behaviours in Beijing and Guangzhou and reveal the continual dominance of danwei in determining residential location, especially the movement to the suburbs. Li (2003) further reveals that while the direction of movement is related to current housing tenure, it is unrelated to previous tenure. Based on retrospective residential histories, Li (forthcoming) finds that in Beijing the rate of residential mobility has exhibited a fluctuating but slightly downward trend since 1985, despite the marketization drive.

2RMB stands for Renminbi, the Chinese currency. At current rate of exchange, RMB 1=US $0.12 approximately. ARTICLE IN PRESS

308 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

The above-cited works often purport to study tenure and housing choice, with the view of unravelling residential preference of individuals and households. Both Ho and Kwong (2002) and Huang and Clark (2002) claim to have estimated a choice model of the McFadden (1973) variety (see also Quigley, 1976; Friedman, 1981). Yet their studies were based on the data collected at a time when the state work units and other institutional forces still dominated the housing provision scene. Even today, the housing market in urban China is still in its infancy. Work units and the municipal housing bureau remain instrumental in deciding who get what, where, and how much. The lack of a market clearing process means that the usual revealed preference approach is inadequate, if not entirely inappropriate, in eliciting housing preferences in urban China. Thus, the results of the aforementioned studies provide no more than anecdotal evidence on the preference structure. To date, very little is known about how individuals and households in urban China make housing decisions in a market context. Are accessibility considerations important? What about neighbourhood safety and availability of social capital in the neighbourhood, especially in light of the gradual dismantling of work unit compounds? How do people in Chinese cities view dwelling attributes, such as type of dwelling, size and layout of the dwelling, etc.? Are they willing to trade one attribute for another, and if so, by how much? Previous studies conducted in western countries have revealed systematic variations in the preferences for housing attributes and hence the choice of housing according to position in the family life cycle and socio-economic status (Quigley, 1976; Friedman, 1981; Clark & Dieleman, 1996). How and to what extent do housing preferences vary across socio-economic classes and demographic groups in the Chinese case? Education, for instance, broadens a person’s horizon and helps inculcate a certain worldview. A person’s preference structure, therefore, varies with his/her education attainment. In a redistributive economy, which until very recently has characterized China, education attainment is particularly instrumental in determining a person’s position in the employment hierarchy and hence the kind of resources at his/her disposal (Szelenyi, 1983). Does education have the same effect on housing preference for Chinese people as it has for people in the West? In this connection, it may be argued that both employment and previous housing experience will have an effect on housing preference formation. In the case of China, work unit compounds used to dominate the landscape. Do people with different housing experiences, in particular do people living in work unit compounds and people living elsewhere, differ systematically in terms of their views on location and dwelling attributes? The answer to all these questions have direct relevance for the understanding of how housing decisions are made in China in an increasingly market- driven setting. To address these questions in the absence of a well functioning housing market, an alternative approach, more speciﬁcally the stated preference approach, is needed. Recently, Wang and Li (forthcoming) conducted a stated preference experiment in Beijing. To our knowledge, this is the ﬁrst serious attempt to study housing preference in China. As the capital of China, Beijing is perhaps the last stronghold of the planned economy. Government and Party organizations and state enterprises dominate the employment scene. The Beijing sample comprises a rather homogeneous group and may not be representative enough to reveal the extent of variability in housing preferences that are emerging in other Chinese cities. The present study builds upon and extends the Beijing study. A companion experiment was conducted in Guangzhou in summer 2001. Guangzhou has served as China’s southern gateway and in many occasions the only trading port to the outside world for more than two millenniums. Its rich ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 309 commercial history and geographic proximity to Hong Kong accorded Guangzhou special positions in the reform period. In comparison with that of Beijing, the Guangzhou economy was much more complex and open. Market elements and foreign investments began to infiltrate Guangzhou in the early reform period. Inflow of foreign capital already made an imprint in the mid 1980s (Li & Chu, 1987). Economic growth accelerated in the 1990s. Per capita GDP reached RMB 38007 in 2001 (SBGP, 2002, p. 75), ranking second amongst all cities in the country. In the realm of housing, Guangzhou was among the first cities in China to introduce the housing provident fund. It also took lead in terminating ‘‘welfare allocation of housing’’ (Guangzhou & Bianzuan, 2001, p. 223). In fact, real estate developments mimic those in Hong Kong already mushroomed in the early 1990s. Li (2000) and Li and Siu (2001a,b) reported that open market housing made up nearly 30% of their sample, which was collected in 1996. In short, the Guangzhou population exhibits high degrees of heterogeneity and the people in Guangzhou have long and rich experience in dealing with the market. In this study, we first depict the general structure of housing preference in Guangzhou, based on choices expressed by subjects of the stated preference experiment. In addition to computing the utilities derived from various location and dwelling attributes and their effects on the choice probabilities, we also examine in more explicit terms the trade-offs between attributes. Attempts will be made to compare our findings with those of Wang and Li (forthcoming) for Beijing to assess to what extent urbanites in China share common housing preferences. The much more heterogeneous Guangzhou sample allows us to explore such effects as previous residential experience and employment on housing decisions, which was not feasible for the Beijing study. In this sense the research findings to be detailed below provide a much richer description of housing preference structures in Chinese cities than hitherto available.

Modelling housing preference in a semi-marketized housing system: methodology and data

The stated preference approach

Many housing studies have adopted the hedonic approach to analyse how the marginal value of housing attributes is priced (Rosen, 1974; Palmquist, 1984; Megbolugbe, 1991). Quigley’s (1976) seminal work, which introduces the discrete choice analysis approach, has stimulated many similar applications (Friedman, 1981; Fischer & Aufhauser, 1988; Timmermans & van Noortwijk, 1995). While most discrete choice-based housing studies use revealed preference data (i.e., housing choice data from the real market), an increasing number of applications adopt the Stated Preference (SP) method and use experimental data (Timmermans & van Noortwijk, 1995; Tu & Goldﬁnch, 1996; Earnhart, 2002; Walker, Marsh, Wardman, & Niner, 2002; Wang & Li, forthcoming). The stated preference method and its advantages and disadvantages in comparison with the revealed preference approach are well documented in the literature. See Earnhart (2002) and Walker et al. (2002). The stated preference method has proved to be particularly useful where there is an absence of actual market information from which preferences can be revealed (Walker et al., 2002). The fact that a full-ﬂedged housing market is yet to be established in Guangzhou implies that no reliable market information is available to derive housing preference and the ARTICLE IN PRESS

310 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 stated preference method is probably the only choice for modelling housing preference in a semi- marketized housing system like that of Guangzhou. Housing choice is a multi-dimensional exercise, involving the choice of tenure, housing type, neighbourhood, location, etc. Most studies examine only one or maximally two choice dimensions. In particular, the stated preference method, almost as a rule, is applied to model a single choice dimension. The ﬁrst attempt to model two housing choice dimensions was made by Wang and Li (forthcoming). The same modelling technique is adopted here. We analyse two choice dimensions: the choice of dwelling and the choice of neighbourhood.

Selection of attributes, experiment design and sampling

The set of housing attributes employed here is similar to that employed in the Beijing study. Table 1 presents details of the selected attributes and their levels. Four attributes are used to define neighbourhood, namely, accessibility, living convenience, security and district. Accessibility is usually defined in terms of distance to work in the literature (for example, Tu & Goldfinch, 1996). This may be appropriate for western cities where the private car is the major mode of transport and people have a good sense of distance. It may not, however, be the best way to define in Chinese cities like Guangzhou where public transport (bus, subway, etc.) is the choice for the majority. Instead, accessibility is defined in this study in terms of access to public transport: (i) highly accessible, with public transport connections to all districts in the city; (ii) reasonably accessible, with public transport connections to major business centres; and (iii) limited accessibility, with very few public transport links with the rest of the city. Living convenience refers to the convenience of daily goods shopping. Urban Chinese families typically undertake

Table 1 Attributes and levels

Attributes Levels

Neighbourhood District 1. Yuexiu; 2. Liwan; 3. Dongshan; 4. Haizhu; 5. Tianhe; 6.Baiyun; 7. Fangcun; 8. Huangpu Accessibility to public 1. Highly accessible, public transport connections to all districts; 2. Reasonably accessible, transport public transport connections to major city centres; 3. Limited accessibility, very few public transport connections with other districts Living convenience 1. Fresh and daily goods markets available within 500 m; 2. Fresh and daily goods markets (shopping) available within 1000 m; 3. Fresh and daily goods markets available beyond 1000 m. Security 1. Good public order; 2. Poor public order Dwellings Price 1. 4000 yuan/m2 or below; 2. 4000–5000 yuan/m2; 3. 5000–6000 yuan/m2; 4. 6000 yuan/m2 or above Orientation 1. East; 2. South; 3. West; 4. North Type 1. Detached house; 2. Apartment building of 4 floors or fewer; 3. Apartment building of 5 floors or more without lift; 4. Apartment building of 5 floors or more with lift Layout 1. Small living room but large bedrooms; 2. Large living room but small bedrooms Property management 1. Pay for property management; 2. No property management ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 311 daily shopping for fresh vegetables, meats, etc. Thus, availability of daily and fresh markets in the vicinity of home is believed to be an important consideration for housing choice. Three levels of living convenience are specified: fresh and daily goods markets available within 500 m, within 1000 m, and beyond 1000 m. The majority of urban Chinese used to live in work unit compounds where most residents were working in the same work unit and security was not a major concern. The new commodity housing estates, however, have a greater mixture of people, which causes concern for security problems. This concern is further aggravated by the rising crime rates in conjunction with much loosened control over migration flows and presence of large numbers of laid-off workers in cities in recent years. It is not surprised to find that in many Chinese cities neighbourhood security has become an important consideration for housing purchase. To gauge this influence, we define two security levels: ‘good public order’ and ‘poor public order’. The district dummy is included to capture the unmeasured components of neighbourhood features such as social class composition and district reputation. The eight urban and inner suburban districts of Guangzhou form the eight levels of the attribute. To help comprehend the modelling result regarding people’s district preference, we briefly introduce the eight districts, the spatial configuration of which is depicted in Fig. 1. Dongshan, Liwan and Yuexiu comprise the pre-Reform (and indeed pre-1949) urban core. While all three suffer from severe crowdedness,

Fig. 1. Spatial conﬁguration of districts in Guangzhou. ARTICLE IN PRESS

312 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 with the population density ranging from 34,800 per km2 for Dongshan to 48,600 per km2 for Yuexiu (SBGZ, 2001, p. 57), these core districts nevertheless contain the bulk of urban amenities in Guangzhou. Planned as a district for government and Party offices, Dongshan is somewhat less crowded and has better provision of parks and other greenery. In many respects Dongshan is the most reputed district in the city. The population of Dongshan comprises a disproportionate number of cadres and intellectuals (Xu, Hu, & Yeh, 1989). Haizhu is a rather peculiar district. A small strip on the waterfront facing Liwan and Yuexiu across the Pearl River, known as Henan in the past, used to constitute part of the inner core. Yet until recently much of Haizhu has remained rural. Also, shipyards and heavy industries dot the landscape. Located to the east of Dongshan, Tianhe has been Guangzhou’s main district for urban expansion from the late 1980s onwards. A new central business district, with the Guangzhou East Station and Tianhe Stadium being the focal points, is fast emerging in the district. Completion of Stage I of Guangzhou Metro in 1998 has substantially improved Tianhe’s accessibility. The other three districts, namely, Baiyun, Fangcun and Huangpu, are Guangzhou’s suburbs. Baiyun Mountain, the city’s largest nature reserve, occupies much of Baiyun District. Despite its rather congenial natural environment, the district’s hilly landscape limits accessibility and acts as a major developmental constraint. In addition, the Guangzhou International Airport, which is located in this district, poses an environmental nuisance. Fangcun, separated from Liwan by the main channel of the Pearl, was formerly known as Huadi or ‘‘flower place’’. Until the early 1990s horticulture was Fangcun’s main economic activity. Completion of a cross- river tunnel in 1996 provided the district with a ready link to the inner core. Accessibility improved further with the extension of Guangzhou Metro to Fangcun. But the river still acts a major impediment to communication. To the mind of Guangzhou people, Fangcun remains an inaccessible and backward rural region. Huangpu, located some 30 km down the Pearl River from the old urban core, is Guangzhou’s deep-water port and heavy industrial base. To many Guangzhou residents, Huangpu is inaccessible and a place with rather undesirable living environments. Five attributes, namely, price, orientation, layout, dwelling type, and whether management fee is needed, are selected to define dwelling. Though price is considered in this study as a factor differentiating dwellings, it should be noted that it is a variable related not only to dwelling but also to neighbourhood and location. The average selling price of commodity housing in Guangzhou was about RMB 3,700 per m2 in 2000 (see SBGZ, 2001, p. 37). Given the fact that housing price had increased during the years proceeding 2001 and we should provide a full range of prices for respondents to consider, the following four price levels are defined (in RMB per m2): 4000 or below; between 4000 and 5000; between 5000 and 6000; and 6000 or above. The direction that the dwelling faces affects sunlight penetration and ventilation. It is an important housing consideration in China. To evaluate the impact of this factor on housing choice, the direction variable is included and the four direction descriptions, namely, ‘North’, ‘South’, ‘East’ and ‘West’, are used as the levels to differentiate people’s preference towards the four different housing directions. The attribute of dwelling type is selected to capture the importance of dwelling type on housing choice and the preference towards different types. Based on the availability of housing types in the market, five levels are identified: ‘detached house’; ‘apartment building of 4 storeys or fewer’; ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 313

‘apartment building of 5 storeys or more without lift’; and ‘apartment building of 5 storeys or more with lift’. Dwellings in China used to have large bedrooms but small living rooms. This is because the functions of living room were not fully recognized and people liked to have many wardrobes in their bedrooms. Nowadays, people, in particular the younger generation, seem to favour the layout of large living room but small bedrooms. To capture this preference change, this variable is selected. Two major layouts of dwelling are identified: ‘small living room but large bedrooms’ and ‘large living room but small bedrooms’. Finally, to see whether property management will affect housing choice, we include presence/ absence of management fee as another attribute in the model. People in Chinese cities have become aware of the importance of estate management, but this requires payment of management fee. It is thus important to see if respondents are willing to pay for property management. There are, of course, other attributes that may be used to define neighbourhood and dwelling. The problem is that respondents are not able to evaluate too many attributes in the experiments and if too many attributes are involved, the problem becomes not tractable. In any case we believe that the above are among the most salient variables that structure the housing preference of urban Chinese. Statistical methods are used to combine the selected attributes into hypothetical choice alternatives of neighbourhood and dwelling. The uniform design method introduced by Wang and Li (2002) and the choice experiment design technique for modelling multi-dimensional choices proposed by Wang, Borgers, Oppewal, and Timmermans (2000) are employed to derive the hypothetical housing choice experiments. Specifically, a total of 72 choice sets are generated. Each choice set consists of three choice alternatives for neighbourhood and three for dwelling. The data were collected in conjunction with an interview survey of 1500 Guangzhou residents in summer 2001. A multi-level probability proportion to size (PPS) sampling strategy was adopted to select the subjects. The subjects were to be distributed in proportion to the population of the original eight urban districts of Guangzhou (Fig. 1). The heads of households, identified as such by members of the family interviewed, were invited to do the choice experiments. To limit the burden on the respondents and ensure data quality, each respondent was presented with only four choice experiments. Perhaps, because the sample comprised only heads of households, male respondents outnumbered females by a factor of 2:1. In terms of age, the sample approximated the distribution of the Guangzhou population aged 20 and over, as given by the 2000 Population Census. As for the current home ownership, 67.8% of the respondents had the ownership of their house either from the housing market (40.4%), or work units (28.8%), or other sources (such as inherited from parents, 30.8%). The rest either rented their house or did not provide this information. With respect to household income, the sample exhibited substantial variability, in line with Guangzhou’s relatively open and complex economy. While 23.3% had annual incomes below RMB 20000, 11.3% earned incomes in excess of RMB 60000. Regarding employment organization, 51.7% of the sampled subjects worked in state work units, and 48.3% in non-state work units. In comparison, the Guangzhou Statistical Yearbook gave corresponding figures of 52.1% and 47.9% (SBGZ, 2001, p. 73). All in all, the sample approximated the Guangzhou population quite well. ARTICLE IN PRESS

314 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

Developing econometric models from stated preference data

Discrete choice models As there are two choice dimensions involved in this study, one may estimate either joint logit models or nested logit models (Ben-Akiva & Lerman, 1985). The speciﬁcation of the joint as well as nested logit model for the choice of neighbourhood and the choice of dwelling is given in Wang and Li (forthcoming). Once the choice model is estimated, one may assess the impact of attributes on choice probability by calculating the marginal choice probability in relation to the marginal utility of attributes. For example, to assess the impact of one unit change of attribute X on choice probability, we may construct two choice alternatives A and B, which are different only on attribute X by one unit: the value of attribute X of A is x, while that of B is x þ Dx. Based on the utility function estimated, we may calculate the choice probability of the two alternatives: expðV þ b xÞ PðAÞ¼ x ; ð1Þ expðV þ bxxÞþexp½V þ bxðx þ DxÞ

exp½V þ b ðx þ DxÞ PðBÞ¼ x ; ð2Þ expðV þ bxxÞþexp½V þ bxðx þ DxÞ where P(A) and P(B) are the choice probability of A and B respectively, V is the systematic component of other attributes and bx is the coefﬁcient of attribute X. The impact of one unit change of attribute X on choice probability can then be calculated by: PðBÞPðAÞ ¼ expðb DxÞ1; ð3Þ PðAÞ x which is the relative choice probability change.

Willingness-To-Pay When a cost variable (such as price) is included in the utility function of the choice model, it is possible to value the other house attributes in dollars, thus providing information similar to that calculated by the hedonic price approach. This is referred to Willingness-To-Pay (WTP), a concept and measurement that are widely used to value travel time savings (Evans, 1971; Hensher & Truong, 1984; McFadden, 1998; Louviere, Hensher, & Swait, 2000, p. 61) and environmental goods and services (Goodman, 1989; Freeman, 1991; Huang, Haab, & Whitehead, 1997). WTP indicates the amount of money that an individual is willing to pay to obtain some benefit and avoid some cost from a specific action so that the level of utility attained remains unchanged (Louviere et al., 2000, p. 61). It is defined as the ratio between the coefficient of other attributes in the utility function and that of the price attribute: b W ¼ a ; ð4Þ bc where ba represents coefficients of other attributes and bc coefficient of the price attribute. The WTP measurement is applied in this study to estimate the differential amount that respondents ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 315 are willing to pay for housing in different districts, of different types, in neighbourhoods characterized by different conveniences and accessibilities, etc.

Modelling impacts of individuals’ socio-economics on housing preference There are two approaches to model the impacts of individuals’ socio-economic characteristics on housing choice behaviour. Firstly, we may develop models for different socio-economic groups and compare the results. We may then conduct sensitivity analysis to see how each group react to change of key attribute in terms of choice probabilities. We may also calculate the WTP for some key variables and to ﬁnd out if there is any difference between groups. Alternatively, we may develop a single model incorporating the socio-economic variables (Walker et al., 2002). In this case, we need to construct choice sets for each individual. This turns out to be impractical to operate for this study because the data set is too large. We therefore choose the ﬁrst option and develop models for different socio-economic groups.

Findings

The choices given by the respondents were aggregated for each choice set. To facilitate estimation of the models, levels of the attributes were coded by an orthogonal coding scheme (Kerlinger & Pedhazur, 1973). A k-level attribute is coded by k 1 vectors. For continuous attributes such as price, these k 1 vectors represent different components of the attribute effect. For example, a two-level attribute is assumed to have only linear effects on housing choice and only one vector will be needed to code this attribute. On the other hand, a three-level attribute is assumed to have a non-linear effect that can be decomposed into two components: linear and quadratic. As a result, two vectors (representing the linear and quadratic effects, respectively) will be needed to code this attribute. Similarly, a four-level attribute will need three vectors (representing the linear, quadratic and cubic effects, respectively) to code. As for the case of discrete attributes such as dwelling orientation, the same coding scheme is employed, but the coding vectors are used to differentiate the utilities of different categories (similar to dummy coding) and do not represent the different components of the non-linear effect. The joint logit models were estimated for the entire sample and for different socio-economic groups. The models were used to compute utilities, choice probability impacts of attributes, and the WTP. Because of limitation of space, only the results for the entire sample are presented here. Table 2 lists the choice model. Coefficient estimates of the coding vectors are presented. To help readers understand the results, the utilities of attribute levels are computed from Table 2 and are given in Table 3. To compare the relative importance of various attributes, the impact of these attributes on choice probability are estimated using the choice model. The results are given in Table 4. Table 5 provides another angle to evaluate the relative importance of the attributes: the WTP estimates, which show in dollar terms how much the respondents would be willing to pay for an improvement of a certain neighbourhood or dwelling attribute. The results for various socio- economic groups, the full set of which is available upon request, are given in conjunction with our discussion of specific neighbourhood and dwelling attributes and of specific socio-economic factors. ARTICLE IN PRESS

316 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

Table 2 The choice model for the entire sample

Dwelling variables Coefﬁcient (sig.) Neighbourhood variables Coefﬁcient (sig.)

Price District Linear 0.133 (0.000) Yuexiu 0.376 (0.000) Quadratic 0.035 (0.082) Liwan 0.385 (0.000) Cubic 0.022 (0.014) Dongshan 0.544 (0.000) Haizhu 0.089 (0.130) Tianhe 0.381 (0.000) Baiyun 0.381 (0.000) Fangcun 0.323 (0.000) Huangpu 0.914 (0.000) Orientation Accessibility Linear 0.068 (0.000) Linear 0.268 (0.000) Quadratic 0.038 (0.062) Quadratic 0.029 (0.037) Cubic 0.094 (0.000) Types Living convenience Linear 0.018 (0.040) Linear 0.251 (0.000) Quadratic 0.070 (0.000) Quadratic 0.045 (0.001) Cubic 0.001 (0.899) Layout Security Linear 0.125 (0.000) Linear 0.340 (0.000) Layout Property management Model ﬁt Linear 0.001 (0.975) w2 1417.8 Property management Rho-square 0.054

Note: see the ﬁrst paragraph of Findings for the explanations of linear, quadratic and cubic.

All estimated choice models are highly significant, although the Rho-square (pseudo-R2) obtained, in line with most choice equation estimates, is not high. Most of the coefficient estimates are also highly significant. Below, we first discuss the housing preferences of the entire sample. We then analyse the differences between socio-economic groups.

Overall housing preference

Restricting ourselves to the equation for the entire sample, we see from Table 2 that, with the exception of property management, all neighbourhood (or location) and dwelling attributes under consideration are highly significant: every variable has at least one attribute level significant at po0:001. The coefficient estimates for the location attributes are generally much larger in magnitude than those for the dwelling attributes; the difference is in the order of 10. Thus, Guangzhou people tend to attach greater importance to neighbourhood attributes than dwelling attributes in choosing a place to live. This finding echoes the one previously found for residents of Beijing (Wang & Li, forthcoming). ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 317

Table 3 Utilities of attribute levels (for the entire sample)

Dwelling variables Utilities Neighbourhood variables Utilities

Price District 4000o 0.412 Yuexiu 0.376 4000–5000 0.164 Liwan 0.385 5000–6000 0.234 Dongshan 0.544 Haizhu 0.089 Tianhe 0.381 Baiyun 0.381 Fangcun 0.323 Huangpu 0.914 Orientation Accessibility East 0.072 Highly accessible 0.239 South 0.388 Reasonable 0.058 West 0.312 Limited 0.297 North 0.148 Types Living convenience Detached house 0.15 Very convenient 0.206 4 storey 0.085 Reasonable 0.090 5+ storey without lift 0.055 Not convenient 0.296 5+ storey with lift 0.125 Layout Security Small living, large bed rooms 0.125 Good public order 0.340 Large living, small bedrooms 0.125 Poor public order 0.340 Property management Yes 0.001 No 0.001

City districts As Table 2 shows, the district dummies generally exhibit high degree of statistical significance, and their coefficient estimates are large in magnitude. The utility figures in Table 3 show that respondents attach significantly different preferences or utilities to different districts of the city: the three inner core districts, namely Dongshan, Liwan and Yuexiu, have large utilities and are strongly preferred. Among the three, Dongshan District is the most preferred. Table 5 shows that, on average, Guangzhou residents are willing to pay an extra RMB 669 per m2 and RMB 633 per m2 (or roughly 20% of the average selling price of commodity housing of about RMB 3700 per m2 for the city in 2000; see SBGZ, 2001, p. 37), respectively, for a dwelling in Dongshan, as compared with buying one in Yuexiu and Liwan. Tianhe is the only district outside the inner core to which respondents attach strong preference, with utility yield almost identical to those given by Liwan and Yuexiu. In that sense Tianhe is now considered a constituent part of Guangzhou’s urban core. Haizhu is geographically and administratively considered an inner core district. Yet the utility attached to it is much lower than those of the three inner core districts on the northern bank of the Pearl River. Table 5 shows that a price premium of RMB 1143, 1179 and 1813 per m2, ARTICLE IN PRESS

318 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

Table 4 Impacts of Key Attributes on Choice Probability

ABEntire sample ðPB PAÞ=PA (%) Sub-samples by household Sub-samples by income nature of work unit

Low Medium High State Nonstate (%) (%) (%) (%) (%)

District Huangpu Dongshan 330 475 448 197 261 431 Liwan Fangcun 51 50 49 54 52 49 Yuexiu Dongshan 18 23 17 19 15 21 Price (1000/m2) 4 or less 4–5 22 21 25 19 17 26 4–5 5–6 33 25 24 45 28 37 5–6 6 or more 10 7 15 15 13 8 Accessibility Limited Reasonable 43 50 48 34 46 38 Reasonable Highly 20 32 15 19 19 22 accessible Living Not Reasonably 47 85 56 23 42 54 convenience convenient convenient Reasonably Very 12 14 19 5 21 4 convenient convenient Security Poor public Good public 97 99 83 111 122 76 order order Layout Small living, Large living 28 30 27 31 42 16 large bed room, small rooms bedrooms East South 37 32 28 49 42 34 South West 50 42 46 58 55 46 Orientation West North 18 31241343

respectively, is needed for the average Guangzhou resident to become indifferent between living in Yuexiu, Liwan and Dongshan, on the one hand, and in Haizhu, on the other. The negative views of Guangzhou people on Baiyun and Fangcun are demonstrated by the utility and WTP computations. As Table 5 shows, on average, dwellings in Dongshan commands a premium of RMB 3685 per m2 and RMB 3454 per m2, respectively, over those in Baiyun and Fangcun. Inaccessibility and rather undesirable living environments render Huangpu the least desirable district of Guangzhou. As Table 3 shows, the utility level obtained for this district (0.914) is much lower than that for Baiyun (0.381), the second lowest district. For reference, Dongshan commands a utlitity of 0.544. The probability impact ﬁgures given in Table 4 show that Guangzhou people are 330% more likely to choose to live in Dongshan than in Huangpu, all other things being equal. People are willing to pay an extra RMB 5809 per m2 to stay in Dongshan, as compared to moving to Huangpu (see Table 5). ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 319

Table 5 Willingness-to-pay (in terms of housing price RMB per square meters)

Yuexiu Liwan Dongshan Haizhu Tianhe Baiyun Fangcun Huangpu

A. District Yuexiu 0 36 669* 1143 20 3016 2785 5139 Liwan 36 0 633 1179 16 3052 2821 5175 Dongshan 669 633 0 1813 649 3685 3454 5809 Haizhu 1143 1179 1813 0 1163 1873 1641 3996 Tianhe 20 16 649 1163 0 3036 2805 5159 Baiyun 3016 3052 3685 1873 3036 0 231 2124 Fangcun 2785 2821 3454 1641 2805 231 0 2355 Huangpu 5139 5175 5809 3996 5159 2124 2355 0 B. Orientation of dwelling facing East South West North

East 0 1259 1530 876 South 1259 0 2789 2135 West 1530 2789 0 653 North 876 2135 653 0 C. Accessibility Highly Reasonably Limited accessible accessible accessibility

Highly accessible 0 721 2135 Reasonably 721 0 1414 accessible Limited 2135 1414 0 accessibility D. Convenience Very Reasonably Not convenient convenient convenient

Very convenient 0 462 2000 Reasonably 462 0 1538 convenient Not convenient 2000 1538 0 E. Type of dwelling Detached 4 storey 5 storey 5 storey house building building no building lift with lift

Detached house 0 936 817 100 4storey building 936 0 120 837 5storey building 817 120 0 717 no lift 5storey building 100 837 717 0 with lift

Note: *Potential home buyers would be willing to pay 669 RMB Yuan higher price for a house in Dongshan than in Yuexiu. ARTICLE IN PRESS

320 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

Computation of the WTP, if conducted at all points in space, generates the loci of the urban bid price function for housing (Alonso, 1966; Arnott, 1987). Under competitive equilibrium, they also give the market housing price function. The above discussion shows that in a fully developed market, given the strong preference for the core districts, the housing price gradient (and hence the urban land rent gradient) in Guangzhou would be quite steep. It falls off quite rapidly outside the urban core. However, it is highly unlikely that the housing price and land rent contours would consist of a series of concentric circle. Factors such as administration boundary, alignment of the Pearl River, government policy especially the Tianhe development plan, and ingrained perception of Guangzhou people (such as the perception of Fangcun as a backward agricultural region) all have an impact on the preference structure and hence the shape of the bid price function. The end product is likely to be a highly elongated oval-shape housing price and land rent contours, probably with multiple maxima at the major business centres in the three inner core districts and in Tianhe, and spurring towards the east downstream along the Pearl River.

Other neighbourhood attributes The other neighbourhood attributes, including accessibility to public transport, shopping convenience and neighbourhood security, are also signiﬁcant and have relatively large coefﬁcient estimates. Table 5 shows that respondents are willing to pay, on average, RMB 721 per m2 and RMB 2135 per m2 extra in order to trade a dwelling with only ‘‘limited accessibility’’ of public transport, respectively, for one with ‘‘reasonable accessibility’’ and one with ‘‘high accessibility’’. Similarly, they are willing to pay RMB 462 per m2 extra in order to trade a dwelling with ‘‘not convenient’’ shopping to one with ‘‘reasonably convenient’’, and a further RMB 1538 per m2 for one with ‘‘highly convenient’’ shopping. Neighbourhood security is of particular importance. A price difference of RMB 2709 per m2 is needed in order to make the average Guangzhou resident indifferent as to choosing a dwelling with ‘‘bad public order’’ and one with ‘‘good public order’’.

Price Price inﬂuences the utility level attained through its effect on the budget constraint. Table 4 shows that the choice probability for Guangzhou people is quite price-sensitive. An increase in the price of dwelling from RMB 4000 per m2 or less to RMB 4–5000 per m2 brings about a 22% reduction in the probability of purchase; further increases from RMB 4–5000 per m2 to RMB 5–6000 per m2, and from RMB 5–6000 per m2 to over RMB 6000 per m2 result in additional reductions of 33% and 10%, respectively.

Other dwelling attributes Three physical dwelling attributes, namely dwelling orientation, type of dwelling, and dwelling layout, are under examination. In line with the results previously obtained for Beijing, south- facing dwellings are the most favoured, given the prevailing wind directions, and given the relative ease of sunlight penetration for south-facing dwellings. Because of the heat brought by the afternoon sun, west-facing dwellings are the least preferred. Table 5 suggests that a price reduction of RMB 2789 per m2 is needed in order to make a person indifferent between residing in a south-facing dwelling and a west-facing residence. As for type of dwelling, ‘‘detached house’’ and ‘‘buildings of ﬁve-storeys or more with elevator’’ are favoured over other types of dwelling. Respondents are willing to pay a price of RMB 837 per m2 higher to buy an apartment in a ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 321 building of ‘‘ﬁve-storey plus with lift’’ than that of ‘‘four-storey or less’’ (Table 5). Table 4 shows that a change of layout from ‘‘small living room and large bedrooms’’ to ‘‘large living room and small bedrooms’’ will increase the choice probability by 28%. The willingness-to-pay data indicate that the potential homebuyers were willing to pay an extra RMB 996 per m2 for the latter (Table 5).

Impacts of selected socio-economic factors and residential experience

Household income Income affects the utility derived from housing consumption mainly through the budget constraint. But people with different incomes also have different life experiences. In Guangzhou and other cities in China the bicycle and the bus are the major transport modes for low-income people. The relative immobility of these people very much curtails their residential location choice. The findings show that the low-income group in Guangzhou, in comparison with the middle and high-income groups, is much more reluctant to choose the newly developed areas, including the new core district, Tianhe. On the other hand, the high-income group, which can readily afford the Metro or even the private car, are more willing to move to the suburbs. In fact, their preference for Tianhe (utility=0.777) is almost identical to that for Dongshan (utility=0.772), traditionally the most reputed district. The differential reliance on public transport is demonstrated by the results on accessibility to public transport. For the low-income group, an improvement from ‘‘limited’’ to ‘‘reasonable’’ accessibility increases the choice probability by 50%; for the high-income group, the increase is only 34% (Table 4). Differential mobility also affects the preference for shopping convenience. For the low-income group, an improvement from ‘‘not convenient’’ to ‘‘reasonably convenient’’ increases the choice probability by 85%. For the high-income group, this only increases the choice probability by 23%. Income also affects the preference for dwelling attributes. In particular, while ‘‘property management’’ is not significant for the overall equation and for the other equations, it is significant at po:005 for the high-income equation. Perhaps this is because the high-income group values more on the better residential environment and the enhanced value of the property that property management may render. The high-income group also shows greater concern for dwelling orientation: they especially dislike west-facing dwellings. The utility difference between south and west facing is equal to 0.870. For the low-income group, the difference is only 0.534.

Education Education is closely correlated with income. But education, which directly moulds a person’s preferences, has an effect that goes beyond its effect through income. In a redistributive economy, which still lingers in many sectors in China, a person’s position in the job hierarchy, and hence housing experience under the danwei system, is particularly tied to his/her education background (Szelenyi, 1987). The inﬂuence of education on residential preference is therefore strong. The estimated choice equations show systematic variations across education groups. In terms of district preference, people with junior secondary education or below have a strong preference for the three inner core districts, Yuexiu, Liwan and Dongshan. They are willing to pay rather high premiums to reside in the inner core. To them, the newly added core district Tianhe is not a favoured choice; they view Tianhe more or less on par with Haizhu. On the other hand, for those ARTICLE IN PRESS

322 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 with college or above education, Tianhe is the most favoured. This group is willing to pay premiums of RMB 1112 per m2, RMB 1699 per m2, and RMB 762 per m2, respectively, to trade a dwelling in Yuexiu, Liwan and Dongshan for one in Tianhe.

Age Studies on residential decisions and mobility invariably point to the importance of age and hence the personal and family life cycle as a factor governing the residential location and relocation process (Clark, 1982). The lack of cases restricts us from adopting a more refined classification. Only two age groups, namely, ‘‘40 or below’’ and ‘‘older than 40’’, are examined. Probably because of the rather coarse classification employed, the findings generally fail to reveal major age differences in the preference for districts. The only exception is Tianhe: the younger age group has a relatively high preference for this district (utility=0.574, which is the second highest among all districts), whereas the older age group has a relatively low preference for it (utility=0.116, which ranks four among the eight districts). Again, this shows that the younger age group is more prepared to trade more familiar surroundings for an environment endowed with better amenities. On the other hand, the older group shows greater concern for neighbourhood familiarity and hence security. They are willing to pay a premium of RMB 4550 per m2 to secure a building with ‘‘good public order’’ over one with ‘‘poor public order’’; in comparison, the younger group is only willing to pay a premium of RMB 1886 per m2 for this purpose.

Nature of employment organization China is a transitional economy. While the market and the private sector are playing an increasing role, state work units or danwei remain to be important job providers. It may be hypothesized that because of their rather different housing experiences, workers in state work units and workers in other organizations have quite different housing preferences. However, the ﬁndings show that the two groups generally exhibit rather similar preferences. But they do differ in some respect to a number of attributes. People working in state work units, who are used to the secluded environment of danwei compounds, are more concerned with neighbourhood security. They are willing to pay a premium of RMB 3635 per m2 to trade a dwelling in a neighbourhood with ‘‘poor public order’’ with one in a neighbourhood with ‘‘good public order’’. People working in non-state sectors are willing to pay a much smaller premium—RMB 1993 per m2—for the same trade off. Moreover, workers in state work units, probably tired of the dull and uniform danwei housing blocks, are more concerned with the internal design of the building: they are willing to pay an extra RMB 1616 per m2 to secure a dwelling with ‘‘large living room and small bedrooms’’ against one with ‘‘small living room and large bedrooms’’. In comparison, workers in other work organizations are willing to pay a premium of only RMB 516 per m2 for this purpose. In addition to the above, the lack of market experience for the state work unit group has resulted in their somewhat insensitivity to price change. Table 4 shows that for workers in state work units, an increase in price from RMB 4000 per m2 or less to RMB 4–5000 per m2 reduces the choice probability by 17%; for workers in non-state work units, the reduction is 26%. Similarly, a price increase from RMB 4–5000 per m2 to RMB 5–6000 per m2 yields a corresponding reduction of 28% and 37%, respectively for the state and non-state groups. Perhaps workers in state work ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 323 units have to learn more about how the market operates before they can make rational choices in a fully marketized environment.3

Current residential location People learn from experience and acquire information about the environment. Thus, the current residence would have an inﬂuence on future housing choice. We partition the sample according to whether the current residence is located in one of the core districts, i.e., Yuexiu, Liwan, Dongshan and Tianhe. The ﬁndings show that residents of the core districts generally have a greater dislike of the non-core districts; conversely residents of the non-core districts show less strong preference for the core districts. To illustrate, core districts residents are willing to pay premiums of RMB 4926 per m2, RMB 4362 per m2 and RMB 6802 per m2, respectively, for a dwelling in Dongshan over one in Baiyun, Fangcun and Huangpu. Residents of the non-core districts are only willing to pay corresponding premiums of RMB 1951 per m2, 2081 per m2 and 4356 per m2.

Conclusions

We employed state-of-the-art experimental design methods to develop choice experiments for a sample of 1500 household heads in Guangzhou, China. We estimated joint logit models for the entire sample and for various sub-samples classified by family income, age, education, nature of employment organizations, and district of current residence. We then used the models to compute utilities for attribute levels, the impacts on the choice probability of attributes, and the relative housing prices that respondents were willing to pay for buying a house in different districts, of different accessibility, and of different types, etc. We revealed that people in Guangzhou attach greater importance to neighbourhood- and location-related attributes than to dwelling-related attributes when considering buying a home. A similar finding was formerly obtained for people in Beijing (Wang & Li, forthcoming). Reputed districts are strongly favoured over other districts. Potential homebuyers in Guangzhou are willing to pay a price difference up to RMB 5000 per m2 to trade a dwelling in the least preferred district for one in the most reputed one. People in Beijing, too, have similarly strong preference for reputed districts. However, unlike the case of Guangzhou in which there is uniformly strong preference for the inner core districts, in the case of Beijing not all inner core districts are preferred (Wang & Li, forthcoming). In addition to preference for district, potential homebuyers in Guangzhou, like those in Beijing, also place great emphasis on the quality of neighbourhood in terms of security image, accessibility and convenience. These facts suggest that despite the differences between Beijing and Guangzhou in terms of lifestyle, extent of marketization, nature of employment organization (more people in Beijing work in state work units), etc. the structures of housing preference in these two cities are quite similar. This is probably due to the strong influence of the central government through its top-down approach in implementing policies,

3In order to sharpen our analysis on danwei Vs non-danwei housing experience, we have also partitioned the sample according to current residence in danwei Vs non-danwei housing. The ﬁndings are very similar to those reported in the text pertaining to the state Vs non-state work unit partition. ARTICLE IN PRESS

324 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 organizing the society and educating people. Consequently, people in different places of the country develop quite similar tastes and preferences. Socio-economic factors were found to be influential in housing preference formation for the urban Chinese. In comparison with the high-income group, the low and medium income groups show stronger preference towards the inner core districts and place more importance on living convenience and accessibility to public transport. The high-income group, however, is more willing to move to outlying district and pay more attention to the quality of dwelling such as orientation. Similarly, people with less education attainment show stronger preference towards the inner core districts. While the age effect is generally not strong, the findings showed that the younger age group particularly favours Tianhe, the suburb that contains the new CBD. These findings have an important implication regarding future urban morphology. Based on them, we may envision a residential pattern not very dissimilar from the ones in the West to emerge in Guangzhou, with the inner core dominated by the urban poor, while the outskirts especially districts with good reputation and congenial natural environment gradually occupied by the rich and well-educated people. Somehow unique to the case of China, nature of people’s employment organization was found to be an important factor structuring housing preference. People in state work units are less price- sensitive than those in non-state work units. This is probably because state work unit workers usually do not have market experience as their housing is in most cases either provided or subsidized by the work unit. Probably because of their previous residential experience, people working in state work units are generally more concerned with neighbourhood security and the internal design of the building. This finding suggests that institutional forces in China not only affects housing choice behaviour through imposing constraints and direct intervention of the housing allocation system, but also act as a structuring factor on the individuals’ housing preference.

Acknowledgement

The authors would like to acknowledge the Centre for Urban and Regional Studies of Zhongshan University for the assistance in the conduct of the choice experiment. They would also like to thank Jiukun Li, Doris Fung and Fion Law for their assistance in preparing and processing the data. This study is sponsored by two Hong Kong Research Grant Council (RGC) projects: HKBU 2018/00 H and HKBU 2080/99 H.

References

Alonso, W. (1966). Location and land use. Cambridge, Mass: Harvard University Press. Arnott, R. (1987). Economic theory and housing. In E. S. Mills (Ed.). Handbook of regional and urban economics, Vol. II (pp. 959–988). Amsterdam: Elsevier Science. Ben-Akiva, M., & Lerman, S. R. (1985). Discrete choice analysis: Theory and application to travel demand. Cambridge, MA: MIT Press. Clark, W. A. V. (1982). Recent research on migration and mobility: A review and interpretation. Pregress in Planning, 18(1), 1–56. ARTICLE IN PRESS

D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326 325

Clark, W. A. V., & Dieleman, F. M. (1996). Households and housing: Choice and outcomes in the housing market. New Brunswick, NJ: Center for Urban Policy Research, Rutgers, The State University of New Jersey. Earnhart, D. (2002). Combining revealed and stated data to examine housing decisions using discrete choice analysis. Journal of Urban Economics, 51, 143–169. Evans, A. (1971). On the theory of the valuation and allocation of time. Scottish Journal of Political Economy, 2, 1–17. Fischer, M. M., & Aufhauser, E. (1988). Housing choice in a regulated market: A nested multinomial logit analysis. Geographical Analysis, 20, 47–69. Freeman, A. M., III (1991). Factorial survey methods and willingness to pay for housing characteristics: A comment. Journal of Environmental Economics and Management, 20, 92–96. Friedman, A. M. (1981). A conditional logit model of the role of local public services in residential choice. Urban Studies, 18, 347–358. Goodman, A. C. (1989). Identifying willingness-to-pay for heterogeneous goods with factorial survey methods. Journal of Environmental Economics and Management, 16, 58–79. Guangzhou Nianjian Bianzuan Weiyuanhui (2001). Guangzhou Nianjian 2001. Guangzhou: Guangzhou Nianjian Chubanshe. Hensher, D. A., & Truong, T. P. (1984). Valuation of travel time savings from a direct experimental approach. Journal of Transport Economics and Policy, 19(3), 237–261. Ho, M. H. C., & Kwong, T. -M. (2002). Housing reform and home ownership behaviour in China: A case study in Guangzhou. Urban Studies, 17(2), 229–244. Huang, Y., & Clark, W. A. V. (2002). Housing tenure choice in transitional urban China: A multilevel analysis. Urban Studies, 39(1), 7–32. Huang, J-C., Haab, T. C., & Whitehead, J. C. (1997). Willingness to pay for quality improvements: Should revealed and stated preference data be combined. Journal of Environmental Economics and Management, 34, 240–255. Kerlinger, F. N., & Pedhazur, E. J. (1973). Multiple Regressions in Behavioral Research. New York: Holt, Rinehart and Winston. Lau, K. M. (2003). Housing Affordability in Beijing from 1992 to 2001: Application of the price-to income ratio. Honours project thesis. Hong Kong: Department of Geography, Hong Kong Baptist University. Li, S. -M. (2000). The housing market and tenure decision in Chinese cities: A multivariate analysis of the case of Guangzhou. Housing Studies, 15(2), 213–236. Li, S. -M. (2003). Housing tenure and residential mobility in urban China: Analysis of survey data. Urban Affairs Review, 38(4), 510–534. Li, S. -M. (2004). Life course and residential mobility in Beijing, China,’’ Environment and Planning A, 36(1), 27–43. Li, S. -M., & Chu, D. K. Y. (1987). The implications of system reforms for urban land uses and city landscape. In J. Chai, & C. K. Leung (Eds.). China’s Economic Reforms (pp. 248–264). Hong Kong: Centre of Asian Studies, The University of Hong Kong. Li, S. -M., & Siu, Y. -M. (2001a). Commodity housing construction and intra-urban migration in Beijing: An analysis of survey data. Third World Planning Review, 23(1), 39–60. Li, S. -M., & Siu, Y. -M. (2001b). Residential mobility under market transition: A study of the newly constructed commodity housing of Guangzhou. The Professional Geographer, 53(2), 219–229. Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated Choice Methods: Analysis and application. Cambridge: Cambridge University Press. McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. In P. Zarembca (Ed.). Frontiers in Econometrics (pp. 105–142). New York: Academic Press. McFadden, D. (1998). Measuring willingness-to-pay for transportation improvements. In T. Gorling, T. Laitila, & K. Westin (Eds.). Theoretical Foundations of Travel Choice Modelling (pp. 339–364). Amsterdam: Elsevier Science. Megbolugbe, I. F. (1991). Hedonic prices and housing programme beneﬁts. Urban Studies, 28, 773–781. Palmquist, R. (1984). Estimating the demand for characteristics of housing. Review of Economics and Statistics, 64, 394–404. Quigley, J. M. (1976). Housing demand in the short run: An analysis of polytomous choice. Explorations in Economic Research, 3, 76–102. ARTICLE IN PRESS

326 D. Wang, S.-m. Li / Habitat International 30 (2006) 305–326

Rosen, S. (1974). Hedonic prices and implicit markets; product differentiation in perfect competition. Journal of Political Economy, 82, 34–55. Statistical Bureau of Guangzhou City (SBGZ) (2001). Guangzhou Statistical Yearbook 2001. Beijing: China Statistics Press. Statistical Bureau of Guangdong Province (SBGP) (2002). Guangdong Statistical Yearbook 2002. Beijing: China Statistics Press. Szelenyi, I. (1983). Urban Inequalities under State Socialism. Oxford: Oxford University Press. Szelenyi, I. (1987). Housing inequalities and housing segregation in state socialist countries. International Journal of Urban and Regional Research, 11, 1–8. Timmermans, H. J. P., & van Noortwijk, L. (1995). Context dependencies in housing choice behaviour. Environment and Planning A, 27, 181–192. Tu, Y., & Goldﬁnch, J. (1996). A two-stage housing choice forecasting model. Urban Studies, 33(3), 517–537. Walker, B., Marsh, A., Wardman, M., & Niner, P. (2002). Modelling tenants’ choices in the public rented sector: A stated preference approach. Urban Studies, 39(4), 665–688. Wang, D., & Li, J. (2002). Handling large numbers of attributes and/or large numbers of levels in conjoint experiments. Geographical Analysis, 34(4), 350–362. Wang, D., Li, S.-M. (2004). Housing preferences in a transitional housing system: the case of Beijing, China. Environment and Planning A, 36(1), 69–87. Wang, D., Borgers, A., Oppewal, H., & Timmermans, H. J. P. (2000). A stated choice approach to developing multi- faceted models of activity behaviour. Transportation Research A, 34(8), 625–643. Wu, F. (1996). Changes in the structure of public housing provision in urban China. Urban Studies, 33(9), 1601–1627. Xu, X., Hu, H., & Yeh, A. G. O. (1989). Guangzhou Shi shehui kongjian jiekou de yinzhi shengtai fenxi (A factorial ecology analysis of the social space of Guangzhou). Acta Geographica Sinica, 44(4), 385–399.