Content uploaded by Kevin Lo
Author content
All content in this area was uploaded by Kevin Lo on Oct 06, 2018
Content may be subject to copyright.
Chin. Geogra. Sci. 2017 Vol. 27 No. 5 pp. 684–696 Springer Science Press
doi: 10.1007/s11769-017-0902-x www.springerlink.com/content/1002-0063
Received date: 2016-10-26; accepted date: 2017-02-24
Foundation item: Under the auspices of National Natural Science Foundation of China (No. 41301143)
Corresponding author: SONG Wei. E-mail: wei.song@louisville.edu
© Science Press, Northeast Institute of Geography and Agroecology, CAS and Springer-Verlag Berlin Heidelberg 2017
Spatial Patterns of Car Sales and Their Socio-economic Attributes in
China
LIU Daqian1, LO Kevin2, SONG Wei3, XIE Chunyan4
(1. Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, Changchun 130102, China; 2. Department of
Geography, Hong Kong Baptist University, Hong Kong 999077, China; 3. Department of Geography and Geosciences, University of
Louisville, Louisville KY 40292, USA; 4. FAW-Volkswagen Automotive Company Ltd., Changchun 130061, China)
Abstract: Using data from the Economic Advisory Center of the State Information Center (SIC), we examined the spatial patterns of car
sales in China at the prefectural level in 2012. We first analyzed the spatial distributions of car sales of different kinds of automakers
(foreign automakers, Sino-foreign joint automakers, and Chinese automakers), and then identified spatial clusters using the local
Moran’s indexes. Location quotient analysis was applied to examine the relative advantage of each type of automaker in the local mar-
kets. To explain the variations of car sales across cities, we collected several socioeconomic variables and conducted regression analy-
ses. Further, factor analysis was used to extract independent variables to avoid the problem of multicollinearity. By incorporating a spa-
tial lag or spatial error in the models, we calibrated our spatial regression models to address the spatial dependence problem. The ana-
lytical results show that car sales varied significantly across cities in China, and most of the cities with higher car sales were the devel-
oped cities. Different automakers exhibit diverse spatial patterns in terms of car sales volume, spatial clusters, and location quotients.
The scale and incomes factor were extracted and verified as the two most significant and positive factors that shape the spatial distribu-
tions of car sales, and together with the spatial effect, explained most of the variations of car sales across cities.
Keywords: car sales; spatial clusters; Location Quotient; socio-economic attributes; China
Citation: Liu Daqian, Lo Kevin, Song Wei, Xie Chunyan, 2017. Spatial patterns of car sales and their socio-economic attributes in
China. Chinese Geographical Science, 27(5): 684–696. doi: 10.1007/s11769-017-0902-x
1 Introduction
International experiences suggest that the proliferation
of automotive sales typically occurs after the economy
of a nation or a region reaches certain levels (Dargay
and Gately, 1999). As incomes increase, the propensity
to purchase auto vehicles increases as they become more
affordable. It is therefore no surprises that, China, one of
the world’s largest and fastest developing economies, is
experiencing a rapid expansion of automobile ownership
and has become one of the largest markets for car sales.
In particular, after the entry to WTO, Chinese automo-
bile markets are becoming more open and competitive.
Car manufacturers from all over the world are increas-
ing their car production and car sales in China. Accord-
ing to the statistics from the China Association of
Automobile Manufacturers (CAAM), China’s automobile
sales in 2015 reached 2.46 × 107, extending the country’s
lead as the world’s largest car market by sales. It is fore-
seeable that, as China’s economy continues to grow, the
number of high-income urban households will expand
greatly, and the infrastructure such as high quality roads
will be improved significantly in the next decades.
Therefore, the market for car sales is expected to maintain
strong growth momentum (Huo and Wang, 2012).
Such rapid growth of car sales has created many
LIU Daqian et al. Spatial Patterns of Car Sales and Their Socio-economic Attributes in China 685
challenges. In addition to problems associated with traf-
fic jams, car use also has implications on local and
global environmental issues such as air pollution, energy
security, energy scarcity, and climate change (Lo, 2014;
Wang et al., 2014). Hence, the phenomenon and its pol-
icy implication need to be analyzed in more details.
Since China has a vast territory with significant regional
heterogeneity, car sales are likely to vary across cities
due to the diverse economic capacities. It is also inter-
esting and meaningful to explore the spatial differences
of car sales across cities because car consumption, as a
common and important consumer durable, is to a certain
degree indicative of the consumption capacity of the
cities.
Nowadays, as the market economy develops, nearly
all the markets of the cities are full of competition from
different car manufacturers, after which various market
structures are shaped across cities. What is the spatial
pattern of the car sales across the nation? What are the
differences with respect to spatial patterns of cars sales
of different manufacturers, such as foreign automakers,
Sino-foreign joint automakers and Chinese automakers?
What about the differences regarding the market shares
of different kinds of manufacturers? What factors could
exert impacts on car sales across cities? The answers to
these questions would provide helpful insights into un-
derstanding the car consumption process, and would be
valuable for relevant policymakers to formulate indus-
trial policies for the automotive sector. This study aims
to address these questions using the data of actual car
sales at the prefectural level, obtained from the Eco-
nomic Advisory Center of the State Information Center
(SIC). The data were analyzed using location quotient,
spatial cluster analysis, factor analysis as well as regres-
sion analysis to illuminate the spatial patterns and socio-
economic associations of car sales among Chinese cities.
2 Literature Review
Many subjects regarding to the production and manu-
facturing of automobile, such as the automobile pur-
chasing, car ownership, and automobile industry, have
been studied extensively over the years in the developed
countries, such as the United States, Great Britain,
Spain, Japan and South Korea (Bandeen, 1957; Ruben-
stein, 1986; Pallares-Barbera, 1998; Giuliano and Dar-
gay, 2006; Clark and Finley, 2010; Lee, 2011). As their
economy grows, more and more developing countries,
especially China, are consuming an increasing number
of automobiles. Thus, researchers are focusing more on
car consumption, automobile industry, and socioeco-
nomic attributes of car ownership in China. There is a
burgeoning literature concerning car consumption in
China (Rits et al., 2004; Qian and Soopramanien, 2014;
2015; Wan et al., 2015). However, most of these studies
focus on the overall characteristics or prediction of sales
of cars (or some particular type of cars, such as green
car), with few focusing on the spatial distribution of car
sale across Chinese cities and the spatial pattern of car
markets shared by different car manufacturers. The
studies on car sales or car markets in China from a spa-
tial perspective are limited, and are mostly conducted at a
macro level, such as national, regional or provincial di-
mensions (Xiao, 2010; Xie and Liu, 2014). Geographical
researchers in China are mainly interested in the produc-
tion and manufacturing of automobiles, as opposed to
their consumption. Therefore, they focus on the devel-
opment of automotive industry in the country, such as the
spatial pattern of car manufacturers or auto industrial
cluster (Liu and Yeung, 2008; Wang and He, 2009; Liu
and He, 2011; Zhao et al., 2014), areal division of auto
industrial chain (Li and Gu, 2010) and spatial evolution of
auto industry (Huang and Zhang, 2014). The reason why
studies on spatial pattern of car sales are scarce is likely
that the data of car sales at such a middle or micro level are
usually not accessible in China. Nevertheless, as argued
previously, it is interesting and meaningful to explore the
spatial differences across cities in terms of car sales.
The purchase of a car is an important indicator of car
use and changes to travel behavior (Clark et al., 2015).
Existing studies have identified several factors affecting
car ownership and purchases. The factors that affect car
ownership and sales include individual and household
income, employment status, ethnic background, house-
hold location, demographic change, car price, access to
and quality of public transportation, parking supply,
population density, parking price, specific life cycle
events (such as employment change and childbirth), and
financing availability (Cullinane, 2002; Guo, 2013;
Ritter and Vance, 2013; Oakil et al., 2014; Clark et al.,
2015; Ling et al., 2015; Oakil et al., 2016; Seya et al.,
2016; Wu et al., 2016; Yagi and Managi, 2016). Consid-
erable discrepancies exist on the relative importance of
these factors that appear to be spatially differentiated
686 Chinese Geographical Science 2017 Vol. 27 No. 5
and are dependent on local contexts. For example, Ling
et al. (2015) found that China’s car ownership is closely
related to household characteristics such as household
income and household size vis-à-vis environmental fac-
tors. In contrast, using a household survey in the New
York City region, Guo (2013) argued that the influence
of parking supply on household car ownership outper-
forms household income and demographic characteris-
tics. These differences suggest that it is important to
understand the spatial patterns by conducting city-level
comparison studies of car sales data, as opposed to
studies that are focused on a single city using house-
hold-level data or using aggregated data at the regional
or national levels (Cao and Huang, 2013). This research
is conducted based on the actual car sales data at the
prefectural cities level, which can be deemed as a valu-
able empirical reference for further recognitions towards
the spatial distribution of car sales across China.
3 Data and Method
3.1 Data and research area
This study focuses on passenger cars, which can be
classified into three types: sedans, multi-purpose vehi-
cles (MPV), and sport-utility vehicles (SUV). Car sales
data used in our research are from the Economic Advi-
sory Center of the SIC. It is composed of the actual
amount of car sales of each manufacturer in each pre-
fectural city in China (excluding Taiwan, Hong Kong,
and Macao) in 2012. This study includes 337 prefectural
cities as the basic analytical units. Other data used in our
research, such as GDP, population, and other socioeco-
nomic indicators, are from the China Statistical Year-
book for Regional Economy 2012. The data contain 107
manufacturers that sold cars in China in 2012, compris-
ing 47 foreign-owned automakers, 24 Sino-foreign joint
automakers, and 36 Chinese automakers.
3.2 Spatial cluster analysis
To capture the local city clusters with higher car sales,
we employed local Moran’s I to detect the hotspots of
car sales of different manufacturers. As one of the most
common local indicators of spatial association (LISA),
local Moran’s I has been applied to identify statistically
significant local clustering of specific phenomena in
many fields of research. In this study, local Moran’s I
efficiently measures the geographical concentration of
car sales clusters by detecting spatial autocorrelation
among cities in China. The local Moran’s I statistic was
derived for each city area and defined as:
,
2
1,
()
n
i
iijj
jji
i
xX
I
xX
S
(1)
where xi is the car sales in city i, X is mean of the car
sales, ,ij
is the spatial weight between city i and j, and
2
1,
2
()
1
n
j
jji
i
xX
SX
n
(2)
where n equals the total number of cities in China
(Anselin, 1995). The significance of the local Moran’s I
statistic is often tested by a normal Z index (Anselin,
1995; Wang, 2006). When the local Moran’s I statistics
are positive and significant (Z > 1.96), we can label the
corresponding areas as ‘hotspots’ (clustering of high
values/High-High) or ‘coldspots’ (clustering of low val-
ues/Low-Low). When the Local Moran’s I statistics are
negative and significant (Z<–1.96), we can label the
corresponding areas as outlier spots that can be further
divided into two types: convex spots (high values sur-
rounded by low values neighbors/High-Low) and con-
cave spots (low values surrounded by high values
neighbors/Low-High). This study focused on all these
spatial relations. Hotspots refer to the cities with higher
car sales values surrounded by neighbors whose car
sales are also above the average, and coldspots refer to
the cities with lower car sales values surrounded by
neighbors whose car sales are below the average. Con-
vex spots refer to the cities with higher car sales sur-
rounded by neighbors whose car sales are below the
average, and concave spots refer to cities with lower car
sales values surrounded by higher car sales values above
the average. We applied cluster and outlier analysis-
(Anselin Local Moran’s I) tool in ArcGIS 9.3, which is
built upon the above equations, to compute local
Moran’s I statistics for all the cities. The spatial weights
matrix that we used to define the spatial relationships is
calculated based on the inverse distance.
3.3 Location quotient
In addition to the general spatial pattern of car sales
across cities in China, we are also interested about the
relative advantage of a manufacturer in each city’s mar-
LIU Daqian et al. Spatial Patterns of Car Sales and Their Socio-economic Attributes in China 687
ket. After long and intensive competition, a city’s mar-
ket is usually shared by many automakers. If a manu-
facturer obtains a larger market share in a city than the
average market share it has across the cities in the whole
country, this automaker has obtained a relative advan-
tage in this city. To measure the relative advantage, we
computed the location quotient as follows:
LQi,j=,/
/
ij j
i
QT
QT
(3)
where LQi, j is the location quotient of manufacturer i in
city j and Qi,j is the amount of car sales of manufacturer i
in city j. Tj represents the total amount of car sales in
city j. Qi is the total amount of car sales of manufacturer
i, and T is the total car sales of all the manufacturers in
the country.
3.4 Factor and regression analyses
To interpret the spatial patterns of car sales, the multi-
variate linear regression was conducted to explore the
socioeconomic factors that may be associated with the
spatial variation. Based on the data from the China Sta-
tistical Yearbook for Regional Economy 2012, a set of
explanatory variables were selected assuming that they
were related to car sales in cities. Theoretically, the car
sales could be related with many factors, such as popu-
lation size, GDP, residents’ income, comprehensive
consumption capacity, investment on infrastructure, etc.
Because these variables are usually highly correlated,
we used the factor analysis to arrive at fewer and more
manageable independent factors from these variables.
The factor extraction method that we used was the de-
fault principal component solution to estimate the coef-
ficients. To simplify the interpretation of factors, the
orthogonal rotation was made to achieve a simple struc-
ture that each factor had large loadings in absolute value
for only some of the variables. The factor analysis was
conducted using the software SPSS 16.0, through which
factor scores for each city could be obtained simultane-
ously. Thus, several regression models could be cali-
brated using fewer but uncorrelated components as in-
dependent variables, and car sales from different kinds
of manufacturers as dependent variables. The regression
model can be described as follows:
0ikiki
k
yx
(i=1, 2, …, n) (4)
where yi is the dependent variable, α0 is the constant, βk
is the regression coefficient for the independent variable
k, εi is the random error or residuals, n is the number of
cases, and k is the number of independent variables.
Since the car sales data had spatial attributes, the spa-
tial dependence problem may exist when we conduct
regression analysis, especially when hotspots and cold-
spots were identified in the process of spatial cluster
analysis. To address this issue, we used GeoDa 1.8.8 to
evaluate spatial dependency in the regression (Anselin,
2005). The first-order queen’s contiguity was used to
specify the spatial weights matrix for car sales and the
Moran’s I (error) as well as Lagrange multiplier statis-
tics (LM lag and LM error) were calculated. The multi-
variate models were estimated using the maximum like-
lihood method. The models and spatial regression proc-
ess can be referred to the work of Wang (2006).
4 Results
4.1 Spatial distribution of car sales
According to the data we collected, the total amount of
car sales in China in 2012 is 1.31 × 107, with 1.04 × 106
cars produced and imported by foreign-owned auto-
makers (7.88%), 8.85 × 106 by Sino-foreign joint ven-
tures (67.71%), and 3.19 × 106 by Chinese automakers
(24.41%). From the descriptive statistics shown in Table
1, there is a great variation of car sales across cities in
China for each type of manufacturer. The maps in Figs.
1(a–d) capture the spatial characteristics of these varia-
tions; we can clearly see the significant heterogeneity in
the spatial pattern of car sales.
Table 1 Descriptive statistics of car sales among different automakers across cities
Variables Maximum Minimum Average SD
Total car sales amount 539494 353 38798 57402.72
Foreign-owned automakers 77356 14 3072 6952.34
Sino-foreign joint automakers 394192 200 26256 40628.94
Chinese automakers 89125 103 9469 11531.60
688 Chinese Geographical Science 2017 Vol. 27 No. 5
Fig. 1 Car sales for total automakers (a), foreign automakers (b), Sino-foreign joint automakers (c), and Chinese automakers (d)
In Fig. 1a, the minimum number of car sales is only
353 in Shennongjia forestry district, but the maximum
reaches 539 494 in Beijing City. The huge gap in terms
of car sales among cities was definitely attributed to the
diverse consumption capacities. The distribution of total
car sales is consistent with China’s economic map to a
large extent: car sales are higher in more developed cit-
ies, such as in the Yangtze Delta Region, Bei-
jing-Tianjin-Hebei (Jing-Jin-Ji) Region, Pearl River
Delta Region, and capital cities in each province. Figs.
1b and 1c show the spatial distributions of car sales for
foreign-owned and Sino-foreign joint automakers, re-
spectively. Car sales of the imported foreign cars, which
were more expensive, were highly concentrated in the
most developed core cities, such as the capital cities and
the largest cities in each province. These cities play a
central role in the economic development of each prov-
ince and represent the highest consumption capacities in
China. The car sales of Sino-foreign joint automakers
were also more concentrated in the developed cities, but
some differences in the spatial pattern were also notable,
such as the relatively higher car sales in Shandong Pen-
insula. In contrast, the car sales of Chinese automakers
show a very different and more scattered pattern, as
shown in Fig. 1d. The second- and third-tier cities, es-
pecially in the North and Northeast China, were very
important target markets for Chinese automakers, al-
though the first-tier cities still recorded relatively higher
car sales. This implies that the Chinese automakers fo-
cused more on the middle and small cities where they
faced less competition because the residents in those
cities had less capacity to consume expensive cars that
were produced by the foreign automakers and
Sino-foreign joint automakers.
4.2 Spatial clusters
The global Moran’s I values for the distribution of car
sales of total automakers, foreign automakers,
Sino-foreign joint automakers, and Chinese automakers
were 0.15, 0.08, 0.15, and 0.22 with a standard normal
z-value of 10.67, 6.45, 10.99, and 15.66, respectively.
These show that the car sales of all kinds of automakers
LIU Daqian et al. Spatial Patterns of Car Sales and Their Socio-economic Attributes in China 689
were distributed non-randomly and displayed positive
spatial autocorrelations. There were clusters of cities
with higher or lower car sales, meaning the cities that
had high car sales were typically surrounded by cities
with high car sales, whereas cities with low car sales
were typically surrounded by cities with low car sales.
To detect these local clusters, the z-score of each city’s
local Moran’s I was computed and reported in Fig. 2.
Since we focused on all the four types of car sale spots
in Chinese prefectural cities (High-High, Low-Low,
High-Low, and Low-High), cities with a z-score more
than 1.96 or less than –1.96 (more than 95% signifi-
cance) were highlighted using different colors, as shown
in Figs. 2(a–d).
Fig. 2a shows the spatial clusters of the total car sales.
We use red and blue colors to represent the areas whose
indexes of local Moran’s I are positive and negative,
respectively, with a statistical significance at the 0.05
level. All the cities with positive indexes of local
Moran’s I were hotspots (High-High), and were mostly
located in the Jing-Jin-Ji region, Shandong Province,
Yangtze Delta Region, and Pearl River Delta Region,
indicating that these areas had significantly higher car
sales and were surrounded by cities with similar higher
values. These clusters were the major places for car
sales in China. Most of these cities were economically
developed and densely populated. While cities with
negative indexes of Moran’s I can be classified into two
types: the convex spots (High-Low) and concave spots
(Low-High). Almost all the clusters were convex spots,
except Zhoushan City in Zhejiang Province, which was
a concave spot. Zhoushan City was a small island city
and located in the Yangtze Delta Region. The demand
for cars in Zhoushan City was relatively small because
of the isolated and orographic conditions as well as the
relatively smaller population. However, Zhoushan’s
neighbors were mostly developed cities with higher car
sales; therefore, the same pattern of Low-High was gen-
erated, as illustrated in Figs. 2(b–d). Other cities with
negative local Moran’s I were convex spots (high car
Fig. 2 Spatial clusters of car sales for total automakers (a), foreign automakers (b), Sino-foreign joint automakers (c), and Chinese
automakers (d)
690 Chinese Geographical Science 2017 Vol. 27 No. 5
sales but neighbors had low car sales), including the
cities of Xi’an, Chongqing, Wuhan, Changsha, Chengdu,
and Kunming. These cities were all capital cities of their
provinces, which played a central role in leading eco-
nomic development and urbanization in the hinterland
provinces they were located. Thus, a High-Low pattern
of car sales was generated.
Fig. 2b shows the spatial clusters of car sales by for-
eign-owned automakers, in which we can see fewer but
more concentrated hotspots, especially in the Jing-Jin-Ji
Region. Only Beijing and Tianjin were identified as car
sale hotspots for the foreign-owned automakers, which
is understandable as only the most developed cities had
enough capacity to consume these relatively expensive
cars. However, the convex spots of car sales for for-
eign-owned automakers were nearly the same as the
ones for total car sales, with only one more city
(Zhengzhou, the capital city of Henan Province) identi-
fied.
Fig. 2c illustrates the spatial clusters of car sales for
Sino-foreign joint venture automakers, which is almost
identical to the pattern of the total car sales. It may be
because more than 70% of the total car sales in 2012
were produced by this type of automakers. Therefore,
the car sales by joint ventures are largely consistent with
the total sales.
In contrast, the car sales for Chinese automakers pre-
sent a distinguishing spatial clustering pattern. As shown
in Fig. 2d, numerous hotspots were located in North
China, particularly in the Jing-Jin-Ji Region and Shan-
dong Province. Only a few hotspots were located in the
Yangtze Delta Region, and none was located in South
China. This shows that car sales by Chinese automakers
were concentrated in North China and East China
rather than South China. The convex spots of car sales
by Chinese automakers were identical to the other
types of automakers, with Shenzhen as an exception,
another city with a large population and developed
economy. Note also that in Fig. 2d, besides Zhoushan,
another city named Laiwu was identified as a concave
spot. It is because Laiwu was the smallest prefectural
city in Shandong Province with a population of 1.30 ×
106, while all its neighboring cities had a population of
over 5 × 106. Hence, the car sales of Chinese auto-
makers in Laiwu were significantly lower than its sur-
rounding neighbors, which in turn make Laiwu City a
concave spot.
4.3 Location quotient of car sales
Location quotient (LQ) is a useful indicator to measure
the comparative advantage of an automaker in a city. It
focuses on the relative share of car markets that a
manufacturer commands, as opposed to the absolute
value of car sales. It is also a valuable way of quantify-
ing the level of importance of a manufacturer in a city’s
market, which further reveals local differences in com-
parison to the national average. If a city has a LQ by a
particular type of automaker higher than 1, it means that
this type of manufacturer managed to achieve more
market share in this city than the average level in the
country.
As illustrated in Fig. 3a, high LQ for foreign auto-
makers appeared mostly in the more developed cities in
each province, such as cities in the southeast coastal
provinces and capital cities in the inner provinces. This
is not surprising because people in these cities can af-
ford more imported cars due to the relatively higher in-
come level. What is surprising is that some inland cities
that were not commonly considered developed, such as
Yulin, Ordos, Da Hinggan Ling Prefecture, and some
cities in Tibet, nevertheless recorded high LQ. There are
several possible reasons for this phenomenon. First,
some types of cars produced by foreign-owned auto-
makers are popular in the western China, especially the
SUV produced by manufacturers such as Toyota, Mitsu-
bishi, and Hyundai. Second, these cities have relatively
higher average income level compared to ordinary
inland cities. For example, residents in Yulin and Ordos
were relatively richer because these two cities had
abundant coal resources.
Fig. 3b shows a different pattern of LQ distribution
for Sino-foreign joint venture automakers. The variances
of LQ for joint ventures were not as large as the other
two types, with the highest values of LQ no more than
1.3. This indicates that the joint ventures had a more
even market shares across China. As shown in Fig. 3b,
cities with higher LQ were concentrated in the eastern
and southeastern China. The LQ for Chinese automakers
exhibits almost an opposite pattern to the other two
types. Most cities in the middle, western, and northeast-
ern part of China had a high LQ. Cities in the western
China, including provinces of Tibet, Xinjiang, and Inner
Mongolia, had LQ values higher than 1.5, which means
that the Chinese automakers obtain above average mar-
ket shares in these cities. Note also that most of these
LIU Daqian et al. Spatial Patterns of Car Sales and Their Socio-economic Attributes in China 691
cities were underdeveloped.
As illustrated in Fig. 3c, the more underdeveloped the
city is, the higher LQ values for Chinese automakers
tend to be. This may be because residents in these un-
derdeveloped cities had lower car consumption than
those in developed cities, and price factor was still the
primary consideration for consumers in these cities;
therefore, cheaper cars produced by domestic auto-
makers were more popular. It also shows that the do-
mestic automakers invested more effort on grabbing the
markets in these developing cities where they face less
competition from joint ventures and foreign-owned
companies. To meet the consumer demand in the west-
ern China, the Chinese automakers have designed
cheaper but prevailing car models, such SUV and MPV.
In a certain sense, these cities contributed significantly
to the development of domestic automakers, although
the amount of car sales might not be large compared to
those in developed cities. In some cities in the western
China, domestic automakers have acquired an absolute
dominant position in the market. Without doubt, these
underdeveloped cities will play increasingly important
roles in the market strategies of Chinese automakers.
4.4 Factor analysis and regression analysis
Several variables capturing the socioeconomic and
demographic characteristics in cities were selected to
examine the factors that may have affected car sales. As
shown in Table 2, these variables exhibit large variations
across cities. For example, the average income of resi-
dents (including both urban and rural) in the city is an
important indicator that may determine the people’s
consumption capacity. The GDP of a city is a basic in-
dicator that may directly reflect the economic develop-
ment level. The resident population variable is usually
positively associated with the car sales because more
people mean more demand for goods. Meanwhile, the
fixed assets investment, the RMB deposits in financial
institutions, and the local fiscal revenue can reflect the
consumption power in the society and the investment
capability of the government and people.
Because these variables may correlate with each
other, it is useful to identify a small number of easily in-
terpretable factors that were responsible for the observed
Fig. 3 Location quotient of car sales for foreign-owned automakers (a), Sino-foreign joint automakers (b), and Chinese automakers (c)
692 Chinese Geographical Science 2017 Vol. 27 No. 5
Table 2 Descriptive statistics of the variables
Va ri a b l e s N Minimum Maximum Mean SD
Population (104 persons) 337 7.60 2919 396.07 324.92
GDP (108 yuan (RMB)) 337 14.53 19195.69 1574.60 2213.22
Fixed assets investment (108 yuan (RMB)) 337 15.37 7579.45 886.33 1034.53
Local fiscal revenue (108 yuan (RMB)) 337 1.26 3429.83 134.51 305.59
Per-capita net income of rural residents (yuan (RMB)) 337 2362 22842 7548.25 2960.45
Per-capita disposable income of urban residents (yuan (RMB)) 337 9759 39513 19181.40 4879.53
Total volume of retail sales of consumer goods (108 yuan (RMB)) 337 3.16 6900.32 545.64 828.80
RMB deposits in financial institutions (108 yuan (RMB)) 337 41.30 69883.87 2313.03 5728.23
correlations among variables. We conducted a factor
analysis to obtain the underlying constructs. The princi-
pal component analysis (PCA) is used to extract factors
from the correlation matrix. Table 3 illustrates the re-
sults of total variance explained, which shows the per-
centage of the total variance attributable to each factor.
According to the extraction sums of squared loadings,
more than 89% of the total variance is explained by the
first two factors and their eigenvalues are more than 1.
Therefore, the first two factors are chosen as the final
factors to represent the variables.
To interpret the two factors, we used the varimax or-
thogonal rotation method. Table 4 shows the rotated
component matrix. The first factor is highly correlated
with population, GDP, fixed assets investment, local
fiscal revenue, total volume of retail sales of consumer
goods, and the RMB deposits in financial institutions.
Thus, we labeled it the scale factor. The second factor
had large loadings on the per-capita disposable income
of urban residents and the per-capita net income of rural
residents, so we labeled it the income factor. Based on
the scores for each factor, two new independent vari-
ables are generated, and regression models are cali-
brated to evaluate the impacts of these factors on car
sales by different manufacturers.
OLS regressions are conducted initially in GeoDa
1.8.8; Table 5 shows the results. To diagnose the spatial
dependence problem, the Moran’s I (error) and La-
grange multiplier statistics are also computed and
shown. As observed in Table 5, the spatial dependence
problem existed in all the models. To address this prob-
lem, we incorporated a spatial lag or spatial error vari-
able into the models following Wang (2006). By com-
paring the results of Lagrange multiplier tests for dif-
ferent models, we chose either a spatial lag model or a
spatial error model to calibrate the spatial regression
models based on whose significances were higher; Table
6 shows the results.
Table 3 Total variance explained
Initial eigenvalues Extraction sums of squared loadings Rotation sums of squared loadings
Component
Total % of variance Cumulative (%)
Total % of variance Cumulative (%) Total % of variance Cumulative (%)
1 6.089 76.117 76.117 6.089 76.117 76.117 4.838 60.481 60.481
2 1.068 13.345 89.461 1.068 13.345 89.461 2.318 28.980 89.461
3 0.410 5.125 94.586
4 0.170 2.120 96.706
5 0.158 1.974 98.680
6 0.059 0.743 99.423
7 0.036 0.449 99.872
8 0.010 0.128 100.000
LIU Daqian et al. Spatial Patterns of Car Sales and Their Socio-economic Attributes in China 693
Table 4 Rotated Component Matrixa
Component
Va ri a b l e s
1 2
Population 0.888 0.111
GDP 0.883 0.441
Fixed assets investment 0.835 0.366
Local fiscal revenue 0.903 0.294
Per-capita net income of rural residents 0.233 0.930
Per-capita disposable income of urban residents 0.332 0.892
Total volume of retail sales of consumer goods 0.902 0.391
RMB deposits in financial institutions 0.883 0.280
As shown in Table 6, the spatial regression models
for car sales by all automakers, foreign-owned auto-
makers, and Sino-foreign joint automakers were better
by incorporating a structure of spatial dependence (spa-
tial error) in the error term. The spatial regression model
for car sales by Chinese automakers was better by in-
cluding a spatial lag of the dependent variable as an ad-
ditional explanatory variable. Compared to the OLS
regression, all the goodness of fit measures, such as the
log likelihood, Akaike’s Information Criterion (AIC),
and Schwartz’s Criterion (SC), demonstrated that all the
spatial regression models show a better model fit. This
demonstrates that the car sales in one city were related
to the car sales of the city’s neighbors. The reasons may
derive from different aspects. People from neighboring
cities tend to have similar socioeconomic status, cultural
background, and hence similar purchasing capability
and consumption habit. Thus, they are inclined to pur-
chase the same brand of cars. Furthermore, if some
automakers have won wide acceptance in a city, they are
also likely to be preferred in the neighboring cities.
Table 5 Results of ordinary least squares regression models
Va ri a b l e s Car sales of all
automakers
Car sales of foreign-owned
automakers
Car sales of Sino-foreign
joint automakers
Car sales of Chinese
automakers
Constant 38455.2*** 3045.03*** 26024.7*** 9385.44***
Factor_1(scale) 48850.3*** 5816.74*** 34459.9*** 8573.67***
Factor_ 2(income) 23106.6*** 2698.15*** 17112.8*** 3295.68***
Adjusted R2 0.8794 0.8466 0.8902 0.6266
F value 1236.56 930.16 1375.13 285.479
Log likelihood –3846.73 –3170.41 –3713.18 –3493.42
AIC 7699.46 6346.83 7432.35 6992.84
SC 7710.95 6358.31 7443.84 7004.32
Moran’s I (error) 3.3819*** 3.6935*** 2.6729*** 6.6378***
LM lag 5.4092** 3.0655* 2.0317 52.4932***
Robust LM (lag) 0.9429 17.4212*** 0.1314 14.3239***
LM error 10.4253*** 12.5162*** 6.3794** 41.7014***
Robust LM (error) 5.9590*** 26.8719*** 4.4791** 3.5321*
Notes: * statistical significance at 10% level; **statistical significance at 5% level; ***statistical significance at 1% level
Table 6 Results of spatial regression models
Va ri a b l e s Car sales of all
automakers
Car sales of foreign-owned
automakers
Car sales of Sino-foreign
joint automakers
Car sales of Chinese
automakers
Constant 38012*** 3061.82*** 25859.1*** 6638.79***
Factor_1(scale) 48720.8*** 5942.93*** 34475.6*** 8157.77***
Factor_ 2 (income) 23150.3*** 2782.69*** 17077.9*** 2738.59***
R2 0.8855 0.8536 0.8938 0.6794
Log likelihood –3841.37 –3164.77 –3709.93 –3472.21
AIC 7688.74 6335.53 7425.87 6952.43***
SC 7700.23 6347.02 7437.36 6967.74***
Spatial lag – – – 0.293119***
Spatial error 0.244387*** 0.23487*** 0.190741*** –
Notes: * statistical significance at 10% level; **statistical significance at 5% level; ***statistical significance at 1% level
694 Chinese Geographical Science 2017 Vol. 27 No. 5
Generally, the models present good prediction ability,
especially for car sales of the total automakers, for-
eign-owned automakers, and Sino-foreign joint auto-
makers, which could explain more than 80% of the
variation of the car sales. Although the R2 is not very
high, the model for car sales for Chinese automakers can
still explain nearly 68% of the variation. Moreover, as
expected, the coefficients of the independent variables
reveal that car sales are positively and significantly re-
lated with the scale factor, income factor, and car sales
in neighboring cities. The coefficients of the scale factor
were the highest in all the models, which means the
scale factor makes the largest contribution to the varia-
tions of car sales. The models also confirm that car sales
in a city are affected by the income level of the urban
and rural residents. In other words, the higher car sales
are more likely to appear in cities with higher income
level. As the previous studies we mentioned revealed,
the car ownership for a household largely depends on
the income level. This research provides another exam-
ple illustrating the importance of income level in affect-
ing the car sales in a city.
By comparing the results of the four different models,
we can further distinguish the differences among the
models in terms of their explanatory power. The regres-
sion model for the car sales of the Sino-foreign joint
automakers has the highest R2 of 0.8938, which indi-
cates the strongest association between the car sales of
Sino-foreign joint automakers and the exploratory fac-
tors. The possible reason may be that the consumers in
cities with larger proportion of population and with
higher income level place more importance on the qual-
ity of cars (produced by the Sino-foreign joint auto-
makers) despite higher prices. Probably due to the same
reason, the car sales of the foreign-owned automakers
are also highly related with the two factors. However,
the R2 for car sales of Chinese automakers is 0.6794,
which is obviously lower than that of the others. The
relative lower explaining capability of this model may
be because more Chinese cars are sold in undeveloped
cities, which exerts a reverse impact on the coefficients
of the model.
5 Discussion and Conclusions
Although the nature of car sales is one of the most en-
during topics among researchers, the understanding of
their spatial distribution across cities in China was lim-
ited due to the lack of reliable data. By mapping spatial
distributions of car sales, we were able to examine the
differences in spatial patterns among types of auto-
makers. In particular, by identifying the different spatial
clusters, we obtained a clearer pattern of spatial ag-
glomeration or heterogeneity. Through calculating and
visualizing the location quotients of different manufac-
turers regarding car sales, we could further examine and
understand the various roles that different types of
automakers play in the car markets in cities across
China. Finally, by conducting the regression analysis,
we further explored the factors that were related to the
car sales volume. Three conclusions emerged from the
findings.
First, car sales varied greatly across cities in China,
and different types of automakers exhibited different
spatial patterns. Generally, the more developed the city
was, the more cars were sold there, although there are
some significant exceptions in car sales by domestic
automakers. Cities with high car sales were usually lo-
cated in the Yangtze Delta Region, Jing-Jin-Ji Region,
Pearl River Delta Region, or were capital cities in inland
provinces. The spatial pattern became clearer after the
identification of hotspots, convex spots, and concave
spots. The developed cities were the most important
markets for all the manufacturers.
Second, different kinds of automakers obtain diverse
advantages in the markets across cities in China. The
foreign-owned automakers obtain more market shares in
the most developed cities, such as cities in the Yangtze
Delta Region and the capital cities of inland provinces.
The Sino-foreign joint automakers had relatively even
market shares in most cities, although people in the
eastern and southeastern part of China tended to buy
more cars produced by the joint ventures. The Chinese
automakers commanded more market shares in most
parts of middle and western China. The more underde-
veloped the city was, the higher proportion of market
share the domestic automakers tended to obtain.
Third, the scale and the income factors, together with
the spatial effects, were strong factors for determining
the amount of car sales. All the coefficients in the mod-
els were positive and statistically significant. The mod-
els for car sales of the Sino-foreign joint automakers
exhibited the highest explanatory power, indicating that
the scale and income factors were key indicators to pre-
LIU Daqian et al. Spatial Patterns of Car Sales and Their Socio-economic Attributes in China 695
dict the amount of car sales of the Sino-foreign joint
automakers. However, the association between car sales
of the Chinese automakers and the contextual factors
was a little weaker than that of the others, which may be
because Chinese automakers obtain more market share
in the underdeveloped cities.
This study could provide some valuable insights into
the process of car purchasing across cities. Different
types of auto manufactures had different marketing ori-
entations and strategies, and hence diverse market
shares across cities. Further, this study also demon-
strated that car sales could be related with the scale fac-
tor, income factor, and the car sales in neighboring cit-
ies, which is helpful for predicting the car sales in cities.
In the future, it will be interesting and necessary to fur-
ther explore the dynamic characteristics of spatial pat-
terns regarding car sales if more data that are accurate in
terms of car sales in different years are available.
Acknowledgement
The authors would like to express their thanks to Eco-
nomic Advisory Center in the State Information Center
(SIC) for providing data used in our research.
References
Anselin L, 1995. Local indicators of spatial association—LISA.
Geographical Analysis, 27(2): 93–115. doi: 10.1111/j.1538-
4632.1995.tb00338.x
Anselin L, 2005. Exploring Spatial Data with GeoDaTM: A Work-
book. Urbana, IL. USA: Spatial Analysis Laboratory, Depart-
ment of Geography, University of Illinois.
Bandeen R A, 1957. Automobile consumption, 1940–1950.
Econometrica, 25(2): 239–248. doi: 10.2307/1910252
Cao X S, Huang X Y, 2013. City-level determinants of private car
ownership in China. Asian Geographer, 30(1): 37–53. doi:
10.1080/10225706.2013.799507
Clark B, Chatterjee K, Melia S, 2016.Changes in level of house-
hold car ownership: the role of life events and spatial context.
Transportation, 43(4): 565–599. doi: 10.1007/s11116-015-
9589-y
Clark S, Finley A O, 2010. Spatial modelling of car ownership
data: a case study from the United Kingdom. Applied Spatial
Analysis and Policy, 3(1): 45–65. doi: 10.1007/s12061-009-
9030-z
Cullinane S, 2002. The relationship between car ownership and pub-
lic transport provision: a case study of Hong Kong. Transport
Policy, 9(1): 29–39. doi: 10.1016/S0967-070X(01)00028-2
Dargay J, Gately D, 1999. Income’s effect on car and vehicle
ownership, worldwide: 1960–2015. Transportation Research
Part A: Policy and Practice, 33(2): 101–138. doi: 10.1016/S
0965-8564(98)00026-3
Giuliano G, Dargay J, 2006. Car ownership, travel and land use: a
comparison of the US and Great Britain. Transportation Re-
search Part A: Policy and Practice, 40(2): 106–124. doi:
10.1016/j.tra.2005.03.002
Guo Z, 2013. Does residential parking supply affect household
car ownership? The case of New York City. Journal of Trans-
port Geography, 26: 18–28. doi: 10.1016/j.jtrangeo.2012.
08.006
Huang Pingting, Zhang Xiaoping, 2014. Spatial evolution of
automobile industry in Beijing-Tianjin-Hebei Metropolitan
Region. Geographical Research, 33(1): 83–95. (in Chinese)
Huo H, Wang M, 2012. Modeling future vehicle sales and stock
in China. Energy Policy, 43: 17–29. doi: 10.1016/j.enpol.2011.
09.063
Lee S M, 2011. A comparative study of the automobile industry in
Japan and Korea: from visible to invisible barriers. Asian Sur-
vey, 51(5): 876–898. doi: 10.1525/as.2011.51.5.876
Li Shaoxing, Gu Chaolin, 2010. Empirical study on the in-
tra-product specialization of Yangtze River Delta: a case study
of auto manufacturing industry. Geographical Research,
29(12): 2132–2142. (in Chinese)
Ling Z W, Cherry C R, Yang H T et al., 2015. From e-bike to car:
a study on factors influencing motorization of e-bike users
across China. Transportation Research Part D: Transport and
Environment, 41: 50–63. doi: 10.1016/j.trd.2015.09.012
Liu W D, Yeung H W C, 2008. China’s dynamic industrial sector:
the automobile industry. Eurasian Geography and Economics,
49(5): 523–548. doi: 10.2747/1539-7216.49.5.523
Liu Zuoli, He Canfei, 2011. Agglomeration, institutions and the
functional location of auto TNCs in China. Geographical Re-
search, 30(9): 1606–1620. (in Chinese)
Lo K, 2014. A critical review of China’s rapidly developing re-
newable energy and energy efficiency policies. Renewable and
Sustainable Energy Reviews, 29: 508–516. doi: 10.1016/j.
rser.2013.09.006
Oakil A T M, Ettema D, Arentze T et al., 2014. Changing house-
hold car ownership level and life cycle events: an action in an-
ticipation or an action on occurrence. Transportation, 41(4):
889–904. doi: 10.1007/s11116-013-9507-0
Oakil A T M, Manting D, Nijland H, 2016. Determinants of car
ownership among young households in the Netherlands: The
role of urbanisation and demographic and economic charac-
teristics. Journal of Transport Geography, 51: 229–235. doi:
10.1016/j.jtrangeo.2016.01.010
Pallares-Barbera M, 1998. Changing production systems: the
automobile industry in Spain. Economic Geography, 74(4):
344–359. doi: 10.2307/144329
Qian L X, Soopramanien D, 2014. Using diffusion models to
forecast market size in emerging markets with applications to
the Chinese car market. Journal of Business Research, 67(6):
1226–1232. doi: 10.1016/j.jbusres.2013.04.008
Qian L X, Soopramanien D, 2015. Incorporating heterogeneity to
forecast the demand of new products in emerging markets:
696 Chinese Geographical Science 2017 Vol. 27 No. 5
green cars in China. Technological Forecasting and Social
Change, 91: 33–46. doi: 10.1016/j.techfore.2014.01.008
Rits V, Kypreos S, Wokaun A, 2004. Evaluating the diffusion of
fuel-cell cars in the China markets. IATSS Research, 28(1):
34–46. doi: 10.1016/S0386-1112(14)60090-X
Ritter N, Vance C, 2013. Do fewer people mean fewer cars?
Population decline and car ownership in Germany. Transpor-
tation Research Part A: Policy and Practice, 50: 74–85. doi:
10.1016/j.tra.2013.01.035
Rubenstein J M, 1986. Changing distribution of the American
automobile industry. Geographical Review, 76(3): 288–300.
doi: 10.2307/214147
Seya H, Nakamichi K, Yamagata Y, 2016. The residential parking
rent price elasticity of car ownership in Japan. Transportation
Research Part A: Policy and Practice, 85: 123–134. doi:
10.1016/j.tra.2016.01.005
Wan Z, Sperling D, Wang Y S, 2015.China’s electric car frustra-
tions. Transportation Research Part D: Transport and Envi-
ronment, 34: 116–121. doi: 10.1016/j.trd.2014.10.014
Wang F H, 2006. Quantitative Methods and Applications in GIS.
Boca Raton, FL: CRC Press.
Wang Junsong, He Canfei, 2009. Agglomeration economies, FDI
spillovers and Chinese automobile enterprises efficiency. Pro-
gress in Geography, 28(3): 337–344. doi: 10.11820/dlkxjz.
2009.03.003(in Chinese)
Wang Y F, Li K P, Xu X M et al., 2014.Transport energy con-
sumption and saving in China. Renewable and Sustainable
Energy Reviews, 29: 641–655. doi: 10.1016/j.rser.2013.08.104
Wu N, Zhao S C, Zhang Q, 2016. A study on the determinants of
private car ownership in China: findings from the panel data.
Transportation Research Part A: Policy and Practice, 85:
186–195. doi: 10.1016/j.tra.2016.01.012
Xiao Yuanfei, 2010. The competitive pattern features of China’s
auto industry and its strategy decision in post-crisis era. Future
and Development, 33(11): 25–30. (in Chinese)
Xie Chunyan, Liu Daqian, 2014. Spatial pattern of passenger-car
consumption market in China. Economic Geography, 34(3):
85–90. (in Chinese)
Yagi M, Managi S, 2016. Demographic determinants of car own-
ership in Japan. Transport Policy, 50: 37–53. doi: 10.1016/
j.tranpol.2016.05.011
Zhao Junzhu, Sun Tieshan, Li Guoping, 2014. Agglomeration and
firm location choice of China’s automobile manufacturing in-
dustry. Acta Geographica Sinica, 69(6): 850–862. (in Chinese)
