HOW DO INCUBATORS AFFECT THE INTRA-CITY SPATIAL PATTERN OF HIGH-TECH STARTUP FORMATION? AN EMPIRICAL STUDY OF SHENZHEN, CHINA A Thesis Presented to the Faculty of the Graduate School of Cornell University In Partial Fulfillment of the Requirements for the Degree of Master of Science by Jiarui Zheng August 2021 © 2021 Jiarui Zheng ABSTRACT Entrepreneurship and innovation have become increasingly important for regional economic development and industry upgradation in recent China. This research focuses on the rapid growth of entrepreneurship in the city of Shenzhen and investigates the relationship between incubation policies and high-tech startup formation from a spatial perspective. Using firm-level data in 2020 and employing a geographical grid-based analysis, it builds spatial econometric models to examine the effect of local incubators on the formation and aggregation of new firms. The spatial Zero-inflated Poisson regression model is selected after a comparison of different models. After controlling other factors, the result shows that incubators have a significant positive effect on attracting new high-tech startups. Besides, the effect of different types and levels of incubators is different. In general, the lower administrative level mass maker spaces have a more significant positive effect than the larger, government-backed technology business incubators. BIOGRAPHICAL SKETCH Jiarui Zheng is currently a Master candidate in Regional Science major at Cornell University. Before studying at Cornell, Jiarui received her dual bachelor’s degrees in Human Geography and Urban-rural Planning and Economics from Peking University in China. Her academic interests lie in economic geography and urban analytics. She has a multi-disciplinary curriculum including urban studies, economics, business and data analytics, and is always passionate about solving real-world problems by applying her knowledge and academic skills. iii ACKNOWLEDGMENTS I would like to express my sincere gratitude to my advisor and committee member Kieran Patrick Donaghy, without whose encouragement and insightful suggestions I would not finish my thesis well. More importantly, thanks to the in-depth talks with Kieran during my three years at Cornell University, which inspired me to explore more opportunities when I was struggling with my academic direction and future career path. I also would like to thank my committee member John Carruthers. John gave a lot of impressive and valuable instructions during my thesis writing, especially on the quantitative methods. Besides, John also gave me a lot of emotional support by recognizing my efforts and progress. I also want to express my thanks to all the facilities and staffs. I really enjoy my study with Regional Science program at Cornell University, which is diverse, flexible, and meaningful. Finally, I would like to thank my families and friends for always being here for me. Thanks for my parents for their understanding and support. Thanks to my friends and classmates for sharing their ideas with me and accompanying me through the difficult time. iv TABLE OF CONTENTS CHAPTER 1. INTRODUCTION ........................................................................................................... 1 1.1 BACKGROUND .................................................................................................................................. 1 1.2 RESEARCH PURPOSE AND SIGNIFICANCE ......................................................................................... 3 1.3 RESEARCH QUESTION AND STRUCTURE ........................................................................................... 6 1.3.1 Research Question ................................................................................................................. 6 1.3.2 Thesis Structure ..................................................................................................................... 7 1.4 RESEARCH CONTRIBUTION .............................................................................................................. 7 CHAPTER 2. LITERATURE REVIEW ............................................................................................... 9 2.1 SPATIAL DISTRIBUTION OF HIGH-TECH START-UPS ......................................................................... 9 2.1.1 Theories Underlying the Unbalanced Firm Spatial Distribution ......................................... 9 2.1.2 Spatial Pattern of High-tech Start-ups ................................................................................ 10 2.1.3 Determinants of Intra-city High-tech Start-up Formation .................................................. 12 2.1.4 Overview of the Empirical Research in China .................................................................... 15 2.2 THE EFFECT OF BUSINESS INCUBATORS ON HIGH-TECH STARTUPS ............................................... 16 2.2.1 The Definition and Function of Business Incubators .......................................................... 16 2.2.2 Chinese Incubator Development and Relevant Research ................................................... 17 2.2.3 Examination of the Effect of Incubators .............................................................................. 19 2.2.4 The Discussion of Spatial Effect of Incubators ................................................................... 19 2.3 SUMMARY AND DISCUSSION OF THE LITERATURE ......................................................................... 20 CHAPTER 3. METHODOLOGY ........................................................................................................ 22 3.1 SPATIAL PATTERN EXPLORATION .................................................................................................. 22 3.1.1 Hot Spot Analysis ............................................................................................................. 22 3.2 RELATIONSHIP ANALYSIS .............................................................................................................. 23 3.2.1 Geographical Grids Based Analysis ................................................................................... 23 3.2.2 Poisson Regression and Negative Binominal Regression Model ........................................ 23 3.2.3 Zero-inflated Poisson and Zero-inflated Negative Binominal Model ................................. 25 3.2.4 Spatial Extension of the Model ............................................................................................ 26 CHAPTER 4. DATA AND VARIABLES ............................................................................................ 28 4.1 STUDY AREA .................................................................................................................................. 28 4.2 DATA SOURCE AND PROCESSING ................................................................................................... 29 4.2.1 Firm Data ............................................................................................................................ 29 4.2.2 Incubator Data .................................................................................................................... 29 v 4.2.3 Venture Capital Firms ......................................................................................................... 32 4.2.4 Spatial POI data .................................................................................................................. 32 4.3 VARIABLES SPECIFICATION ............................................................................................................ 33 CHAPTER 5. DESCRIPTION OF HIGH-TECH STARTUP SPATIAL PATTERN .................... 36 CHAPTER 6. HOW DO INCUBATORS AFFECT HIGH-TECH STARTUP SPATIAL PATTERN? ............................................................................................................................................. 40 6.1 SUMMARY AND TEST OF DATA ...................................................................................................... 40 6.2 MODEL SELECTION .................................................................................................................... 43 6.2.1 Basic Model Selection ......................................................................................................... 43 6.2.2 Spatial Extension Exploration ............................................................................................. 47 6.3 MODEL RESULTS AND INTERPRETATION .................................................................................... 50 6.3.1 Determinants of New High-tech Firm Distribution ......................................................... 50 6.3.2 Determinants of the Possibility of New High-tech Firm Existence ................................. 52 6.4 SUMMARY .................................................................................................................................. 57 CHAPTER 7. CONCLUSION AND DISCUSSION ........................................................................... 59 7.1 CONCLUSION .................................................................................................................................. 59 7.2 LIMITATION AND FUTURE DIRECTION ............................................................................................ 60 REFERENCES ....................................................................................................................................... 62 vi LIST OF FIGURES FIGURE 1.1 NEWLY REGISTERED HIGH-TECH STARTUPS IN THE CITY OF SHENZHEN, CHINA ...................... 2 FIGURE 1.2 THE NUMBER OF INCUBATORS AND THEIR TENANTS IN CHINA ................................................. 3 FIGURE 4.1 MAP OF SHENZHEN ................................................................................................................ 28 FIGURE 4.2 MAP OF INCUBATORS IN SHENZHEN (2020) ........................................................................... 32 FIGURE 5.1 THE NUMBER OF NEW TECHNOLOGY AND INTERNET FIRMS IN SHENZHEN (2000-2020) ........ 36 FIGURE 5.2 MAP OF NEW HIGH-TECH FIRM LOCATION (2005-2020) ......................................................... 37 FIGURE 5.3 LISA CLUSTER MAP FOR NEW FIRM NUMBERS (2005-2020) ................................................... 39 FIGURE 6.1 PLOTS OF PREDICTED AND ACTUAL COUNTS OF FIRMS USING DIFFERENT MODELS ................. 46 vii LIST OF TABLES TABLE 4.1 COMPARISON OF MASS MAKER SPACE (MMS) AND TECHNOLOGY BUSINESS INCUBATORS (TBI) ............................................................................................................................................... 30 TABLE 4.2 THE QUALIFICATION OF OFFICIAL IDENTIFICATION OF MASS MAKER SPACE (MMS) ............. 30 TABLES 4.3 THE QUALIFICATION OF OFFICIAL IDENTIFICATION OF TECHNOLOGY BUSINESS INCUBATOR (TBI) ............................................................................................................................................... 31 TABLE 4.4 SUMMARY OF INCUBATOR NUMBER ........................................................................................ 32 TABLE 4.5 DEFINITION OF DEPENDENT AND INDEPENDENT VARIABLES ................................................... 34 TABLE 5.1 SUMMARY OF STATISTICS OF NEW IT FIRMS BY GRIDS (2005-2020) ...................................... 36 TABLE 5.2 MORAN’S I TEST FOR THE NEW FIRM (2000-2020) .................................................................. 38 TABLE 6.1 STATISTICS OF VARIABLES ...................................................................................................... 40 TABLE 6.2 CORRELATION BETWEEN INDEPENDENT VARIABLES ............................................................... 41 TABLE 6.3 VIF TESTS OF THE VARIABLES ................................................................................................. 42 TABLE 6.4 CATEGORIES OF NUMBER OF FIRMS BY GRID CELLS ................................................................. 44 TABLE 6.5 COMPARISON OF FOUR BASIC MODELS .................................................................................... 44 TABLE 6.6 COMPARISON OF NON-SPATIAL AND SPATIAL ZIP MODELS ..................................................... 47 TABLE 6.7 SPATIAL ZIP MODEL OF THE NUMBER OF NEW HIGH-TECH FIRMS ........................................... 53 viii CHAPTER 1. INTRODUCTION 1.1 Background China has been lagged behind in entrepreneurship for a long time, compared with many developed countries in North America and Europe. The turning point occurred in 2014. In late 2014, Chinese government put forward “Mass Entrepreneurship and Innovation” Strategy, which is an initiative to encourage entrepreneurships for all. This is primarily based on China’s development needs for the economic transformation in the new era, when the traditional development model is no longer suitable and new drivers are to be built. Innovation and entrepreneurship, then, were viewed as the key to lead a more energetic and sustainable economic growth. By launching a large number of policies such as subsidies, tax reduction, talent introduction and incubation, the innovation environment has been greatly improved and the number of new startups has increased dramatically. In 2015, under the background of internet prevalence and technology revolution, another initiative “Internet Plus” was proposed, which is a strategy and roadmap leveraging the innovation and application of technology such as cloud computing, big data and the Internet of Things to benefit and transform a variety of industries. Then in 2017 and 2018, more initiatives such as the deepening of entrepreneurship development and the upgraded version of “Mass Entrepreneurship and Innovation” were proposed. Those initiatives together encouraged a wave of new startup formation, especial high-tech and internet related startups. The number of newly registered firms as well as the proportion of technology firms increased a lot. For example, according to IT Juzi, a startup and venture capital database, the new internet startups formed in 2015 is 28,288, nearly twice as many as the number in 2013. As expected, mass entrepreneurship and innovation has been increasingly supporting the growth of economics and the employment restructuring. By the end of 2018, 1 the number of employers of small-and-medium sized and micro-sized companies reached 233 million in China, increased by 5.5% since 2013. The asset of those firms is 402.6 trillion RMB and the total revenue is 188.2 trillion RMB, make up 77.1% and 68.2% of all firms’ asset and revenue, separately. (China’s National Bureau of Statistics, 2019) Compared with developed countries, the defining feature of Chinese entrepreneurship development is government-led and policy oriented. Incubation is among the core of all the policies at different administrative levels. Economic and urban planning as well as policies intend to leverage the ability of incubators to attract and spur new startups, promote new industry agglomeration and improve local innovation and economy. The increasing trend of incubators is similar to that of new startups, making it reasonable to further explore the relationship between the two. Moreover, because of the size and capability of incubators, their effect is more likely to happen at a smaller scale, for example within the city, rather than macro geographical scale. Figure 1.1 Newly registered high-tech Startups in the city of Shenzhen, China 2 Figure 1.2 The number of incubators and their tenants in China 1.2 Research Purpose and Significance Based on the background stated above, this study would like to explore the effect of incubators on Chinese high-tech startup formation from a spatial perspective, and further explore the heterogeneity of different levels and types of incubators. The result could: (1) add to the research of entrepreneurship spatial distribution and the effect of incubators with the latest data; (2) provide implications for the government in entrepreneurship and innovation policies. The research significance is based on the importance of four perspectives. First, it’s the importance of entrepreneurship. Schumpeterian entrepreneurship theory identifies new high-growth startup companies as key factors in technological innovation and economic growth (Schumpeter, 1934a, 1934b). By actively founding new firms, making use of new technology and creating new business models, entrepreneurs not only make profits, but restructure the industry, grow the economy, and improve employment (Adler et al., 2019). Recent empirical research also concludes that entrepreneurship has positive effect on economic growth, especially for emerging economies (Braunerhjelm et al., 2010; Chaşovschi 3 et al., 2014). After the 1990s, with the development of information and communication technology (ICT) and the trend of globalization, entrepreneurs become even more important in economic and social evolution as they commercialize new idea and knowledge (Audretsch et al, 2006). Compared with large firms, the younger and smaller firms are tested to have positive effect on job creations, innovations and productivities (Van Praag and Verloot, 2008; Haltiwanger et al., 2013). Recent Evolutionary Economic Geography considers innovation at the core of economic system evolution (Boschma and Frenken, 2011, Shi and He, 2014). Under this perspective, the only way to keep sustained competitive advantages and offset the negative effects caused by market saturation or the decline of mature industries is to develop new industries and enterprises and promote continuously evolution of clusters (He and Li, 2020). Second, it’s the importance of the spatial perspective of entrepreneurship. Economic geographers and urban economists are interested in the spatial perspective study of entrepreneurship, because spatial characteristics are important at both micro and macro level (Feldmam and Kogler, 2010; Adler et al., 2019). At the micro level, a firm chooses a geographical location in order to maximize its profit by reducing the cost and growing the revenue potentials. For example, the shorter distance to resources and markets will reduce the transportation cost, and the better proximity to the labor market or academic institutes may provide firms with employees of higher quality. The choice of location is crucial for the survival opportunity and long-term success of a firm. At the macro level, entrepreneurship distribution is part of economic and urban planning. A reasonable distribution of new firms will promote local economic growth and industry upgradation as well as improve the balanced and efficient development within the region. Third, it’s the importance of incubation to entrepreneurship agglomeration. Incubation, as a widely used policy-driven instrument to promote entrepreneurship, has important and complex 4 influence on the new firm development and distribution. A business incubator is a subtle integration of subsidized rental space, collectively shared facilities, business assistance and most importantly, resources like capital investment and business network (Schwartz, 2013). By improving the accessibility to professional services, investment, information and reducing the cost to market, incubators aim to benefit the long-term survival and success of startups and also improve the local economic development. A lot of research has stated the positive effect of incubators (Avnimelech et al., 2007; Li and Zhu, 2017; Wu, 2020), while others give the opposite conclusion (Wang, 2019). Though the incubator is such an important factor, the research that relates incubators with the spatial distribution of entrepreneurship is limited. Fourth, it’s the lack of empirical study in the Chinese context. Chinese incubator policies are among the core of its economic development and urban planning but are still at an early stage. The empirical research is lagged behind by its recent expansion (Zhu, 2014). Previous startup spatial research concentrates more on the regional level study, taking cities and counties as the unit for analysis, while the relationship within the city is not subject to sufficient research (Du and Zheng, 2020). Moreover, the study about incubators is mainly from policy, economics and management field, focusing on their efficiency or economic effect assessment. Given that the spatial impact of incubators is important, more research is needed. The lag of deep research has two main reasons. First, it’s the lack of practice. China does not have a long history of market-oriented entrepreneurship and innovation. Until the end of the 20th century, China had been learning from developed countries (Wang, 2019). The initiatives were just proposed seven years ago, before which the development of entrepreneurship and incubator establishment was slow and concentrated at some major regions. Second, it’s due to the lack of data. The effect of incubators often occurs at a small scale (for example within the city), but the detailed spatial data of firms and POIs were hard to get several years ago. 5 Therefore, the recent development in China provides a good case study of entrepreneurship and its reaction to business incubators. The opening of government data and spatial big data makes it possible to do more accurate analysis about the topic. 1.3 Research Question and Structure 1.3.1 Research Question Given the above theory and practice, this research would like to explore the spatial effect of incubators on Chinese high-tech startups. The city of Shenzhen, Guangzhou Province is chosen as the study area because it is a pioneer and representative of Chinese high-tech industry development and innovation. After the first Chinese Special Economic Zone opened there in the 1980s, Shenzhen enjoyed the favorable policies and strong technology industry basis. Shenzhen’s High-tech industries added value became 923 billion RMB in 2019, making 34. 28% of its total GDP (Deloitte, 2020). Firm-level registration data of 79,465 firms in high-tech industry between 2000 to 2020 were collected from the government open data platform and aggregated to the map. 405 incubators and mass maker spaces opened before 2018 were employed to examine the effect. Analysis is based on the regression models. This study seeks to answer the following three questions. 1) How to describe the spatial pattern of the high-tech startup formation within the city of Shenzhen recent years? 2) Given the rapid establishment of policy-backed incubators, does it effectively attract new high-tech startups and influence its intra-city spatial pattern? 3) Is there any heterogeneity of the effect of incubators of different types and levels? If so, 6 which types of incubators have a larger or more significant effect on the spatial pattern? 1.3.2 Thesis Structure To answer the research questions, this paper is structured to comprise seven sections. Section 2 reviews existing literature about start-up spatial distribution and the effect of incubators on new firm spatial pattern. This section also summarizes empirical results and methodology used in Chinese case studies and identifies the research gap in Chinese context. Section 3 discusses the methodology used in the study, including the spatial analysis method, count data regression models and their spatial extension. Section 4 provides explanation of the study area, data, measures and variable specifications. Section 5 does an exploratory analysis of the spatial pattern of new high-tech startups in Shenzhen using spatial analysis methods and identifies the key features of the distribution. Section 6 focuses on the relationship between incubator location and new startup distribution. It presents the model selection, results, and interpretation. Section 7 summarizes the research, provides the policy implications and limitation. It also discusses the direction for future research. 1.4 Research Contribution The contribution of this research can be summarized as the following four aspects. First, it adds to the micro level studies of entrepreneurship distribution. It uses geographical grid cells as the units of analysis instead of districts or subdistricts, which can help understand entrepreneurship location behavior more accurately. 7 Second, it provides in-depth study of incubators from the geographical perspective and discusses its spatial effect. By understanding the effect, urban planners and policy makers can assess whether the establishment of incubators does contribute to a heathier or more balanced development in regional entrepreneurship and innovation. Third, it focuses on high-tech startups with the most up-to-date data sources, which are relatively new but emerging in China, thus are not subject to a lot of research. This study adds to the empirical results in the existing literature. Fourth, it compares four count data models using their prediction accuracy and extends them to SAR and SLX models. The method is creative and worth trying for the topic. 8 CHAPTER 2. LITERATURE REVIEW The literature review section is divided into three parts. Section 2.1 introduces the study of the new enterprise spatial distribution, especially the determinants of the intra-city spatial pattern of high-tech startups. A subsection also pays attention to Chinese research. This part serves as the theoretical basis of the variable specification in the model. Section 2.2 summarizes the development of incubators and reviews the relevant research, including their purpose, function, and effects. Based on the literature review, section 2.3 summarizes current research progress and identifies the gaps this study seeks to fill. 2.1 Spatial Distribution of High-tech Start-ups The theory of entrepreneurship was brought up by Schumpeter (1934a, 1934b, 1954), focusing on the process and impact of business and technological innovation. The Neo-Schumpeterian theory has three levels: 1) the micro level focuses on the enterprise learning behavior and R&D competition; 2) the industrial level focuses on the market structure, industry and firm dynamics changes; 3) and the macro level is mainly about the economic growth of countries (Cohen, 2010). Economic geographers and urban economists are often interested in the last two topics, investigating the geographical pattern of innovation and how spatial factors influence entrepreneurship and technological innovation. 2.1.1 Theories Underlying the Unbalanced Firm Spatial Distribution One important theoretical basis of startup distribution is the growth poles theory proposed by Perroux in 1955. The idea is that economic growth is not uniform over all the sectors but takes places at some specific innovative “poles”. The “poles” are industries and firms that have a high degree of interactions with others and are dominant in the interactions. The leading or dominant companies are often the core of a growth center and the driver for economic growth. 9 The “poles” have features of polarization and diffusion. The former means the process of the development of a polarized firm or industry, while the later means the effect on related industries or firms (Darwent, 1969). While growth pole is initially discussed in abstract economic space, it was then expanded to geographic space (Boudeville and Montefiore, 1966) and became popular in regional and policy studies. The growth center could also be a geographical region with polarized industries. The unbalanced regional growth is viewed as a result of the early development and accumulation of self-reinforcing advantages and growth in those early-developed regions (Ke and Feser, 2010). Same as the economic growth pole, the geographical growth center also has positive effect and negative effect on the surrounding area, called trickling-down effect and polarization effect, separately. Amos (1990) has divided growth pole process into three stages. The first is the concentration of a single center. The second is the diffused concentration at multiple centers and the third is the diffusion to the periphery. This is typical for many economic and innovation activities in the spatial dimension, according to empirical research. The growth centers also related to the concept of agglomeration. Agglomeration has two underlying drivers, localization economies from Marshall (1890) and urbanization economies from Jacobs (1969, 1984). “Marshallian” externalities emerge from the spatial proximity of firms within the same industry, while the “Jacobsian” externalities are about inter-industry knowledge spillovers (Glaeser et al. 1992; Li and Zhu, 2017). The agglomeration is often the distribution pattern of industries and firms. Benefiting from agglomeration, firms in the cluster enjoy the labor market pooling, the linkages with potential suppliers and customers, and knowledge spillover (Kerr & Kominers, 2015). Firms also minimize their production and transportation costs due to the geographical proximity (Du and Zheng, 2020). 2.1.2 Spatial Pattern of High-tech Start-ups 10 An extensive literature is devoted to discussing the spatial pattern and its evolution of high- tech entrepreneurship in different geographical scales, from global, country to city and county levels (Adler et al., 2019, Du and Zheng, 2020). According to empirical studies, the distribution of start-ups, especially high-tech enterprises, shows a geographical centralized pattern and organized by spatially limited ecosystems (Adler et al., 2019). The pattern is often a result of the uneven distribution of resources, marketplace and policies between regions. And the formation of technology start-ups, in turn, contributes to an even more unbalanced economic and industry development. For example, the high-tech start-ups are concentrated at some major metropolitans globally, including San Francisco, New York, Boston and Los Angeles metropolitans in the US, London and surrounding area in England and South and West part in China (Adler et al., 2019, Zhang and Li, 2014). Fischer et al. (2018) study knowledge intensive firms in Brazil using kernel density analysis and find a significant spatial agglomeration pattern at city level. While a lot of studies focus on the macro level distribution, the micro-level research is relatively limited (Du and Zheng, 2020; Adler et al., 2019). Cities are often treated as “black boxes” due to lack of micro data in the past decades. The answer to questions such as the scale of the agglomeration economies and its attenuation with distance remains unclear (Andersson et al., 2019). However, with the growth of city size and the increasing complexity of business flows and activities in the urban area, the micro study is of rising importance recently (Batty, 2013). Policies are expected to be more effective in promoting an efficient and healthy development of intra-city entrepreneurship and innovation. Noticing the demand, there is a growing literature exploring the entrepreneurship within the city. Duan et al. (2019) state that firms are clustered at even denser districts and neighborhoods. For example, the most famous business agglomeration, Silicon Valley in the US, has its technology complex made up of several distinctive technology spillover zones (Kerr & Kominers, 2015). In China, the clusters 11 of internet start-ups in Shanghai are located in several subdistricts including Lujiazui, Wujiaochang, Zhangjiang and Jingan (Duan et al, 2019, Wang, 2019). 2.1.3 Determinants of Intra-city High-tech Start-up Formation A lot of empirical research also explored the influencers of startup or new firm distribution. A large variety of factors are tested. Here the factors that influence entrepreneurship and especially high-tech startup distribution are organized into six categories as follows. 1) Agglomeration Agglomeration or localization economy is tested to be a crucial determinant of firm distribution within the city, especially for high-tech firms. The Evolutionary Economic Geography thinks that the birth of new firm obeys the rule of path dependency, meaning that the more the existing firms in the local region, the higher the formation rate of new firms (Shi and He, 2014). Li and Zhu (2017) explore the determinants of the location process of high- tech firms from the micro level perspective and find significant positive effect of existing agglomeration. Artz et al. (2016) further conclude that the agglomeration effect is not limited to the same industry but can be extended to the related industries. The authors use the location quotient of the same industry, the proximity to upstream and downstream firms to measure agglomeration effect and find all the results are consistent to show that agglomeration will increase the probability of firm entry in both urban and rural area in Iowa and North Carolina. Kerr and Kominers (2015) state that knowledge flow acts over a smaller geographical scale than input-output interactions. High-tech industries such as software show greater elasticity to the clustering (Rosenthal & Strange, 2004). Duan (2019) finds that the proximity to large companies has a positive effect for internet startup location choice. 2) Centrality 12 Centrality means the distance to urban center or commercial centers. Fritsch (2002) finds that the innovation at peripheral regions is less efficient than that in the central region. Wang (2019) finds new start-ups tend to locate near urban centers. The author also concludes that startups are less sensitive to land price as other factors seem to be more attractive to offset the high price at central area. 3) Professional Entrepreneurship Environment Professional entrepreneurship environment includes capital, talents, technology and business services. A lot of research has confirmed that professional business environment has significant influence on the distribution of startups. Lutz (2013) confirms the importance of spatial proximity between investors and investees in a dense economy in Germany. Lin et al. (2019) find Venture Capital Firms (VCF) have a close relation with the location of high-tech industries. Fischer (2018) finds strong evidence in Brazil that the local presence of research- oriented universities, access to capital, and business concentration are correlated to KIE (knowledge intensive enterprise) emergence and density. Adler et al. (2019) find that many of the venture capital-backed high-tech start-ups tend to locate near top universities and institutes, for example, Stanford University and University of California San Francisco, and Harvard and MIT in Boston. Wang (2019) finds new start-ups tend to locate near academic institutes. 4) Transportation Transportation factors include the distance to transit stations such as bus stops, subways, highways, train stations and airports. Transportation, on the one hand, largely improves the accessibility to other factors, lowers the commuting cost and time, and contributes to the agglomeration economies. (Du and Zheng, 2020). On the other hand, the side effect such as traffic congestion can hinder the formation of firms. Yao and Hu (2020) examine the impact 13 of an urban transit line on startup location pattern in Hangzhou, China, and find the effect of transit stations on nearby business formation is significant: the number of retail and technology firms increases by 42% and 29% respectively and the spatial attenuation effect is the strongest for technology firms. Osman et al. (2019) find that the access advantages new firms accruing from clustering with same industries strongly outweigh the impact of traffic congestions for many high-tech industries in Bay Area, USA. Du and Zheng (2020) employ the expansion of subway from 2010 to 2015 as a quasi-experiment to study the effect of accessibility on new firm formation within the city. The authors also conclude that skill- intensive firms have a higher response to the change of accessibility. 5) Local Facility and Amenity Local facilities and infrastructures such as office buildings and hotels are tested to be influential to new firm location. Duan et al. (2019) test the number of office buildings and hotels in the subdistrict and find significant positive effect on internet firm formation in Shanghai. Amenity is another influencer especially for knowledge-incentive and creative firms. Moeller (2018) investigates internet start-ups and confirmed the positive influence of local amenity including bars/pubs, cinemas, theaters, clubs, operas, beer gardens, cafés, restaurants, and art places. Using the fall of the Berlin Wall as the exogeneity shock and the number of internet firms per statistical housing block as the dependent variable, the author finds that a 1% increase in amenity density raises the probability of start-up formation by almost 2%. 6) Planning and Policy Planning and regional policies are another fact that should not be ignored, especially when looking at the Chinese context. The policies include development zones, industrial parks, scientific and high-tech parks and incubators. However, as the policies and services are 14 complex, the outcome is sometimes controversial. While much research shows positive effect of planning on firm formation, others show controversy result. Wang (2019) finds the effect of incubators is not significant for start-ups in Shanghai. Huang et al. (2017) provide empirical justification of development zones in Shanghai by confirming their positive effect on the land use efficiency of electronic firms. Wu (2020) studies internet startup formation in Hangzhou and finds that incubation has positive effect on the spatial agglomeration of startups. 2.1.4 Overview of the Empirical Research in China The empirical study in China has just started from the past decade, and the methods are still limited. From the geographical level aspect, most research about innovation and enterprise behavior are at the macro level, using city or county as the analysis unit (Du and Zheng, 2020). Guo (2020) studies startups in electronic information manufacturing industry in China using Tobit model and identifies several hotspots for new firm formation at city cluster level, such as Yangtze River Delta and Pearl River Delta. The author also identifies the heterogeneity of location choice by region and time period. Ding et al. (2017) study the entrepreneurship activities in Guangdong Province and figure out that capital, social services, talents, technology and economic environment are the main factors that influence startups. Fu and Deng (2017) investigate the location choice of software and IT startups in the city of Guangzhou and its effect on the industry clusters. The research also concludes that new firm entry is a key driver for industry cluster evolution. Study has turned to look at the intra-city pattern of startups in China in recent years. Wang (2019) focuses on the startups in Shanghai and investigates the evolution of their spatial pattern, especially the agglomeration characteristics. She uses OLS, spatial error and spatial lag models to capture the factors of the intra-city distribution of startups and finds the accessibility to city center, talents and public services are significant. Wang et al. (2018) and 15 Xie et al. (2017) study the internet firms in Yangzhou and software firms in Ningbo separately and find new firms agglomerate in main urban districts and spread to suburbs. Wu (2020) researches the internet startup formation in Hangzhou and finds that existing agglomeration, universities and incubation have positive effect on the spatial distribution of startups. Herein, this paper will follow the intra-city topics and methods and investigate the distribution pattern in recent China to add on to the existing empirical literatures. 2.2 The Effect of Business Incubators on High-tech Startups 2.2.1 The Definition and Function of Business Incubators Business incubator is a tool widely adopted by government, universities and private corporations to promote new ventures and foster regional economic development (OECD, 1999, Lewis, 2001). The concept of business incubator (BI) was first developed and practiced in the 1950s when Joseph L. Mancuso opened the Batavia Industrial Center in a Batavia, New York. After that, incubators have expanded in the US in the 1980s and spread to other countries. The definition of business incubators is often based on a description of its function. Smilor and Gill (1986) identify the concept of business incubators as building a link between technology, know-how, entrepreneurial talent, and capital. Chinese government defines an incubator as “a carrier for cultivating and supporting small high-tech enterprises” (Yang, 2020). BI seeks to achieve two types of objectives: a) At the macro-level, it is to enhance the local or regional economic development, technological innovation and create jobs in the long term; b) At the micro-level, it seeks to promote the growth of startups and improve their long- term survival. Different from science parks, industry parks or technology accelerators, business incubators are dedicated to early-stage firms and start-ups and seem like a "one-stop shop" for 16 entrepreneurial support (White et al, 2017). They provide a large range of business resources and services, including the following five aspects. 1) low-rent workspace, shared facilities (Hackett and Dilts, 2004; Aerts et al., 2007; Chan and Lau, 2005) and flexible leasing terms (Schwartz, 2013), which could lower the fixed cost and upfront investment for the early-stage firms. 2) Mentorship and assistance for business development, from product development to marketing, accounting and management (Rice, 2002). 3) Accessibility to a social network, including the internal network and a much broader external network with the potential investors, academic institutes, suppliers and customers. The social network is regarded as a critical strategic resource that benefits the firms even in the long run (Granovetter, 1985, Peters et al., 2004). 4) Accessibility to capital. An incubator can either co-fund the startups or it can have relationships with third-party investors to invest in potential firms. 5) Credibility and firm image building. Through the election process, large and famous incubators also build market recognition and firm credibility for the promising start-ups. The old generation incubators focus more on the physical services, such as shared space, business facilities for the incubated firms, while the more recent incubators position services and network as the most important values (Rubin et al., 2015). Because of the stated functions, incubation initiatives are often put at the heart of the technology and innovation policies to promote local and regional entrepreneurship, technology development, innovation activities and local economic development (Hackett and Dilts 2004; OECD 1997; Schwartz, 2013). 2.2.2 Chinese Incubator Development and Relevant Research Chinese incubators in the past years are really policy oriented. In 1987, the first business incubator, Wuhan East Lake New Technology Entrepreneurship Service Center, was born in China. Then in 1988, Chinese government set up the Torch Program, which is a guiding plan for the development of Chinese high-tech industries. One important objective for the Torch Program is to establish and manage high-tech entrepreneurship service centers, which are 17 combinations of incubator practice overseas and Chinese condition. The purpose is to create a link between laboratories and enterprises, promote the commercialization of high-tech achievements and cultivate new economic growth points. As a result, business incubators become larger in terms of number, size and capacity (Chandra and Fealey, 2009). Zhu (2014) divides Technology Business Incubators into four generations: a) In the late 1980s, the 1st generation incubators provided physical facilities and policy preference mainly based on the economy of scales theory. b) In the 1990s, the 2nd generation incubators started to offer services, such as training, consulting on management, laws and accounting, supplement the lack of market and management experience of startups (Bruneel et al., 2012). 3) In the early 21st century, the 3rd generation incubators started to focus on expertise and professional entrepreneurship services as an added value to the incubated firms. During this period, the incubators became more specialized other than comprehensive. 4) After 2005, the network creation has become the core of the 4th generation incubators. Incubators lead a network among the governments, universities and institutes, enterprises, investors and business services such as accounting, law and financing, and thus found the innovation synergy. By 2019, China has 5,206 Technology Business Incubators (TBI) and 216,828 tenants in the incubators (Torch High Technology Industry Development Center & Ministry of Science and Technology, 2020). China is actively developing a reasonable system and hierarchy of incubators, made up of mass maker spaces, technology business incubators and business accelerators. Different levels of governments have their criteria for identifying an incubator, making the potential effect varies. 18 2.2.3 Examination of the Effect of Incubators It is always a concern whether the wide-spread business incubators really have a positive effect as a policy tool. A lot of empirical research has tested the outcome of incubators on new firms and local development. The research can also be categorized into two levels, company- level and regional-level. Most existing research focuses on the former. For the company level, topics includes long-term survival, sales revenue and growth of firms. Colombo and Delmastro (2002) find that incubated firms show higher growth rates than their non-incubated counterparts. Vásquez-Urriago et al. (2016) conclude that incubated firms can benefit from the agglomeration created by incubators or science parks, which decrease the uncertainty and establish diverse relationship. However, some recent research does not find any significant positive effect as regards the above outcomes (Schwartz, 2013; Lukeš, 2019). For the regional level, topics include geographical distribution of technology start-ups, job creation and regional economic development. Noticing the negative effects of venture capital on technology start-ups, such as narrowing the geographical distribution and narrowing technological diversification, Avnimelech et al. (2007) examine the new incubators in Isreal’s peripheral areas and confirmed that they do attract firms to peripheral areas and less popular technologies, while the success rate of the start-ups is low. Lukeš et al. (2019) analyze 2544 innovative Italian start-ups both in short-term and long-term and find no significant effect on job creation, even though job creation is designed to be a goal of incubators (Bruneel et al., 2012). Phan et al. (2005) also find little evidence for the effect of business incubators on job and wealth creation. 2.2.4 The Discussion of Spatial Effect of Incubators From the geographical perspective, the impact of policy-oriented incubators is not subject to a lot of discussion, and the empirical research shows controversial results in different regions and industries. Taking China as an example, Li and Zhu (2017) find that the number of 19 innovative incubators has a positive effect on new high-tech firm entry at a subdistrict level in the city area, while Wang (2019) finds that state-level mass maker spaces do not has any significant effect on the location choice of new firms. Despite the development of over half a century, technology business incubator is still a complex and multi-faceted issue to study, according to Mian et al. (2016). Under the circumstance that many local governments in China are building incubators to attract market- oriented organizations, more study is needed to assess whether they are promoting an efficient development of new ventures and a healthy pattern of regional economic development. 2.3 Summary and Discussion of the Literature The gap in the literature can be organized into three categories. 1) The understanding of the spatial dimension of new enterprises and its determinants at the intra-city level is limited. Viewing cities as “black boxes”, research hasn’t formed a systematic and comprehensive framework to explain the new startup formation within the city. The empirical research about intra-city spatial distribution only concentrates on some major cities. This becomes an important issue for now and for the future, as cities grow to be larger and more complex, making a healthy development of entrepreneurship and resource allocation necessary for policy makers and urban planners. Thus, room is left for future study on micro-level enterprise spatial research. 2) The analysis of the geographical effect of incubators is limited. Although there is a rich body of literature about the effect of incubators, discussion from the spatial perspective is still limited. Spatial pattern and organization, especially agglomeration, as stated above, is an important characteristic and influencer of entrepreneurship. Therefore, the association between the two deserves more attention and research. 20 3) The empirical research in Chins is limited and lags behind the rapid growth of high-tech entrepreneurship practice in recent years. The policy-backed entrepreneurship and incubation is a key feature of Chinese practice. Compared with the exponential increase of high-tech startups in China, the research to justify the effect of the supporting policies from various aspects is not enough. Herein, this study also would like to add to the literature about the Chinese situation and provide some implications to the policymakers and urban planners. 21 CHAPTER 3. METHODOLOGY 3.1 Spatial Pattern Exploration 3.1.1 Hot Spot Analysis Spatial autocorrelation is the interdependency of the characteristics of the observations neighboring to each other. In the exploratory data analysis, global Moran’s I and Local Indicators of Spatial Association (LISA) are often used to measure the significance and extent of autocorrelation of spatial activities. Moran’s I is applied to measure the overall spatial autocorrelation of firm formation. The statistic is designed to reject the null hypothesis of spatial randomness in favor of an alternative of clustering. By doing Moran’s I test, the autocorrelation feature of the firms in different years can be recognized and compared. The formula is written as: ∑# ∑# 𝑤 𝐼 = !%& "$! !" (𝑥! − ?̅?)+𝑥" − ?̅?, 𝑆'∑#!%&∑# "$!𝑤!" Where ?̅? = &∑# ' & # ' # !%& 𝑥!; 𝑆 = ∑!%&(𝑥! − ?̅?) ; 𝑤# !" is the spatial weight matrix. LISA is suggested in 1995 by Anselin to compliment Moran’s I. It generates a statistic for each location to assess its autocorrelation characteristics. The LISA map provides hot spots (high value surrounded by high values) and cold spots (low value surrounded by low values). By doing this, it is easier to identify the clusters of the firms. Comparing the LISA maps of different years can help recognize the change of pattern. The formular can be written as: # (𝑥! − ?̅?)𝐼! = ' .𝑤!"+𝑥" − ?̅?, 𝑆 "$! Where ?̅? = &∑# 𝑥 ; 𝑆' = &!%& ! ∑#!%&(𝑥! − ?̅?)'; 𝑤!" is the spatial weight matrix. # # 22 In this study, n is the number of grid cells; 𝑥! and 𝑥" is the characteristics of two grid cells i and j. 3.2 Relationship Analysis 3.2.1 Geographical Grids Based Analysis The study uses geographical grid cells as the unit of analysis instead of districts or subdistricts often used in Chinese firm distribution research. This method is widely used in the spatial distribution study when there is no official boundaries or boundaries doesn’t matter a lot (Andersson et al., 2019; Yao and Hu, 2020). The choice of the size of grid cells follows the research of Chinese cities by Du and Zheng (2020) and Yao and Hu (2020). The geo-coded firm-level data and other spatial characteristics are aggregated into grids of 1km×1km on the map. It has two major advantages: First, it can fit well with the situation that high-tech startups often cluster at a smaller scale, which cannot be explained by district or subdistrict level data. Second, it can avoid the effect of the district area on the number of firms. 3.2.2 Poisson Regression and Negative Binominal Regression Model To analyze the firm location choice or the influencers of new startups, two kinds of models are widely used. First is discrete choice model, including Logit and Probit model. This method takes single firm as the unit and study whether it chooses a certain location under the profit maximization consideration. For example, Osman et al. (2019) used Tobit regression to explore the influence of traffic congestion on new firm birth. Mejia-Dorantes et al. (2012) use multinominal logit model to analyze the effect of a new metro line on the spatial pattern of firms. The other is the count data model, using territories as the units of analysis (Li and Zhu, 2017). The distribution of dependent variable could be neither continuous nor normal. For 23 example, in Chinese studies, Duan et al. (2019) use Negative Binominal Model to examine the factors affecting the intra-city distribution of internet startups in the Shanghai, China. Huang (2018) also uses Negative Binominal Regression Model explore the internet startup distribution in Chinese cities. Li and Zhu (2017) use Spatial Lagged Poisson regression models to study the location process of high-tech firms in Nanjing, China at township-level. Yao and Hu (2020) employ difference-in-difference Poisson regression with matching methods to test the impact of urban transit on nearby startup formation. The method is also widely used in studies in other countries and regions studying firm distribution and other economic activities (Arauzo Carod, 2005; Brülhart and Schmidheiny, 2012). As this study seeks to answer the question for policy makers and urban planners, the unit of territories is used, and the Poisson and Negative Binominal Regression methods are employed. The assumption underlying the models is that the firm’s choice of location is independent of one another. The choice between the specific models will rely on the features of data and prediction accuracy of the models. In the Poisson Regression model, the number of new firms located in each territory unit in a given year is the dependent variable. Poisson regression assumes the dependent variable obeys Poisson distribution, and its expected value can be modeled by a linear combination of unknown parameters. The conditional probability of Poisson distribution is given as follows. e()!𝜆 *! 𝑃(𝑌! = 𝑦! | 𝜆 ) = ! ! 𝑦!! where 𝑦! is the number of new firms located in the territorial unit; 𝑃(𝑌! = 𝑦! | 𝑋!) represents the probability density function of 𝑦!; 𝜆! depends on a series of explanatory variables 𝑋!: 𝜆! = 𝑒+,!. And the parameters could be solved using max likelihood-based method. 24 # 𝑙𝑛𝐿 = .[−𝜆! + 𝑦!𝛽𝑋! − ln(𝑦!!)] !%& In the Poisson Regression, the variance of 𝑦! is required to be identical to the mean of 𝑦!. However, the over-dispersion of dependent variable often occurs in the firm birth and location study due to the clustering or heterogeneity of firms. It will cause an underestimation of the standard error and an overestimation of the significance. To adjust the potential over- dispersion problem, this study will also test Negative Binominal Regression. Negative Binominal (NB) Regression is a generalization of Poisson regression by including a random variable 𝜏 which follows a Gamma distribution. The distribution of Negative Binominal is as follows. Γ(𝑦 + 𝜏) 𝜏 𝜆 𝑃(𝑌 = 𝑦 ) = ( )-( )*! ! 𝑦!! Γ(𝜏) 𝜆 + 𝜏 𝜆 + 𝜏 Where y = 0, 1, …, 𝜆, 𝜏 > 0; 𝛼 = 1/𝜏 is the dispersion parameter. 𝑉𝑎𝑟(𝑌) = 𝜆 + 𝜆'𝛼. The variance of Y is larger than mean in Negative Binominal model. In this model, the conditional distribution of 𝑦! on 𝑋! and random effect 𝑢! is still Poisson distribution. It can be estimated using maximum likelihood method as well. 3.2.3 Zero-inflated Poisson and Zero-inflated Negative Binominal Model Another way to address the problem of the Poisson model is to split the data and fit them with different models. Poisson or Negative Binominal regression is not good enough if many zero values exist in the dependent variable. Thus, a Zero-inflated Poisson (ZIP) model or Zero- inflated Negative Binominal (ZINB) model is proposed to solve that problem. The underlying assumption of these two models is that there is a second process determining whether a region has new firm or not. Once the number of new firms is determined to be non-zero, then it obeys the basic Poisson or Negative Binominal model. 25 In a ZIP model, the function is: Φ+ (1 − Φ)𝑒(.! 𝑦! = 0 𝑃(𝑌 = 𝑦!; Φ, 𝜆) = P e()!𝜆 *! (1 − Φ) !𝑦 ! 𝑦! > 0! Where 𝜆 = 𝑒+,! !. In a ZINB model, the function is: 𝜏 - ⎧ Φ+ (1 − Φ) T U 𝑦 = 0𝜆 + 𝜏 𝑃(𝑌 = 𝑦!; Φ, 𝜆, 𝜏) = ⎨ Γ(𝑦! + 𝜏) 𝜏 - 𝜆 *! ( ⎩ 1 − Φ ) 𝑦 ! Γ(𝜏) T𝜆 + 𝜏U V𝜆 + 𝜏W 𝑦 > 0! 𝛼 = 1/𝜏 is the coefficient for dispersion. When 𝛼 → 0, Zero-inflated Negative Binominal model becomes the same as the Zero-inflated Poisson model. In this study, data exploration shows that over half of the grid cells do not have firms in them, which fits the assumption of ZIP and ZINB models. The two models will be tested to choose the best one. 3.2.4 Spatial Extension of the Model In the spatial topics, the models not accounting for the correlation of observations are likely to cause problematic parameter estimates and incorrect standard error estimates. Thus, many techniques are adopted to solve the spatial dependency problem. Two types of models are employed in the following analysis. The first is Spatial Autocorrelation (SAR) model, in which the dependent variables of other observations will influence a given grid cell. The other is spatially lagged covariate models (SLX), in which the independent variables of other observations affect the dependent variable of a given grid cell. The two are separated to avoid the strong correlation between the lagged X terms and lagged Y terms. 26 Similar to an OLS model, a weighted matrix of all other observations is added to the independent variables. In the original Poisson model, the relationship between 𝜆! and 𝑋! is as follows. 𝜆 = 𝑒+,! ! For SLX model, # 𝜆 = 𝑒(+,!0∑"$%2!","3)! Where 𝑤!" is the spatial weight matrix, 𝑤!"𝑋" is the spatially lagged independent variables of observation j on observation i, 𝜃 measures the relationship between dependent variable and the exogenous variables of other observations. According to LeSage (2008), the direct and indirect impact on the observation i can be measured by 𝛽 and 𝜃. For SAR model, # 𝜆 = 𝑒(+,!0∑"$%2!"*"5)! Where 𝑤!" is the spatial weight matrix, 𝑤!"𝑦" is the spatially lagged dependent variables of observation j on observation i, 𝜌 measures the relationship between dependent variable and the endogenous variables of other observations. The Zero-inflated Poisson model and Zero-inflated Negative Binominal Model are adjusted in the similar way. All the models will be defined and estimated using Python. 27 CHAPTER 4. DATA AND VARIABLES 4.1 Study Area The study area is the city of Shenzhen in China. Shenzhen is a sub-provincial city in Guangdong Province. It is a coastal city located at the South of China and close to Hong Kong. When China was opened to capitalism and foreign investment in 1979, Shenzhen was selected to be the first special economic zone (Gladstone, 2015). With the favorable policies, the young city enjoyed a dramatic growth of economy and a boom of innovation. After 2003, Shenzhen has adjusted its development to introduce a new production method and achieve higher level of innovation. A characteristic is that Shenzhen has some large enterprise but also a lot of SMEs (small and medium enterprises). Based on the complex network-type supply chain relationships between large and medium-sized firms, a more efficient collaborative innovation model has been built up (United Nations Human Settlements Programme, 2019). Figure 4.1 Map of Shenzhen Shenzhen serves as a good case for exploring the new technology, entrepreneurship and innovation, because it is one of the most energetic and open cities in China, widely considered “China’s Silicon Valley”. In July 1995, Shenzhen City defined a transformation strategy with 28 high-tech industry as the forerunner and began to implement a series of policy measures to support the development of high-tech industries (United Nations Human Settlements Programme, 2019; Li, 2000). With attracting research institutes, talents and the build-up of enterprise-oriented and market-oriented innovation system, Shenzhen’s High-tech industries added value became 923 billion RMB in 2019, making up 34.28% of its total GDP, according to Deloitte (2020). Shenzhen also provided a good entrepreneurial environment for early-stage startups, including supportive policies, services and infrastructure. So far, it has 20 unicorn companies such as DJI (a producer of commercial unmanned aerial vehicles), Royole Corporation (a manufacturer of flexible displays and sensors that can be used in a range of human-machine interface products), WeBank (a digital financial service provider for everyone), etc. Within the 12,965 PE/VC-invested startups in IT and internet from 2016 to 2020, Shenzhen has 1,544, accounting for 12% of all deals in China (ITjuzi database, 2021). 4.2 Data Source and Processing 4.2.1 Firm Data To capture new firm activities and agglomeration effect, firm-level data are collected from Shenzhen Administration for Market Regulation. It has a detailed official dataset for all business entities registered from the 1980s to 2021. Different from some widely used datasets as listed firms or Annual Survey for Industrial Firms (ASIF), this dataset gives a good view of newly founded and small firms, which are the focus of this study. The dataset provides firm name, unique record ID, year and date founded, business description, registered capital, firm location and current status (as of Jan 2021). The high-tech (including technology, IT and internet) firms that this study focuses on are selected using relevant keywords in the firm name and description. After geocoding and spatial joining, 79,465 firms founded in year 2000 to 2020 are effectively included in the study, within which 76% are still in operation. For the year 2020, it has 10825 new firms in total. 4.2.2 Incubator Data Along the life cycle of a technology firm, there are different types of incubators in China, from mass maker space (MMS), technology business incubator (TBI) to business accelerators, 29 according to government documents. As the accelerators have just begun to appear in China recently, this study will focus on the first two types of incubators. Tables 4.1 shows a comparison between the two according to policies and government documents, summarizing the difference of their targe groups, function and goals. Table 4.1 Comparison of Mass Maker Space (MMS) and Technology Business Incubators (TBI) Moreover, China also identifies incubators of different administrative levels, from high to low are state-level, provincial-level and city-level. The qualification for government official identification is different (Table 4.2 and 4.3), indicating the potential heterogeneity of the effect. Table 4.2 The Qualification of official identification of Mass Maker Space (MMS) Category Level Qualification State-level operating for more than 18 months before official identification Operation Time Provincial-level operating for more than 12 months before official identification City-level operating for more than 12 months before official identification State-level service space no less than 500 square meters, or 30 working stations Physical Space Provincial-level service space no less than 300 square meters, or 20 working stations City-level service space no less than 500 square meters, or 30 working stations no less than 20 entrepreneurial teams and enterprises each year; The State-level number of registered entrepreneurial teams shall be no less than 10 every year Number of Incubatees Provincial-level no less than 15 entrepreneurial teams and enterprises each year no less than 20 entrepreneurial teams and enterprises each year; The City-level number of registered entrepreneurial teams shall be no less than 6 every year State-level / Investment Fund Provincial-level has the function of angel investment City-level / 30 Table 4.2 continued State-level / has an open online service platform with integrated service capability. Business Service Provincial-level The number of cooperative institutions that actually provide services is Network not less than 5. City-level has signed contracts with more than 6 science and technology service agencies State-level no less than 3 mentors in total Entrepreneur- ship mentorship Provincial-level at least 2 mentors for every 10 entrepreneurial teams and enterprises city-level no less than 3 mentors in total Tables 4.3 The Qualification of official identification of Technology Business Incubator (TBI) Category Level Qualification State-level operating for more than 36 months before official identification Operation Time Provincial-level operating for more than 24 months before official identification City-level operating for more than 24 months before official identification State-level service space of not less than 10,000 square meters Physical Space Provincial-level service space of not less than 6,000 square meters City-level service space of not less than 3,000 square meters State-level incubating no less than 50 enterprises and the accumulative number of graduated enterprises of the incubator should reach more than 20 Number of Provincial-level incubating no less than 35 enterprises and the accumulative number of Incubatees graduated enterprises of the incubator should reach more than 15 City-level incubating no less than 20 enterprises and the accumulative number of graduated enterprises of the incubator should reach more than 8 State-level self-owned seed funds or cooperative incubation funds of no less than RMB 5 million Investment Provincial-level self-owned seed funds or cooperative incubation funds of no less than Fund RMB 3 million City-level self-owned seed funds or cooperative incubation funds of no less than RMB 3 million State-level / has an open online service platform with integrated service capability. Business Service Provincial-level The number of cooperative institutions that actually provide services Network is not less than 5 City-level has signed contracts with more than 6 science and technology service agencies State-level at least 1 mentor for every 10 entrepreneurial teams and enterprises Entrepreneur- ship mentorship Provincial-level at least 2 mentors for every 10 entrepreneurial teams and enterprises city-level No less than 3 mentors in total An official list of technical incubations is acquired from Science and Technology Innovation Committee of Shenzhen Municipality, including 405 incubators founded by the year 2018. 31 This dataset also provides their operating entity. To assess their effect on startup formation, the location data are collected manually and geocoded using Gaode map API. Table 4.4 Summary of Incubator Number Type State-level Provincial and City-level Total Technology Business Incubator (TBI) 30 107 137 Mass Maker Space (MMS) 110 158 268 Figure 4.2 Map of Incubators in Shenzhen (2020) 4.2.3 Venture Capital Firms Venture capital corporation (PE/VC firms) dataset is from IT juzi database (https://www.itjuzi.com), one of the most comprehensive investment databases for technology and internet startups in China. Using years in operation and the number of deals as a filter, the active Venture capital firms are selected for each year and are geocoded using Gaode map API. 4.2.4 Spatial POI data All other control factors including professional business services environment (law and accounting firms, universities), transportation convenience (subway stations coverage and bus stops), centrality (commercial centers), facilities (office buildings and hotels), local amenity 32 (parks and libraries) and land use (green spaces) are assembled from Gaode Map (amap.com) historical POI data using API services. Gaode map is the largest professional map provider in China. All the data are geocoded and then spatially joined to maps using GIS. The coordinates are adjusted to WGS84 and projected. 4.3 Variables Specification As stated above, square grid cells are used as the units of analysis. Shenzhen consists of 2196 grid cells in total. The data in 2020 are merged together to produce a cross-sectional dataset for regression. The assumption is that firms make the location decision based on the spatial condition before their entry, so the independent variables are measured in year 2019. This method has been used by a lot of research (Artz et al., 2016). The variables are as follows. Dependent Variable: Following the research about new firm location choice and spatial distribution (Du and Zheng, 2020; Yao and Hu, 2020), this study chooses the number of new firms in each grid founded in the study year as the dependent variable. Incubator variable: Incubator variable is of the interest of this study. Different Measures of incubators are used in previous research. Wu (2020) uses a binary variable to describe the incubator policy in grids, in which 1 represents having incubators or mass maker spaces and 0 represents not. Li and Zhu (2017) and Duan et al. (2019) use the number of incubators in each subdistrict. Based on the purpose, this study follows the method of using incubator number and generates several variables to describe different types and levels of incubators. Besides the number of all incubators (Incubator), it includes the number of technology business incubators (TBI), mass maker spaces (MMS) and their levels (L1 for state-level and L2 for provincial and city-level). The model examines the effect of these three types of variables separately. Control Variables: Drawing upon the factors discussed by existing literature, control variables in the model are categorized into the following 5 groups, including agglomeration, centrality, professional business environment, transportation, and local facility and amenity. 1) Agglomeration: Agglomeration effect is measured using the number of incumbent large 33 high-tech firms in the same grid cell, following the measurement employed by Li and Zhu (2017) and Du and Zheng (2020). 2) Centrality: Centrality is measured using the distance of the grid cell centroid to the nearest commercial centers. Commercial centers are calculated using K-means clustering method based on major shopping centers. 3) Professional Business Environment: Professional business environment describes the favorable factors in a region for a new business development. Four variables are included, including distance to investment (PE/VC firms), distance to law firms, distance to accounting firms and distance to universities. 4) Transportation Convenience: Transportation has two variables, the accessibility to subway stations and bus stops, which are two most widely used public transportation in Chinese urban area. Subway uses a ratio indicator. This is because the accessibility to subway is more like a binary variable. Whether people can reach a subway station in a threshold (often around 10-15 minutes) is important, but within or without the threshold, the time doesn’t make much difference. Thus, a buffer of 1km of subway stations is made in GIS and a ratio of buffer to the total land area of each grid is calculated. The overlapping only adds ones, because more available subways don’t increase the utility of people. 𝑙𝑎𝑛𝑑 𝑎𝑟𝑒𝑎 𝑐𝑜𝑣𝑒𝑟𝑒𝑑 𝑏𝑦 𝑡ℎ𝑒 1 𝑘𝑚 𝑏𝑢𝑓𝑓𝑒𝑟 𝑜𝑓 𝑎 𝑠𝑢𝑏𝑤𝑎𝑦 𝑠𝑡𝑎𝑡𝑖𝑜𝑛 𝑠𝑢𝑏𝑤𝑎𝑦 = 𝑡𝑜𝑡𝑎𝑙 𝑙𝑎𝑛𝑑 𝑎𝑟𝑒𝑎 𝑜𝑓 𝑎 𝑔𝑟𝑖𝑑 5) Local Facilities and Amenity: The number of office buildings, hotels of three-star or higher in the grid are used to assess the local facility. The distance from the grid centroid to the nearest park. Green land percentage and a dummy variable indicating whether the land is unusable for business development is also added. While some subdistrict level analysis has population, labor and land price indicators, the factors are not included in this research as they tend to influence at a larger geographical scale. The following table shows the definition of all the variables. Table 4.5 Definition of dependent and independent variables Variables Definition Expected Sign Dependent Variable New firms n_firms Number of new-born high-tech firms in the year Independent Variables Incubation Incubator Number of incubators in a grid (TBI and MMS at all levels) + 34 Table 4.5 continued TBI Number of Technology Business Incubators (TBI) in a grid + MMS Number of Mass Maker Spaces (MMS) in a grid + TBI_L1 Number of State-level Technology Business Incubators (TBI) in a grid + TBI_L2 Number of provincial and city-level Technology Business Incubators (TBI) in a grid + MMS_L1 Number of state-level Mass Maker Spaces (MMS) in a grid + MMS_L2 Number of provincial and city-level Mass Maker Spaces (MMS) in a grid + Agglomeration Industry Number for the firms founded in the past 20 years with registered capital over RMB 10 million + Centrality Dis2Cntr Distance between a grid's centroid to the nearest commercial center - Dist2Inv Distance between a grid's centroid to the nearest PE/VC firm - Professional Dis2Law Distance between a grid's centroid to the nearest law firm - Business Environment Dis2Acc Distance between a grid's centroid to the nearest accounting firm - Dis2Univ Distance between a grid's centroid to the nearest university or college - Percentage of grid area within the 1000-meter buffer of subway Transport Subway stations + Convenience Busstop Number of bus stops in a grid + Office Number of office buildings in a grid + Hotel Number of three-star and above hotels in a grid + Local Facility Dis2Park Distance between a grid's centroid to the nearest park or plaza - and Amenity Library Number of libraries in a grid + Green Land use variable: percentage for green area + Unusable Dummy variable whether a grid is usable to host any firms - 35 CHAPTER 5. DESCRIPTION OF HIGH-TECH STARTUP SPATIAL PATTERN To explore the spatial distribution of high-tech firms, it is useful to look back on the past 20 years to understand its growth and characteristics. There is a dramatic increase of the new high-tech startups in Shenzhen in recent years. The total number of new firms born in 2005 is only 446, which becomes over 26 times in 2020, reached 11,691 new firms. The number started to increase quickly after 2010 and then had a boom in 2015, following the concept of “Mass Entrepreneurship and Innovation” put forward by Chinese government in late 2014 and the "Internet Plus" initiative proposed in 2015. 12000 10000 8000 6000 4000 2000 0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Figure 5.1 The number of new technology and internet firms in Shenzhen (2000-2020) Table 5.1 Summary of Statistics of New IT firms by Grids (2005-2020) Year N Sum of firms range mean Std. Dev. Grid with firms 2005 2196 446 0-11 0.203 0.765 257 2010 2196 670 0-18 0.305 1.129 304 2015 2196 9,773 0-2828 4.450 61.080 767 2020 2196 11,691 0-680 5.324 20.841 860 As for the spatial pattern, three characteristics can be recognized. First, it shows a strong path dependency trend and a reinforcement of existing agglomerations. Newly formed firms tend to cluster at existing agglomeration of large technology firms, universities and institutes, reinforcing the Qianhai cluster at the southwest part of Nanshan District. An unbalance of new firm location become strongest in 2015. The grid with most new firms has reached over 2000 new firms, making nearly 30% of new firms of the year. The situation becomes less severe in 2020, but agglomeration is still dominant in the spatial 36 distribution pattern. Second, there is a dynamic dispersion of high-tech startup birth activities and a continuous formation of new entrepreneurship hubs. As shown in figure 5.2, most new technology and internet firms were located only at some discrete places around the western coastal area in 2005. The business areas grew gradually and by the year of 2015, three large entrepreneurial centers were gradually formed, which were in Nanshan district (including central part of Nanshan and Qianhai area), Futian Central District and Bantian District. After 2015, the spillover effect became more important in shaping the geographical pattern. The clusters became larger and connected each other. Some remote regions like Longgang and Pingshan in the eastern part also started to have their high-tech firms. Then in 2020, another large cluster was formed at Longgang, which is previously a manufacturing-oriented region, but recent years developed a trillion RMB-level ICT industry cluster, the hundred billion RMB-level AIoT (Artificial Intelligence Internet of Things) industry cluster. (2005) (2010) (2015) (2020) Figure 5.2 Map of new high-tech firm location (2005-2020) Third, there is strong evidence for spatial autocorrelation. The global Moran’s I test based on 37 the count of new firms in the past 20 years shows the firm distribution is clustered rather than randomly distributed (table 5.2). The autocorrelation means that the firm distribution in nearby areas may affect the location decision of entrepreneurs. This fact confirms that the spatial justification to the original regression model is necessary. Two kinds of weight matrix are used, queen and KNN. Except for the year 2015 due to an outlier grid, all other years show strong spatial autocorrelation at 0.01 level. Table 5.2 Moran’s I test for the new firm (2000-2020) KNN KNN KNN Year Queen p value p value p value p value K=4 K=5 K=10 2000 0.045 0.000 0.234 0.000 0.222 0.000 0.199 0.000 2005 0.097 0.000 0.290 0.000 0.269 0.000 0.227 0.000 2010 0.081 0.000 0.344 0.000 0.318 0.000 0.298 0.000 2015 0.002 0.861 0.029 0.052 0.025 0.054 0.024 0.008 2020 0.063 0.000 0.289 0.000 0.270 0.000 0.256 0.000 Regarding the local Moran’s I test, figure 6.3 shows the LISA cluster map based on the queen weight matrix. It shows the heterogeneity in the distribution of new high-tech firms. The south and west part of Shenzhen (Nanshan and Futian Districts) is where most HH observations locate, showing the possibility of positive effect on the nearby area. 38 (2005) (2010) (2015) (2020) Figure 5.3 LISA cluster map for new firm numbers (2005-2020) 39 CHAPTER 6. HOW DO INCUBATORS AFFECT HIGH-TECH STARTUP SPATIAL PATTERN? 6.1 Summary and Test of Data The empirical model uses the data of 2020 with different types of incubators and control variables to test the significance and heterogeneity of incubator effects. One outlier with over 600 businesses due to a special policy of affiliated business is dropped for the regression. There are 2195 observations in total. The unusable is a binary variable to control the land use. 204 of the grids are hills or parks that cannot host business. A statistical summary of variables is shown as table 6.1. Table 6.1 Statistics of Variables Variable N sum min max mean Std.Dev. n_firms 2195 10825 0 234 4.93 14.75 Incubator-all 2195 404 0 24 0.18 0.93 Incubator- TBI 2195 137 0 12 0.06 0.4 Incubator- MMS 2195 267 0 12 0.12 0.63 Incubator- TBI_L1 2195 30 0 3 0.01 0.13 Incubator- TBI_L2 2195 107 0 9 0.05 0.33 Incubator- MMS_L1 2195 110 0 7 0.05 0.36 Incubator- MMS_L2 2195 157 0 6 0.07 0.35 Industry 2195 4381 0 189 2 8.26 Dis2Cntr 2195 19218 0 19 8.76 3.78 Dis2Inv 2195 12138 0 20 5.53 4.43 is2Acc 2195 14038 0 33 6.4 7.02 Dis2Law 2195 6160 0 14 2.81 2.33 40 Table 6.1 continued Dis2Univ 2195 7754 0 15 3.53 2.46 Subway 2195 63049 0 100 28.72 40.03 Busstop 2195 6099 0 29 2.78 3.64 Office 2195 6441 0 172 2.93 7.97 Hotel 2195 616 0 20 0.28 0.97 Dis2Park 2195 12107 0 19 5.52 3.66 Library 2195 311 0 6 0.14 0.51 Green 2195 54313 0 100 24.74 36.56 Unusable 2195 204 0 1 0.09 0.29 To exclude the problem of multicollinearity, correlation and VIF tests are used. Table 6.2 shows the Pearson correlation of all the independent variables. According to experience, 0.7 is set as a threshold for collinearity. Distance to a law firm (Dis2Law) is found to have strong correlation with distance to an accounting firm (Dis2Acc) and distance to a university (Dis2Univ), with Correlation of 0.74 and 0.76 separately. Therefore, Dis2Law is considered to be excluded in the model. Table 6.2 Correlation between independent variables The VIF test (1) gives the similar result as correlation analysis, showing a high VIF statistic (9.59) for distance to a law firm. After excluding Dis2Law, other variables are acceptable, using VIF < 10 as a threshold suggested by a lot of research. The distance to park (Dis2Park) also has a high VIF, but this variable is found to be very significant in the following 41 regressions and greatly influence the accuracy of the model prediction. Therefore, Dis2Park is kept in the model as it still meets the requirement of VIF < 10. As there is perfect collinearity among the incubator variables, they will be employed to model one by one, as shown in VIF (2)-(4). Table 6.3 VIF tests of the variables Variables VIF (1) VIF (2) VIF (3) VIF (4) Incubator 1.29 1.29 TBI 1.68 MMS 1.92 TBI_L1 1.31 TBI_L2 1.73 MMS_L1 2.04 MMS_L2 1.71 Industry 1.04 1.04 1.04 1.05 Dis2Cntr 6.98 6.90 6.90 6.91 Dis2Inv 4.69 4.43 4.43 4.44 Dis2Acc 5.49 4.74 4.74 4.74 Dis2Law 9.59 Dis2Univ 7.05 5.38 5.38 5.38 Subway 2.31 2.30 2.30 2.30 Busstop 3.20 3.07 3.07 3.08 Office 2.63 2.63 2.65 2.65 Hotel 1.84 1.84 1.84 1.84 Dis2Park 8.84 8.79 8.79 8.81 Coffee 2.42 2.41 2.42 2.43 Library 1.36 1.36 1.37 1.37 Green 2.98 2.96 2.96 2.96 42 Table 6.3 continued Unusable 1.96 1.96 1.96 1.96 6.2 Model Selection 6.2.1 Basic Model Selection Before testing the models, the distribution of the dependent variable, the number of new firms, is calculated (table 6.4) to better understand the features of data. Four basic models are then run to select the most suitable one, including Poisson model, Negative Binominal model, Zero-inflated Poisson model and Zero-inflated Negative Binominal model (table 6.5). There are many criteria that can be employed for model selection, and it is always a trade-off between the performance from different dimensions. Four types of indicators are calculated for model comparison: 1) the smaller Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) indicate a better goodness-of-fit. (2) A smaller value of Akaike Information Criterion (AIC) indicates a better quality of the model. (3) A larger Log- likelihood and Pseudo R-square indicates the goodness of fit. (4) Some data points are predicted and plotted as a test of model. The visualization of the actual counts and predictions gives a pattern of the model fit (figure 6.1). The analysis and result of selection process of this study is as follows. First, both the NB model and ZINB models are rejected for much larger RMSEs and failure to predict the points with higher values (figure 6.1). Although they have lower AIC scores in total, they do not perform well in predict non-zero values, which may indicate that negative binominal regression is not a good fit of the data. Then, comparing the Poisson and Zero-inflated Poisson (ZIP) models, the latter fit the data better as the data contains an excess of zeroes (table 6.4). The statistics including Log- likelihood, AIC, RMSE and MAPE all show that ZIP model is better. The plots also show a better fitness of the ZIP model. Taking a zero-inflated Poisson regression models, the hypothesis is that there is a process determining whether a grid cell will have new firms or not. Once a cell has new firms, the counts of firms is determined by the Poisson probability mass function. For each process, the relationship will be discussed later in the spatial and more detailed analysis. 43 Based on the above discussion, the Zero-inflated Poisson (ZIP) Regression model is chosen for further spatial analysis. Table 6.4 Categories of number of firms by grid cells Number of Firms count of grid cells 0 1493 (1, 5] 302 (5, 10] 130 (10, 15] 79 (15, 20] 51 (20, 30] 50 (30, 50] 47 (50, 80] 27 (100, 200] 10 (80, 100] 5 (200, 300] 1 Table 6.5 Comparison of four basic models Poisson Model NB Model ZIP Model ZINB Model Incubator 0.03*** 0.14*** 0.03*** 0.04 0 -0.05 0 -0.03 Industry 0.02*** 0.06*** 0.02*** 0.04*** 0 -0.01 0 0 Dis2Cntr -0.07*** -0.07*** -0.05*** -0.05*** 0 -0.01 0 -0.01 Dis2Inv -0.11*** -0.07*** -0.11*** -0.07*** -0.01 -0.01 -0.01 -0.01 Dis2Acc -0.26*** -0.18*** -0.17*** -0.13*** -0.01 -0.02 -0.01 -0.02 Dis2Univ -0.18*** -0.13*** -0.11*** -0.05* -0.01 -0.03 -0.01 -0.03 Subway 0.01*** 0.01*** 0.00*** 0.00*** 0 0 0 0 Busstop 0.03*** 0.13*** 0.01* 0.01 0 -0.01 0 -0.01 Office 0.00*** 0.02** 0.00*** 0.02*** 44 Table 6.5 continued 0 -0.01 0 0 Hotel -0.01 -0.10** 0.01* -0.03 0 -0.05 0 -0.03 Dis2Park 0.02*** 0.02 0.01 0.03* -0.01 -0.02 -0.01 -0.02 Library 0.04*** 0.02 0.04*** 0.01 -0.01 -0.07 -0.01 -0.04 Green -0.01*** -0.00** -0.01*** -0.00* 0 0 0 0 Unusable -2.15*** -1.98*** -1.10*** -0.95 -0.36 -0.41 -0.39 -0.6 Intercept 2.99*** 1.62*** 3.24*** 2.30*** -0.05 -0.16 -0.05 -0.15 inflate_Incubator -1.08*** -1.40** -0.32 -0.71 inflate_Industry -1.03*** -2.44*** -0.14 -0.46 inflate_Dis2Cntr 0.03 0.03 -0.02 -0.03 inflate_Dis2Inv -0.02 -0.02 -0.02 -0.03 inflate_Dis2Acc 0.06* 0.06 -0.03 -0.04 inflate_Dis2Univ 0.05 0.1 -0.06 -0.08 inflate_Subway 0 0 0 0 inflate_Busstop -0.29*** -0.41*** -0.04 -0.06 inflate_Office -0.24*** -0.42*** -0.05 -0.1 inflate_Hotel -0.08 0.05 -0.15 -0.22 inflate_Dis2Park 0 0.01 -0.03 -0.05 45 Table 6.5 continued inflate_Library -0.18 -0.07 -0.23 -0.38 inflate_Green 0 0 0 0 inflate_Unusable 1.05* 1.35* -0.61 -0.78 inflate_Intercept 1.53*** 1.38*** -0.31 -0.39 alpha 1.55*** 0.67*** -0.09 -0.05 N 2195 2195 2195 2195 Log-likelihood -6596.7 -3451 -5076.4 -3144.9 Pseudo R2 0.68 0.2 0.56 0.27 AIC 13223.34 6933.96 10182.83 6319.81 RMSE 9.3 53646.23 8.67 1358.3 MAPE 0.51 0.46 0.37 0.35 Figure 6.1 Plots of predicted and actual counts of firms using different models 46 6.2.2 Spatial Extension Exploration As the grid cell is small in geographical scale, there is great concern that the adjacent characteristics can also affect the result. The significant Moran’s I statistic also indicates strong positive spatial autocorrelation. Therefore, the spatial lag model is employed to adjust the spatial dependency and refine the previous models. Two types of spatial lag terms are examined here in this paper, spatial lag of Y model (SAR) and spatial lag of X term (SLX). The former tests the effect of number of new firms of adjacent area on the given grid cell, while the later assesses the exogenous factors of surrounding area on the given grid cell. By analyzing the coefficients of original and lag terms, the direct and indirect effects of independent variables can be explored. For the SLX model, all the distance variables are dropped for spatial terms, as spatial terms would have strong linear relation with the original distance variables. Tables 6.6 shows the results of non-spatial ZIP model, SAR-ZIP model and SLX-ZIP model. The Y lag term as well as some of the X lag terms are significant in model 2 and 3, showing that the hypothesis of spatial independency is false. The queen spatial weight matrix is employed and the variable starting with “W_” are spatial terms calculated using adjacent grid cells. The spatial weight matrix is row standardized. For example, the positive coefficient of W_n_firms shows that the count of the firms in the surrounding area has positive effect on the new high-tech firm agglomeration. The negative sign of incubators is likely to indicate a competing relationship between the neighboring areas. The more incubators in the nearby area, the fewer the new firms in a given grid cell. Thus, both SLX model and SAR model will be tested to explore the effect of incubators and its heterogeneity. Table 6.6 Comparison of non-spatial and spatial ZIP models ZIP model SZIP model - SAR SZIP model - SLX Incubator 0.03*** 0.02*** 0.03*** 0.00 0.00 0.00 Industry 0.02*** 0.02*** 0.02*** 0.00 0.00 0.00 Dis2Cntr -0.05*** -0.05*** -0.05*** 0.00 0.00 0.00 Dis2Inv -0.11*** -0.10*** -0.10*** -0.01 -0.01 -0.01 Dis2Acc -0.17*** -0.17*** -0.13*** 47 Table 6.6 continued -0.01 -0.01 -0.01 Dis2Univ -0.11*** -0.12*** -0.07*** -0.01 -0.01 -0.01 Subway 0.00*** 0.00*** 0.00*** 0.00 0.00 0.00 Busstop 0.01* 0.00 0.00 0.00 0.00 0.00 Office 0.00*** 0.00*** 0.00*** 0.00 0.00 0.00 Hotel 0.01* 0.00 -0.01** 0.00 0.00 -0.01 Dis2Park 0.01 0.02** 0.02*** -0.01 -0.01 -0.01 Library 0.04*** 0.04*** 0.03*** -0.01 -0.01 -0.01 Green -0.01*** -0.01*** -0.01*** 0.00 0.00 0.00 Unusable -1.10*** -1.07*** -1.04** -0.39 -0.39 -0.4 W_n_firms 0.01*** 0.00 W_Incubator -0.03** -0.01 W_Industry 0.00 0.00 W_Subway 0.00** 0.00 W_Busstop 0.01 -0.01 W_Office 0.02*** 0.00 W_Hotel -0.03** -0.01 W_Library 0.22*** -0.03 W_Green -0.00** 48 Table 6.6 continued 0.00 W_Unusable 1.27*** -0.32 Intercept 3.24*** 3.10*** 2.73*** -0.05 -0.05 -0.08 inflate_Incubator -1.08*** -1.05*** -1.18*** -0.32 -0.32 -0.32 inflate_Industry -1.03*** -1.01*** -0.53*** -0.14 -0.15 -0.09 inflate_Dis2Cntr 0.03 0.02 0.00 -0.02 -0.02 -0.02 inflate_Dis2Inv -0.02 -0.03 0.00 -0.02 -0.02 -0.03 inflate_Dis2Acc 0.06* 0.04 0.04 -0.03 -0.03 -0.03 inflate_Dis2Univ 0.05 0.03 0.01 -0.06 -0.06 -0.06 inflate_Subway 0.00 0.00 -0.01** 0.00 0.00 0.00 inflate_Busstop -0.29*** -0.30*** -0.21*** -0.04 -0.04 -0.04 inflate_Office -0.24*** -0.22*** -0.15*** -0.05 -0.05 -0.05 inflate_Hotel -0.08 -0.06 -0.11 -0.15 -0.15 -0.15 inflate_Dis2Park 0.00 -0.01 0.00 -0.03 -0.03 -0.03 inflate_Library -0.18 -0.18 -0.33 -0.23 -0.23 -0.24 inflate_Green 0.00 0.00 0.00 0.00 0.00 0.00 inflate_Unusable 1.05* 0.99 0.66 -0.61 -0.61 -0.62 inflate_W_n_firms -0.05*** -0.02 inflate_W_Incubator -0.56* 49 Table 6.6 continued -0.3 inflate_W_Industry 0.01** 0.00 inflate_W_Subway 0.02*** -0.01 inflate_W_Busstop -0.50*** -0.07 inflate_W_Office -0.02 -0.05 inflate_W_Hotel -0.03 -0.26 inflate_W_Library 0.59 -0.47 inflate_W_Green -0.01** -0.01 inflate_W_Unusable 2.17** -1.02 inflate_Intercept 1.53*** 1.86*** 2.78*** -0.31 -0.34 -0.41 N 2195 2195 2195 Log-likelihood -5076.4 -5020.3 -4872.9 Pseudo R2 0.56 0.57 0.58 AIC 10182.83 10072.62 9793.82 6.3 Model Results and Interpretation Based on the model and spatial term selection in the last part, both Y lag and X lag SZIP models are employed to examine the relationship between incubators and the counts of new firms, controlling other factors and spatial effects. Section 6.3.1 and 6.3.2 discusses the determinants of new high-tech firm numbers and the existence of new high-tech firms separately. The first three models are SAR and the latter three are SLX. 6.3.1 Determinants of New High-tech Firm Distribution To explore the influential factors of the number of local new firms, we focus on the non- inflated variables. Models 1 and 4 show a significantly positive coefficient of Incubator-all, 50 meaning the total number of incubators has a positive effect of new firm birth. For the new tech companies, the policy and entrepreneurship convenience provided by incubators could be an attractiveness for their starting the business. The result is consistent with many of the existing research, and further confirms that incubators attract new high-tech firms at a smaller geographical scale. Breaking it down to the types of incubators, there is a significant and negative effect of Technology Business Incubators (TBI) but a positive effect of Mass Maker Spaces (MMS), showing that the features and influence are different between the two (model 2 and 5). TBIs do not show significant attractiveness to new high-tech firms, while the MMSs seem to perform better in attracting new firms. According to the detailed information of incubator data, the TBIs are more government-backed and many of them are affiliated to government departments, development zones, technology institutes and large universities, for example, among them are Shenzhen Institute of Advanced Technology Chinese Academy of Science, Futian District High-tech Entrepreneurship Center Incubator and PKU-HKUST Shenzhen– Hong Kong Institution. Many have been developed for a long time, for example over 10 years. As a comparison, MMSs are more diverse and market oriented. According to the country-level data, over 70% of MMSs are private-owned. For example, the well-known Y Combinator and TechCode are all large international incubation networks. There are also many relatively small and local mass maker spaces such as Open-source Maker Space and Inno Park. They are more flexible and easier to enter and exit for new firms and emphasize more on soft services (6 activities per year required for city-level and 10 activities required for state and provincial- level). The difference to some extent indicates stronger competitiveness and attractiveness of market-oriented and diverse mass maker spaces to the new high-tech firms. Models 3 and 6 further discuss the difference between different levels of incubators, state- level (L1) and provincial and city level (L2). The results show that the difference between the levels of TBI is not significant, while different levels of MMS show opposite effect. The state- level MMS has significant negative effect while city and provincial-level has positive effect. State-level incubators often have higher requirement for size, age, and services. For example, an MMS could only apply for the state-level identification after 18 months’ operation, while a city-level identification only takes 12 months. The state-level MMS has higher requirements for the entry of new firms and startup graduation rate. One possible explanation worth testing is that higher entry requirements, greater competition and limited space might push the new firms away. 51 As for the control variables, most of them are consistent with previous studies. The industry agglomeration of indicated by large firms (Industry), subway station (Subway), the number of office buildings (Office) and libraries (Library) have significantly positive effect. The coefficients of distance to commercial center (Dis2Cntr), investment firms (Dis2Inv), accounting firms (Dis2Acc) and Universities (Dis2Univ) are significantly negative, meaning that the closer the factors, the more the new firms. The distance to park (Dis2Park) has positive sign, which is not intuitive, as entrepreneurs and creative class are found to care about the local amenity and physical environment. This could attribute to the geography of Shenzhen, where there is large forest coverage as 44.6%. The area around parks, especially in some remote area, are often not attractive for business due to inconvenience and lack of development. The green area (Green) and unusable dummy (Unusable) variable show the land use has significant effect. The spatial terms are also significant and show the indirect effect from surrounding area. The spatial lag term of Y is positive in models 1 to 3, showing reinforcement process of new firm agglomeration and positive spatial autocorrelation. A new high-tech business is more likely to enter an area if the surrounding areas are hosting more new firms. However, most of the lag terms of incubators are negative, showing that it has significant negative indirect effect on nearby area. This also shows the competition of neighboring areas. New firms can be attracted away by nearby incubators on other grid cells. This fact kind of confirms that incubators’ effect at small scales. 6.3.2 Determinants of the Possibility of New High-tech Firm Existence Besides the count of firms, the Zero-inflated Poisson model also provides the technique to understand the factors influencing whether a grid cell has any new high-tech firms. The logit model is used to deal with the excess of zeros. The coefficients of “Inflated” variables show the relationship between factors and zero values. A significant and positive coefficient means that the larger value the factor has, the higher likelihood for “zero” new firms. A significant and negative coefficient means that the larger the factor, the higher likelihood for the existence of new firms. The coefficient of the total number of incubators is significantly negative at 0.01 level, indicating the having an incubator of any type is positively associated with existence of new high-tech firms. When breaking down to Technology Business Incubator (TBI) and Mass Maker Space (MMS), both coefficients are negative and the value has little difference, 52 meaning that both TBI and MMS encourage the existence of new high-tech firms, and the extent of effect is similar. When it comes to the level of incubators, only the provincial and city-level MMSs have significant negative effect on “zero” firms. For the control variables, the result also makes sense as industrial agglomeration of large firms, bus stops and office building all have significant positive influence on the birth of new firms in a grid. For the spatial terms, the nearby factors are less significant for the existence than for the count of new firms. The coefficient of Inflate_W_n_firms is negative, showing the more the new firms in the adjacent lands, the higher likelihood of for the existence of new high-tech firms in a given grid cell. The coefficient of inflate_W_Incubator is significant at 0.1 level, showing the total number of surrounding incubators has positive effect on firm birth. But the effect of most of the single types and levels of incubators are not such significant (model 4 to 6). Table 6.7 Spatial ZIP model of the number of new high-tech firms SZIP – SAR SZIP - SAR SZIP - SAR SZIP - SLX SZIP - SLX SZIP - SLX (1) (2) (3) (4) (5) (6) Incubator - all 0.02*** 0.03*** 0 0 Incubator-TBI -0.03** -0.05*** -0.01 -0.01 Incubator-MMS 0.06*** 0.10*** -0.01 -0.01 Incubator-TBI_L1 0.01 -0.02 -0.03 -0.04 Incubator-TBI_L2 -0.04** -0.01 -0.02 -0.02 Incubator-MMS_L1 -0.12*** -0.13*** -0.02 -0.02 Incubator-MMS_L2 0.23*** 0.25*** -0.01 -0.01 Industry 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0 0 0 0 0 0 Dis2Cntr -0.05*** -0.05*** -0.05*** -0.05*** -0.05*** -0.05*** 0 0 0 0 0 0 Dis2Inv -0.10*** -0.10*** -0.09*** -0.10*** -0.10*** -0.09*** 53 Table 6.7 continued -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 Dis2Acc -0.17*** -0.17*** -0.16*** -0.13*** -0.12*** -0.12*** -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 Dis2Univ -0.12*** -0.12*** -0.13*** -0.07*** -0.07*** -0.07*** -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 Subway 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0 0 0 0 0 0 Busstop 0 0 0 0 0 0 0 0 0 0 0 0 Office 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0 0 0 0 0 0 Hotel 0 0 0 -0.01** -0.01** -0.02*** 0 0 0 -0.01 -0.01 -0.01 Dis2Park 0.02** 0.02** 0.01** 0.02*** 0.02*** 0.02*** -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 Library 0.04*** 0.03*** 0.03** 0.03*** 0.03** 0.01 -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 Green -0.01*** -0.01*** -0.01*** -0.01*** -0.01*** -0.01*** 0 0 0 0 0 0 Unusable -1.07*** -1.03*** -0.92** -1.04** -0.97** -0.84** -0.39 -0.39 -0.39 -0.4 -0.4 -0.4 W_n_firms 0.01*** 0.01*** 0.00*** 0 0 0 W_Incubator -0.03** -0.01 W_TBI -0.19*** -0.05 W_MMS 0.04 -0.03 W_TBI_L1 0.68*** -0.15 W_TBI_L2 -0.16** -0.07 W_MMS_L1 -0.11** -0.04 W_MMS_L2 0.14** 54 Table 6.7 continued -0.06 W_Industry 0 0 -0.00** 0 0 0 W_Subway 0.00** 0.00** 0 0 0 0 W_Busstop 0.01 0.01 0.02** -0.01 -0.01 -0.01 W_Office 0.02*** 0.02*** 0.01*** 0 0 0 W_Hotel -0.03** -0.04*** -0.02* -0.01 -0.01 -0.01 W_Library 0.22*** 0.20*** 0.16*** -0.03 -0.03 -0.03 W_Green -0.00** -0.00** 0 0 0 0 W_Unusable 1.27*** 1.18*** 1.32*** -0.32 -0.32 -0.32 inflate_Incubator -1.05*** -1.18*** -0.32 -0.32 inflate_TBI -1.15* -1.10* -0.63 -0.59 inflate_MMS -1.00*** -1.01*** -0.39 -0.39 inflate_TBI_L1 -1.58 -1.5 -1.21 -1.19 inflate_TBI_L2 -0.94 -0.96 -0.71 -0.68 inflate_MMS_L1 -0.61 -0.62 -0.85 -0.83 inflate_MMS_L2 -1.08** -1.10** -0.48 -0.48 inflate_Industry -1.01*** -1.00*** -0.99*** -0.53*** -0.92*** -0.94*** -0.15 -0.15 -0.15 -0.09 -0.14 -0.14 inflate_Dis2Cntr 0.02 0.02 0.02 0 0 0 -0.02 -0.02 -0.02 -0.02 -0.02 -0.03 inflate_Dis2Inv -0.03 -0.03 -0.03 0 0 0 55 Table 6.7 continued -0.02 -0.02 -0.02 -0.03 -0.03 -0.03 inflate_Dis2Acc 0.04 0.04 0.05 0.04 0.04 0.04 -0.03 -0.03 -0.03 -0.03 -0.03 -0.03 inflate_Dis2Univ 0.03 0.03 0.03 0.01 0.01 0 -0.06 -0.06 -0.06 -0.06 -0.06 -0.06 inflate_Subway 0 0 0 -0.01** -0.01** -0.01** 0 0 0 0 0 0 inflate_Busstop -0.30*** -0.30*** -0.30*** -0.21*** -0.21*** -0.21*** -0.04 -0.04 -0.04 -0.04 -0.04 -0.04 inflate_Office -0.22*** -0.22*** -0.22*** -0.15*** -0.16*** -0.15*** -0.05 -0.05 -0.05 -0.05 -0.05 -0.05 inflate_Hotel -0.06 -0.06 -0.06 -0.11 -0.07 -0.08 -0.15 -0.15 -0.15 -0.15 -0.15 -0.16 inflate_Dis2Park -0.01 -0.01 -0.01 0 0 -0.01 -0.03 -0.03 -0.03 -0.03 -0.04 -0.04 inflate_Library -0.18 -0.18 -0.18 -0.33 -0.27 -0.26 -0.23 -0.23 -0.23 -0.24 -0.24 -0.24 inflate_Green 0 0 0 0 0 0 0 0 0 0 0 0 inflate_Unusable 0.99 1 1.04* 0.66 0.76 0.79 -0.61 -0.61 -0.61 -0.62 -0.65 -0.65 inflate_W_n_firms -0.05*** -0.05*** -0.05*** -0.02 -0.02 -0.02 inflate_W_Incubator -0.56* -0.3 inflate_W_TBI -0.33 -0.83 inflate_W_MMS -0.67 -0.52 inflate_W_TBI_L1 -1.2 -1.97 inflate_W_TBI_L2 -0.68 -0.99 inflate_W_MMS_L1 2.32* -1.26 inflate_W_MMS_L2 -2.64*** 56 Table 6.7 continued -0.94 inflate_W_Industry 0.01** 0 0.01 0 -0.01 -0.01 inflate_W_Subway 0.02*** 0.02*** 0.01** -0.01 -0.01 -0.01 inflate_W_Busstop -0.50*** -0.49*** -0.49*** -0.07 -0.07 -0.07 inflate_W_Office -0.02 0 0.01 -0.05 -0.05 -0.05 inflate_W_Hotel -0.03 -0.05 -0.05 -0.26 -0.27 -0.27 inflate_W_Library 0.59 0.61 0.63 -0.47 -0.49 -0.49 inflate_W_Green -0.01** -0.01** -0.01** -0.01 -0.01 -0.01 inflate_W_Unusable 2.17** 2.06** 2.17** -1.02 -1.04 -1.05 Intercept 3.10*** 3.11*** 3.06*** 2.73*** 2.74*** 2.64*** -0.05 -0.05 -0.05 -0.08 -0.08 -0.08 inflate_Intercept 1.86*** 1.87*** 1.84*** 2.78*** 2.87*** 3.01*** -0.34 -0.34 -0.34 -0.41 -0.42 -0.43 N 2195 2195 2195 2195 2195 2195 Log-likelihood -5020.3 -5007.5 -4883.2 -4872.9 -4842.7 -4725.5 Pseudo R2 0.57 0.57 0.58 0.58 0.58 0.59 AIC 10072.62 10049.01 9804.38 9793.82 9737.3 9511.01 6.4 Summary After selecting the most suitable model, adding control variables, and adjusting the spatial effects, the regression result confirms that incubators have a significant positive relationship with new high-tech startup location. As a whole, a larger number of incubators in a grid cell tend to encourage the existence and also positively affect the amount of newly formed high- tech firms. However, when it comes to different types and levels of incubators, there exists great heterogeneity. Although the effect on existence of new high-tech firms tests to be similar, MMS has a positive relationship with firm numbers while TBI has a negative 57 relationship. It shows that MMS performs better than TBI in attracting new firms. As for the level of incubators, only the provincial and city-level MMS shows a significant positive effect on both new firm birth and counts. As for the indirect effect from adjacent area, incubators have a negative effect, meaning that neighboring areas are in competition of new high-tech firms, confirming that incubators can make influence at a small scale. This result is very interesting as the effect of TBI is far from the expectation at least recently. The government or institute backed TBIs often have better hardware, resources input and policy supporting. Many of them have developed for a long time but failed to encourage the new firm spatial aggregation recently. MMS, especially lower-level MMS, seems to do better. Based on recent research in China, some possible explanations are as follows: (1) Most technology business incubators are supported by the government and are non-profit. Their income is largely based on rent and subsidies and thus they lack market vitality and competitiveness. This may make them less attractive (Wang, 2010). (2) During the past years, the identification of incubators emphasizes more on its input rather than result or efficiency. This causes redundant input and unreasonable resource allocation. Some large state-owned incubators are running in low efficiency while the development of private-owned incubators is hindered by size (Yang, 2020). (3) Compared with MMS, TBI is not as good in market association, creativity and globalization (Wang, 2010; He and Hu, 2016). (4) Many MMSs are more focusing on niche market, meaning they are more professional in specific industries. Moreover, it is easier for a new firm to enter an MMS than TBI, may making their effect stronger. As for other variables influencing new high-tech firm location, existing industry agglomeration, convenient public transportation, business environment and office buildings all have significant positive effects, which is consistent with previous studies. 58 CHAPTER 7. CONCLUSION AND DISCUSSION 7.1 Conclusion Entrepreneurship and the relevant policy instruments including incubation have been a fever in China since around 2014. Technology Business Incubators and Mass Maker Spaces assist the formation, commercialization, development, and success of new high-tech firms, and thus impact the spatial pattern of such new startups. A lot of studies have discussed the determinants of new firm spatial distribution, while the empirical research on the incubators’ effect is insufficient in quantity and the level of detail. Thus, this study follows the line of existing literature on new firm distribution but further explores the effect of incubators. Using the latest firm-level data of 2020 in Shenzhen, China, it selects the Spatial Zero-inflated Poisson model to examine the relationship between incubators and new firms’ existence and counts, after controlling for other spatial factors and indirect spatial effects. It further explores the heterogeneity of incubator effects between types and levels. Overall, the empirical results provide supportive evidence that the total number of incubators in a grid cell will encourage the birth and aggregation of new high-tech firms at a very significant level. As for the heterogeneity of incubators, this study finds that both Technology Business Incubator and Mass Maker Space significantly incent the birth of new firms but only Mass Maker Spaces have positive influence on the aggregation of new high-tech firms. Further breaking down to different levels, evidence shows that the lower-level MMS such as provincial and city-level seems to perform better in attracting new high-tech firms. The conclusion casts doubt on the resource allocation and policies for Technology Business Incubators, which has long been at the core of entrepreneurship and creativity supportive policies. Those government-backed and higher-level incubators do not function as well as more diverse and private-owned mass maker spaces in encouraging new high-tech firm entry and increase. In terms of the indirect effect from neighboring areas, it has been found that most of the incubator variables has positive indirect effect on new firm birth but negative effect on new firm aggregation (counts). The negative effect shows the neighboring areas are in competition for new firms instead of synergy. This also confirms that incubators are a force that shape the entrepreneurship distribution within the city. As Chinese entrepreneurship and corresponding incubation policies are still experiencing 59 rapid expansion, the study conclusion provides implications for policy makers and urban planners in the following three ways. (1) Incubators do benefit the agglomeration of high-tech firms with the city, but government should carefully revalue the resource allocation and regional economic impact of incubators, especially for large, state-owned and non-profit technology business incubators. More emphasis should be given to the efficiency and achievement instead of expansion in capability. (2) More research should be done on market- oriented Mass Maker Spaces and to figure out the underlying mechanism that they perform best in attracting new high-tech, for example, whether the global linkage, flexible policies or market operation makes them more effective. This could give direction for the future policy making. (3) As evidence shows competition between neighboring grid cells, another question arises that whether incubators do promote new firm birth or only attract firms from one grid to another. Urban planners are supposed to optimize industrial distribution to balance the specialization and urbanization effect and to avoid potential cannibalization between regions. 7.2 Limitation and Future Direction This study still has some limitations and is expected to be extended in several directions in the future. First, for the study object, it only considers the high-tech industry new firm entry behavior in the city of Shenzhen and the model only focuses on the last year. The city of Shenzhen is representative as stated but it is also useful to have more comparative studies on other types of cities and industries to better test the effect of incubators. Moreover, incubators not only influence the firm birth or entry activities but also affect their development, relocation and exit behavior. This study only examines the former due to the limitation of data. But as other startup activities are also important, they should be left for further studies. Also because of the lack of some detailed data in previous years, a cross-sectional analysis instead of a penal data analysis is adopted, while the latter may provide more evidence in the causal effect. Second, there are some other good methods for spatial analysis that are worth trying. (1) This study uses square grid cells as units for analysis, while the hexagonal cells are proposed as an alternative way that has a lot of advantages. For example, it can avoid the complication of the effect from diagonal and orthogonal neighbors and has greater clarity when used for visualization. (2) For the spatial weight matrix, the Queen’s method is used with the underlying hypothesis that every adjacent grid has the same effect. However, this could be 60 optimized by trying different method such as distance or nearest neighbor to test the heterogeneity of spatial indirect effects. Third, this study is more concentrated on the quantitative evidence of relationship between incubators and firm aggregation. It will be worthwhile to further examine the spatial effect channels and the reason underlying the attractiveness of incubators in qualitative or quantitative ways. By doing more detailed field study, the question why TBIs perform far from than expectation in attracting new high-tech firms recently could be better answered. 61 REFERENCES Adler, P., Florida, R., King, K., & Mellander, C. (2019). The city and high-tech startups: The spatial organization of Schumpeterian entrepreneurship. Cities, 87, 121-130. Aerts, K., Matthyssens, P., & Vandenbempt, K. (2007). Critical role and screening practices of European business incubators. Technovation, 27(5), 254-267. Amos Jr, O. M. (1990). Growth pole cycles: a synthesis of growth pole and long wave theories. Review of Regional Studies, 20(1), 37-48. Andersson, M., Larsson, J. P., & Wernberg, J. (2019). The economic microgeography of diversity and specialization externalities–firm-level evidence from Swedish cities. Research Policy, 48(6), 1385-1398. Arauzo Carod, J. M. (2005). Determinants of industrial location: An application for Catalan municipalities. Papers in Regional Science, 84(1), 105-120. Artz, G. M., Kim, Y., & Orazem, P. F. (2016). Does agglomeration matter everywhere?: new firm location decisions in rural and urban markets. Journal of Regional Science, 56(1), 72-95. Audretsch, D. B., & Feldman, M. P. (1996). R&D spillovers and the geography of innovation and production. The American economic review, 86(3), 630-640. Audretsch, D. B., Keilbach, M. C., & Lehmann, E. E. (2006). Entrepreneurship and economic growth. Oxford University Press. Avnimelech, G., Schwartz, D., & Bar-El, R. (2007). Entrepreneurial high-tech cluster development: Israel's experience with venture capital and technological incubators. European planning studies, 15(9), 1181-1198. Bandera, C., & Thomas, E. (2018). The role of innovation ecosystems and social capital in startup survival. IEEE Transactions on Engineering Management, 66(4), 542-551. Batty, M., 2013. The New Science of Cities. MIT Press. 62 Bergek, A., & Norrman, C. (2008). Incubator best practice: A framework. Technovation, 28(1- 2), 20-28. Boschma, R., & Frenken, K. (2011). The emerging empirics of evolutionary economic geography. Journal of economic geography, 11(2), 295-307. Boudeville, J. R., & Montefiore, C. G. (1966). Problems of regional economic planning. Edinburgh: Edinburgh UP. Braunerhjelm, P., Acs, Z. J., Audretsch, D. B., & Carlsson, B. (2010). The missing link: knowledge diffusion and entrepreneurship in endogenous growth. Small Business Economics, 34(2), 105-125. Brülhart, M., Jametti, M., & Schmidheiny, K. (2012). Do agglomeration economies reduce the sensitivity of firm location to tax differentials?. The Economic Journal, 122(563), 1069-1093. Bruneel, J., Ratinho, T., Clarysse, B., & Groen, A. (2012). The Evolution of Business Incubators: Comparing demand and supply of business incubation services across different incubator generations. Technovation, 32(2), 110-121. Carlton DW. (1979). Why new firms locate where they do: An econometric model. Joint Center for Urban Studies of MIT and Harvard University Carlton, D. W. (1983). The location and employment choices of new firms: An econometric model with discrete and continuous endogenous variables. The Review of Economics and Statistics 65: 440–449 Chan, K. F., & Lau, T. (2005). Assessing technology incubator programs in the science park: The good, the bad and the ugly. Technovation, 25, 1215–1228. Chandra, A., & Fealey, T. (2009). Business incubation in the United States, China and Brazil: A comparison of role of government, incubator funding and financial services. International Journal of Entrepreneurship, 13, 67. Chaşovschi, C., Bordeianu, O., & Clipa, D. (2014). Entrepreneurial culture in transition economies. The case of Romania and Republic of Moldova. Procedia Economics and Finance, 15, 1507-1514. 63 Chesbrough, H., Morten, T., Hansen, N. N., & Sull, D. N. (2000). Networked incubators: Hothouses of the new economy. Harvard Business Review, 78(5), 74-84. China’s National Bureau of Statistics. (2019). The Fourth National Economic Census Series Report (12). The Official Website of Chinese National Bureau of Statistics http://www.stats.gov.cn/tjsj/zxfb/201912/t20191218_1718313.html Cohen, J. P., & Paul, C. J. M. (2005). Agglomeration economies and industry location decisions: the impacts of spatial and industrial spillovers. Regional Science and Urban Economics, 35(3), 215-237. Cohen, W. M. (2010). Fifty years of empirical studies of innovative activity and performance. Handbook of the Economics of Innovation, 1, 129-213. Colombo, M., & Delmastro, M. (2002). How effective are technology incubators? Evidence from Italy. Research Policy, 31, 1103–1122. Darwent, D. F. (1969). Growth poles and growth centers in regional planning—a review. Environment and Planning A, 1(1), 5-31. Deloitte. (2020). Shenzhen IT/BT Industry Development Environment Observation Report. https://www2.deloitte.com/content/dam/Deloitte/cn/Documents/finance/deloitte-cn-fa- shenzhen-itbt-zh-201204.pdf (In Chinese) Ding, Y., Zhu, S., & He, Q. (2017). The Regional Differences and Determinants of Entrepreneurship in Guangdong Province: Based on Spatial Autocorrelation Analysis. Economic Geography, 37(11). (In Chinese) Du, R., & Zheng, S. (2020). Agglomeration, housing affordability, and new firm formation: The role of subway network. Journal of Housing Economics, 48, 101668. Duan, L., Du, D., & Huang, X. (2019). Spatial and temporal changes and influencing factors of the location of internet start-ups in Shanghai, China. Progress in Geography, 38(3), 383- 394. (In Chinese) Feldman, M. P., & Kogler, D. F. (2010). Stylized facts in the geography of innovation. Handbook of the Economics of Innovation, 1, 381-410. 64 Fischer, B. B., Queiroz, S., & Vonortas, N. S. (2018). On the location of knowledge-intensive entrepreneurship in developing countries: lessons from São Paulo, Brazil. Entrepreneurship & Regional Development, 30(5-6), 612-638. Fritsch, M. (2002). Measuring the Quality of Regional Innovation Systems: A Knowledge Production Function Approach. International Regional Science Review 25 (1): 86–101. Fu, W., & Deng, J. (2017). Locational Choice of Start-ups and Spatial Evolution of Clusters in the Context of Industrial Restructuring. Scientia Geographica Sinica, 37(6): 833-840. (In Chinese) Gladstone, R. (2015). Shenzhen: The City Where China’s Transformation Began. The New York Times. https://www.nytimes.com/2015/12/22/world/asia/shenzhen-site-of-landslide- embodies-chinas-rapid-growth.html Glaeser EL, Kallal HD, Scheinkman JA, & Shleifer A. (1992). Growth in cities. Journal of Political Economy 100: 1126–1152 Glaser, S. (2017). A review of spatial econometric models for count data. Hohenheim Discussion Papers in Business, Economics and Social Sciences. 19-2017. University of Hohenheim, Faculty of Business, Economics and Social Sciences. Granovetter, M. S. (1985). Economic action and social structure: A theory of embeddedness. American Journal of Sociology, 91, 481–510. Guo, J. (2020). Temporal-spatial distribution patterns and location choice of new firms: Taking electronic information manufacturing as an example. [Master’s Thesis, Henan University]. (In Chinese) Hackett, S. M., & Dilts, D. M. (2004). A systematic review of business incubation research. The Journal of Technology Transfer, 29(1), 55-82. Haltiwanger, J., Jarmin, R. S., & Miranda, J. (2013). Who creates jobs? Small versus large versus young. Review of Economics and Statistics, 95(2), 347-361. He, C., & Li, W. (2020). Evolutionary Economic Geography and Regional Development. Regional Economic Review, 2020 (1). (In Chinese) 65 He, Y., & Hu, P. (2016). Research on the Development of Hi-tech Incubators in Guangdong Province, Science and Technology Management Research, 2016 (2). Hochberg, Y. V. (2016). Accelerating entrepreneurs and ecosystems: The seed accelerator model. Innovation Policy and the Economy, 16(1), 25-51. Huang, X. (2018). Research on the Space-time Evolution and Mechanism of Internet Entrepreneurship and Financing Network in China. [Master’s Thesis, East China Normal University]. (In Chinese) Huang, Z., He, C., & Zhu, S. (2017). Do China’s economic development zones improve land use efficiency? The effects of selection, factor accumulation and agglomeration. Landscape and urban Planning, 162, 145-156. ITjuzi database. (2021). Company Data Searching. https://www.itjuzi.com/company Jacobs, J. (1969). The economy of cities. New York: Random House. Jacobs, J. (1984). Cities and the wealth of nations. New York: Random House. Ke, S., & Feser, E. (2010). Count on the growth pole strategy for regional economic growth? Spread–backwash effects in Greater Central China. Regional Studies, 44(9), 1131-1147. Kerr, W. R., & Kominers, S. D. (2015). Agglomerative forces and cluster shapes. Review of Economics and Statistics, 97(4), 877–899. Lavesson, N. (2018). How does distance to urban centres influence necessity and opportunity‐ based firm start‐ups?. Papers in Regional Science, 97(4), 1279-1303. LeSage, J. P. (2008). An introduction to spatial econometrics. Revue d'économie industrielle, (123), 19-44. Lewis, D. A. (2001). Does technology incubation work?: A critical review. Washington, DC: Economic Development Administration, US Department of Commerce. Li, Y., & Zhu, K. (2017). Spatial dependence and heterogeneity in the location processes of new high-tech firms in Nanjing, China. Papers in Regional Science, 96(3), 519-535. 66 Lin, D., Wood, L. C., & Lu, Q. (2012). Improving business incubator service performance in China: the role of networking resources and capabilities. The service industries journal, 32(13), 2091-2114. Lukeš, M., Longo, M. C., & Zouhar, J. (2019). Do business incubators really enhance entrepreneurial growth? Evidence from a large sample of innovative Italian start- ups. Technovation, 82, 25-34. Lutz, E., Bende, M., Achleitner, A., & Kaserer, C. (2013). Importance of spatial proximity between venture capital investors and investees in Germany. Journal of business research ,66(11). Marshall, A. (1890). Principles of economics (8th ed.). London: Macmillan Mejia-Dorantes, L., Paez, A., & Vassallo, J. M. (2012). Transportation infrastructure impacts on firm location: the effect of a new metro line in the suburbs of Madrid. Journal of Transport Geography, 22, 236-250. Mian, S., Lamine, W., & Fayolle, A. (2016). Technology business incubation: an overview of the state of knowledge. Technovation 50–51, 1–12. Moeller, K. (2018). Culturally clustered or in the cloud? How amenities drive firm location decision in Berlin. Journal of Regional Science, 58(4), 728-758. Mowery, D. C., & Rosenberg, N. (1991). Technology and the pursuit of economic growth. Cambridge University Press. Organization for Economic Co-Operation and Development (OECD). (1997). Technology incubators: Nurturing small firms. Paris: OECD. Organization for Economic Cooperation and Development (OECD). (1999). Business Incubation: International Case Studies. Paris: OECD. Osman, T., Thomas, T., Mondschein, A., & Taylor, B. D. (2019). Does traffic congestion influence the location of new business establishments? An analysis of the San Francisco Bay Area. Urban Studies, 56(5), 1026-1041. 67 Peters, L., Rice, M., & Sundararajan, M. (2004). The role of incubators in the entrepreneurial process. The Journal of Technology Transfer, 29(1), 83-91. Phan, P. H., Siegel, D. S., & Wright, M. (2005). Science parks and incubators: observations, synthesis and future research. Journal of business venturing, 20(2), 165-182. Rice, M. P. (2002). Co-production of business assistance in business incubators: an explorative study. Journal of Business Venturing, 17, 163–187. Rosenthal, S. S., & Strange, W. C. (2004). Evidence on the nature and sources of agglomeration economies. In J. V. Henderson, & J.-F. Thisse (Vol. Eds.), Handbook of regional and urban economics. Vol. 4. Handbook of regional and urban economics. 2119– 2171. Rubin, T. H., Aas, T. H., & Stead, A. (2015). Knowledge flow in technological business incubators: evidence from Australia and Israel. Technovation, 41, 11-24. Salman, A. K., Zampatti, D., & Shukur, G. (2013). Macroeconomic Determinants, Innovation and the Birth of New Firms: Negative Binomial Regression Approach: Negative Binomial Regression Approach. International Journal of Economics and Finance, 5(11), 72-81. Schumpeter, J. A. (1934a). Capitalism, socialism, and democracy. London: Allen and Unwin. Schumpeter, J. A. (1934b). The theory of economic development: An inquiry into profits, capital, credit, interest, and the business cycle. Cambridge, MA: Harvard University Press. Schumpeter, J. A. (1954). History of economic analysis. New York: Psychology Press. Schwartz, M. (2013). A control group study of incubators’ impact to promote firm survival. The Journal of Technology Transfer, 38(3), 302-331. Shi,j & He, C. (2014). Research progress in spatial dynamics of firms. Process in Geography. 33(10), 1342-1353. (In Chinese) Shepherd, D. A., Douglas, E. J., & Shanley, M. (2000). New venture survival: Ignorance, external shocks, and risk reduction strategies. Journal of Business Venturing, 15(5-6), 393- 410. 68 Smilor, R., & Gill, M. (1986). The New Business Incubator: Linking Talent, Technology, Capital and Know-how. Lexington Books, Lexington, MA, p. 224. Torch High Technology Industry Development Center & Ministry of Science and Technology. (2020). China Torch Statistical Yearbook-2020. China Statistics Press. United Nations Human Settlements Programme. (2019). The Story of Shenzhen: Its Economic, Social and Environmental Transformation. UN-Habitat. https://unhabitat.org/the-story-of- shenzhen-its-economic-social-and-environmental-transformation Van Praag, M., & Versloot, P. H. (2008). The Economic Benefits and Costs of Entrepreneurship: A Review of the Research. Foundations and Trends in Entrepreneurship, 4(2), 65-154. Vásquez-Urriago, Á. R., Barge-Gil, A., & Rico, A. M. (2016). Science and technology parks and cooperation for innovation: Empirical evidence from Spain. Research Policy, 45(1), 137- 147. Wang, D., Fang, B., & Chen, Z. (2018). Spatial pattern and evolution of internet companies based on community scale ---- A case study of Yangzhou. Economic Geography. 38(6). (In Chinese) Wang, Y. (2010). Research on the Technology Business Incubators Development Mechanism from the Views of Industrial Cluster. [Doctoral dissertation, Chongqing University]. (In Chinese) Wang, Y. (2019). Study on Agglomeration and Evolution Characteristics of Start-ups in Shanghai from the Perspective of Multi-source Big Data. [Master’s Thesis, Shanghai Normal University]. (In Chinese) White, Doug & Polly. (2017). No One Can Match Silicon Valley, But Other Cities Benefit From Fostering Startups. Entrepreneur. https://www.entrepreneur.com/article/296808 Woodward, D., Figueiredo, O., & Guimaraães, P. (2006). Beyond the Silicon Valley: University R&D and high-technology location. Journal of Urban Economics. 60: 15–32 69 Wu, M. (2020). Location Research on the Corporate Venture Capital and Entrepreneurship of Internet. [Master’s Thesis, East China Normal University]. (In Chinese) Xie, M., Zhao, H., Zhu, N., & Zhao, Y. (2017). Spatial Pattern Evolution and Location Selection of Software Industry in Ningbo. Economic Geography. 37(4). (In Chinese) Yao, L., & Hu, Y. (2020). The impact of urban transit on nearby startup firms: Evidence from Hangzhou, China. Habitat International, 99, 102155. Yang, Y. (2020). Research on Operating Efficiency and Business Model of Business Incubator. [Doctoral dissertation, Jilin University]. (In Chinese) Zhang, Y. & Li, C. (2014). Analysis on Spatial Agglomeration and Its Influencing Factors of China’s Venture Capital, China Soft Science Magazine, 2014(12). (In Chinese) Zhu, Y. (2014). Technology Business Incubators’ Efficiency and Their Influences on the Regional Innovative Economic Development in China. [Doctoral dissertation, University of Science and Technology in China]. (In Chinese) 70