What made Indian Cities and Towns Grow in the 2000 s? Stylized Facts and Determinants

Transcription

1 What made Indian Cities and Towns Grow in the 2000 s? Stylized Facts and Determinants Rana Hasan Asian Development Bank Ji Yiang Asian Development Bank Debolina Kundu National Institute of Urban Affairs India Policy Forum July 11 12, 2017 NCAER is celebrating its 60 th Anniversary in NCAER National Council of Applied Economic Research 11 IP Estate, New Delhi Tel: , NCAER Quality. Relevance. Impact

2 The findings, interpretations, and conclusions expressed are those of the authors and do not necessarily reflect the views of the Governing Body or Management of NCAER.

3 What made Indian Cities and Towns Grow in the 2000 s? Stylized Facts and Determinants* Rana Hasan Asian Development Bank Ji Yiang Asian Development Bank Debolina Kundu National Institute of Urban Affairs India Policy Forum July 11 12, 2017 Abstract We take a first step at trying to understand the determinants of city size and growth in India by combining information on city and town characteristics from three main sources of data: the 2001 and 2011 population censuses; the 1998 economic census; and GIS-enabled road network data. We use the data assembled to understand whether and how industrial structure, transport infrastructure, and various socio-economic characteristics (including those related to human capital) influence city size and growth. We find that various dimensions of economic activity and better connectivity to other locations are positively associated with the population growth of cities. To the extent that faster growing cities are also ones that provide higher real incomes to their residents, policymakers aiming to improve development outcomes and maximize the benefits of the urbanization process underway in India should focus more attention on what features of urbanization are conducive for the growth of economic activity and design policy interventions accordingly. JEL Classification: R11, O14, O18 Keywords: City growth, Migration, Human capital, Market access, Economic activity, Economic census * Preliminary draft. Please do not circulate beyond the discussion at NCAER India Policy Forum 2017, for which this paper has been prepared. rhasan@adb.org yijiang@adb.org dkundu@niua.org

4

5 Rana Hasan, Ji Yiang and Debolina Kundu 1 What made Indian Cities and Towns Grow in the 2000 s? Stylized Facts and Determinants 1. Introduction Rana Hasan, Ji Yiang and Debolina Kundu 1 Cities are widely believed to be engines of economic growth and good jobs. In this context, the urbanization process under way in India is good news. According to data from the Census of India, the share of India s population residing in urban areas increased from 20% in 1971 to 31% in 2011, with the urban population of 377 million representing the 2nd largest urban community in the world. All expectations are that the process of urbanization in India will continue, if not accelerate, with various models suggesting that the country might add another 400 million people by 2050 (United Nations, 2014). However, the link between urbanization and economic dynamism is not assured and its strength is likely to vary on account of various factors. In India s case, for example, several commentators have noted that the potential of its cities and towns to spur economic activity may not be met fully due to several factors, including limited investment in infrastructure and the prevalence of unsynchronized spatial and economic planning systems. Ahluwalia, Kanbur, and Mohanty (2014) note that in comparison to other fast-growing economies in Asia, urbanization in India has been relatively slow, largely unplanned, and characterized by underinvestment in urban infrastructure and public service delivery. Driven by such concerns, and the emergence of data disaggregated at district and city levels, researchers have recently turned their attention to understanding both the nature of urbanization underway in India, as well as the relationship between urbanization and various outcomes. The relationships studied include the size and determinants of agglomeration economies associated with urban locations (Chauvin et al. 2016; Desmet et al 2013; and Hasan, Jiang and Rafols, 2017), the relationship between transport infrastructure and city/urban growth (Lall, Wang, and Deichmann 2010 and Alder, Roberts, and Tewari 2017), and that between the form and shape of cities and economic growth (Harari 2016 and Tewari et al 2017). Some studies have also focused on documenting the spatial development of industry (manufacturing and services) and understanding its determinants (Ghani, Kerr, and Tewari 2014). With some notable exceptions, such as the recent work of Harari (2016) and Tewari et al (2017), most of these studies treat districts, the country s second administrative level (after states), as the urban unit of analysis (see, for example, Chauvin et al. 2016; and Ghani et al. 2013). This is because most available data from labor and enterprise surveys contain geo-information down to the district level. However, Indian districts often cover large rural areas and many contain multiple, 1 We thank Arvind Pandey and Pris Villanueva for their superb technical assistance on the analysis of population census data. Thanks are also due to Michelle Rafols and Rhea Molato for their dedicated work on economic census data and NSS employment-unemployment survey data, respectively.

6 2 India Policy Forum 2017 geographically independent urban areas, i.e., cities and towns. A more appropriate unit of analysis is thus likely to be the city or town, which is the administrative division below the district level whose rural counterparts are villages. Additionally, for those studies that look at the relationship between urbanization and economic activity, city level measures of the latter are captured in a highly aggregated manner (for example, through the intensity of nightlights as a proxy measure for aggregate economic activity as in Tewari et al). Nuances that arise from different structures of economic activity, for example, differences in the relative importance of manufacturing, the prevalence of formal and large firms, get missed in this approach. In this paper, we take a first step at trying to understanding the determinants of city size and growth in India by combining information on city and town characteristics from three main sources of data: the 2001 and 2011 population censuses, which provide demographic, amenities, and infrastructure related information; the 1998 economic census, which provides information on the structure of economic activity; and GIS-enabled road network data, which is used to construct transport connectivity and market access measures. Accordingly, we use the city level data we have assembled to understand whether and how industrial structure, transport infrastructure, and various socio-economic characteristics (including those related to human capital) influence city size and growth. We find a positive association between various dimensions of economic activity and population growth of cities, emphasizing the importance of cities as centers of production. We also find connectivity between cities to influence city growth. While the causal interpretation of our results is tentative at this point of time, we believe that the analysis contained here is useful in building a set of stylized facts about Indian urbanization. The remainder of this paper is organized as follows. Section 2 motivates the need for understanding the drivers of urban growth from a policy perspective and provides key features of the urbanization process underway in India, including major policy initiatives undertaken for improving urban outcomes. Section 3 provides a discussion on the empirical framework used in this paper to analyze urban growth and the data used in the analysis. Sections 4 and 5 describe the results of our analysis on the determinants of city size and city growth, respectively. Section 6 concludes. 2. Urban Growth Cities need many things to support a growing population and thrive economically. They need to provide their residents affordable housing; access to infrastructure encompassing transport services, power supply, and water supply and sanitation; and institutions for building and sustaining human capital. They also need to provide an ecosystem that encourages economic activities of various types. This, in turn, requires managing land use and a synchronized approach to economic and spatial planning, among others. Since no city is an island existing on its own, connectivity to other cities/locations is also important. Providing all the necessary infrastructure and institutions is not easy. Infrastructure can be quite expensive, and good and effective planning requires considerable coordination across different agencies as well as technical expertise that may be scarce. Where should policymakers interested in making the most out of urbanization focus? For example, do cities of a particular size class say, small and

7 Rana Hasan, Ji Yiang and Debolina Kundu 3 medium size cities versus large ones have the most potential to grow? At another level, cities have traditionally been seen as centers of production. Does the nature of economic activity in a city have implications for its future growth? For example, do cities with a more diversified structure of production have greater potential to grow? Are cities that provide a conducive environment for new establishments to set up business economic firms likely to grow faster? 2 Having a good understanding of the drivers of city growth can be particularly useful for policymakers in developing countries where the urbanization process has yet to run its course. This is certainly the case in India Urbanization in India: Trends and Patterns Contrary to popular perceptions of an urban explosion, urban India has experienced a deceleration in the growth of its population during the last three decades. The urban growth rate came down from 3.1% in the 1980s to 2.73% in the 1990s, and increased marginally over to 2.76%. The consistent decline in the growth rate of urban population over the past decades led Government of India s plan documents to express concern over the moderate pace of urbanization. Nevertheless, the percentage of population in urban areas has continued to increase, up from 23.3% in 1981 to 27.78% in 2001 and 31.16% in It is important to analyze the reasons for this growth; i.e., whether it has been driven by: (a) natural increase in the population of existing cities, (b) migration into cities, (c) emergence of new urban centers, and/or (d) expansion in municipal limits and urban agglomerations. The increase in urban growth in the past decade cannot be attributed to a spurt in natural growth in population as the latter has declined more or less uniformly both in urban and rural areas during the last couple of decades. The annual growth rate in population in the present decade is 1.62% registering a decline from 1.95% in the preceding decade (Kundu, 2011). Migration due to economic compulsions seems to have reduced among ruralurban migrants as suggested from the 55th and 64th rounds of the National Sample Survey as well as the Provisional Tables from Census Also, the share of adult male migrants in the corresponding adult male population in urban areas has declined slightly from 32% in to 31% in suggesting that migration to existing urban centers may not have been a major factor in the growth of the urban population. Urbanization seems to have been driven by the last two components of growth noted above, manifested especially by an increase in the number of urban agglomerations (UAs) in 2011 and somewhat helped by an expansion of new census towns. 3 Notably, 91 new urban agglomerations, comprising a continuous urban spread 2 Indeed, given the potential importance of agglomeration economies associated with the urbanization process, rather than try to spread economic development everywhere and equally, it might be better to focus on certain types of locations and types of cities. 3 In addition to statutory towns towns notified under law by the concerned state/union territory governments and having local bodies like municipal corporations, municipalities, etc., regardless of their demographic characteristics urban centers include statutory towns, census towns and urban agglomerations. Census towns are administrative units formally classified as villages but which satisfying the following three criteria simultaneously: (i) a

8 4 India Policy Forum 2017 over adjoining towns, census towns and their outgrowths, have come up due to an expansion of city limits reflecting a "passive" type of migration. The new urban agglomerations are especially prevalent in the states of West Bengal, Kerala and Tamil Nadu and an analysis of disaggregated data of the new urban agglomerations, many of which became million plus in 2011, are attributed to the growth and merger of new census towns in close proximity to a statutory town. Larger cities and UAs have played an especially important role in India s urbanization process. This may be seen by examining Census of India data that classifies urban centers into six classes. A center with a population of more than 100,000 is called a city while that with less than 100,000 is called a town. Cities with population between 1-5 million are called metropolitan cities while those with more than five million are mega cities. Table 1a shows that Class I cities and UAs (i.e., urban centers with population of 100,000 or more) accounted for approximately 70% of the urban population in both 2001 and These cities/uas accounted for 71% of total urban growth between 2001 and Among these, the size class of one million plus, especially cities with 1-5 million residents, recorded the highest growth. This is because many new million plus UAs came up in this category in the last decade by the merger of statutory and census towns. The share of incremental population in class-i cities comes down from over 70 per cent to 56 per cent when cities are considered individually and not part of UAs. In this analysis, one finds a more equitable distribution of population across size classes. The highest growth rate of 7 per cent is registered by cities reporting a population of more than 5 million. This may be attributed to a substantial increase in the boundaries of cities like Bangalore and Hyderabad, which had an impact on their population as well Policy Perspective on Urban Development The nineties saw an era of opening up of the country s economy. Following the balance of payment crisis, a program of economic liberalization was launched. Roughly, during the same time the 74 th Constitutional Amendment Act was enacted which decentralized powers and essential functions related to city planning, local economic development, poverty alleviation and provision of basic services to the urban local bodies 4 (ULBs). However, even after several decades, many states are yet to transfer the mandated funds, functions and functionaries to ULBs, thereby increasing the dependency of the ULBs on higher authorities (Vaidya and Vaidya, 2010). The Approach paper to the Tenth Plan admitted that the past approach to the process of urbanization had been largely reactive and called for a change for urban planning to become more anticipatory with an integrated approach to addressing various dimensions of urban development. Accordingly, the Jawaharlal Nehru National Urban Renewable Mission (JNNURM) was launched in 2005 to create economically population of at least 5,000; (ii) 75% or more of the male population (working population in 1961 and 1971) engaged in nonagricultural pursuits; and (iii) a population density of at least 400 persons per square kilometer. Urban agglomerations are continuous urban areas consisting of a statutory town and its adjoining outgrowths or two or more physically contiguous statutory towns (with or without the outgrowths of such towns). For more details, see Section 3. 4 Urban local bodies are statutory towns/cities.

9 Rana Hasan, Ji Yiang and Debolina Kundu 5 productive, efficient, equitable and responsive cities. It was the single largest initiative of the government for planned urban development that tried to integrate the two pressing needs of urban India: massive investments required for infrastructure development and reforms that were required to sustain investments. JNNURM attempted infrastructural development and reform in governance in 65 cities through 70 percent of the total stipulated funds. The remaining was expected to provide the poor in these cities access to basic services and housing through a Basic Services for the Urban Poor component. Never before had the selected large cities under any urban development program received per capita allocation on such a large scale for infrastructural investment coming as project grants through additional central assistance. The Approach Paper to the Eleventh Plan reaffirmed the concern regarding concentration of demographic and economic growth in a few large cities and deteriorating infrastructural situation in these that nonetheless provide large economies of agglomeration. The Eleventh Plan envisaged the government strategy to establish the macroeconomic preconditions for rapid growth and support key drivers of this growth. It further added that the strategy must also include sector-specific policies to ensure that the structure of growth and the institutional environment in which it occurs, achieves the objective of inclusiveness in all its many dimensions. The Approach Paper to the Twelfth Plan added that since it takes time to create urban infrastructure, it is necessary to have a sufficiently long term focus on urban planning with thrust on accelerated investment in infrastructure, as this is critical for sustaining and accelerating growth. During the Eleventh Plan, in pursuance with the vision to make India slum-free, and meet the housing deficit of million, the Rajiv Awas Yojana was launched in 2011 which was replaced by Housing for All in Also, during the 12 th Plan, the Swachh Bharat Mission and the National Urban Livelihood Mission were launched to cover all the 4041 statutory towns to provide sanitation and livelihood respectively. The AMRUT and Smart Cities Mission were also launched in 2015 to provide basic services and build amenities in select Class I cities. Only a few small towns are covered under these Missions. Further, these missions have a much lower per capita allocation as compared to JNNURM. The latter succeeded in getting the state and the city governments to commit themselves to structural reforms which the central government failed to achieve despite adopting several measures and incentive schemes through other programs and legislations (Kundu et. al. 2007). It was also effective in renewing focus on the urban sector across the country. This was more evident in the developed states of the western and southern regions of the country. Yet, many states in the central and eastern regions lagged behind in program utilization due to lack of enabling capacity and capacity to generate matching funds. Also, the share of government fund allocated under JNNURM was largely biased towards the large cities (Kundu and Samanta, 2010). The impact of the reform measures and urban missions in urban areas on the growth process remain to be examined.

10 6 India Policy Forum Analyzing Urban Growth: Framework and Data Requirements 3.1. A Simple Framework for Analyzing Urban Growth In an accounting sense, an increase in the growth of the urban population can be seen to be the result of an increase in the population of existing cities i.e., the sum of births minus deaths plus net migration as rural residents move to urban areas and the reclassification of rural areas to urban areas. The latter process can involve either the creation of entirely new cities and towns or the expansion of existing cities and towns as rural hinterlands get absorbed into adjoining urban areas. There are several factors that influence how much growth a city experiences (through migration and/or expansion of a city s de-facto boundaries). First, there is the structure and nature of economic activity undertaken in a city. These can be expected to be a major driver of productivity and labor demand, and thus the wages and incomes a city offers. The latter, in turn, will drive how much migration a city experiences; it may also influence urban growth through an expansion of the physical size of a city. This expansion may be captured by an expanding administrative boundary of the city, or by urban outgrowth (i.e., the urbanization of areas outside the administrative boundaries of the city). Second, there are factors such as are skills and education available in a city, transportation and connectivity (to other cities and locations, but also within the city), amenities and the cost of living. A third factor is the spatial layout of cities, which as the recent work of Harari (2016) has shown, interacts with geography and regulatory factors (such as land use regulations) and can influence the location choices of consumers and firms. In this paper, we focus on the first two sets of factors. We examine their empirical importance by estimating regressions of the following type: (1) ln popct ln popct-1 = a*ln popct + Xct + ect where pop represents population, X is a vector of city characteristics, and c and t denote cities and year. As discussed in Duranton (2016b), this formulation, whereby city growth is regressed on the initial levels of the determinants of city growth (rather than growth of the determinants) is justified when labor mobility is imperfect, so that city populations adjust slowly. Our main challenge is that data on productivity, wages and incomes at the city level are hard to come by. Close proxies, however, are available through India s economic census which provides two key pieces of information on establishments in manufacturing and most services: the industry to which an establishment belongs and the number of employees it has. 5 Given the close relationship between establishment size and wages (and even industry and wages), measures that can proxy for productivity, wages and incomes at the city level can be generated. We now turn in detail to data issues. 5 The economic census carried out in 1998 also provides information on the age of enterprises.

11 Rana Hasan, Ji Yiang and Debolina Kundu Data To explore the determinants of city growth in India, our main sources of data include the Census of India for 2001 and 2011; establishment-level information from the fourth economic census (EC) carried out in 1998; and GIS-enabled road network data to measure city s connectivity and market access. Some of the analysis also utilizes the individual-level survey data from various rounds of the employment unemployment survey of the National Sample Survey Office. Before getting into details on these data and the variables we use and construct for our analysis, a few remarks on the population census from the perspective of analyzing urbanization issues are in order. First, at least since the 1961 census, the criteria for identifying urban areas have been broadly similar. In addition to statutory towns (with a municipal corporation, municipality, cantonment board, notified town area committee, or town council), urban areas include census towns, i.e., places that meet three criteria: (i) a population of at least 5,000; (ii) 75% or more of the male population (working population in 1961 and 1971) engaged in nonagricultural pursuits 6 ; and (iii) a population density of at least 400 persons per square kilometer. Census towns are administrative units formally classified as villages but which satisfying the three criteria above simultaneously. In addition to statutory and census towns, the census also tracks urban agglomerations, continuous urban areas consisting of a town and its adjoining outgrowths or two or more physically contiguous towns (with or without the outgrowths of such towns). 7 Urban agglomerations must consist of at least a statutory town, and its total population (all constituents combined) should not be less than 20,000. A common practice is to refer to cities as towns with population of more than 100,000. Both towns and urban agglomerations are treated equally in our analysis and referred to as cities. Second, while the concepts of census town, outgrowths, and urban agglomerations try to capture the urbanization process taking place outside the administrative boundaries that mark statutory towns, it is possible that official statistics understate the extent of urban growth. One reason is the complex and time-consuming process of redrawing municipal boundaries as cities and towns expand (Colmer, 2015). Thus, reliance on census data may understate the true extent and growth of urban areas and, therefore, urban population. An increasingly popular approach around this problem is to use data on nighttime lights and use these to delineate urban areas. The underlying logic is that urban areas are characterized by a concentration of people and incomes such that a threshold value of the luminosity of nighttime lights can be used as a proxy for whether a given area of land is urbanized or not (see, for example, the application of this approach by Tewari et al 2017). However, the use of nighttime lights in the analysis of urbanization issues is not without its own drawbacks. For example, from the perspective of defining urban areas, scattering of nightlights over wide 6 The definition for the 2011 census entailed a slight change. Now, only main male workers are included in the criteria; marginal workers are excluded. 7 An outgrowth is an area lying outside the boundary of the town and within the revenue limit of a village or group of villages. It should possess varies urban features in terms of infrastructure and amenities. Examples of outgrowth include railway colonies, university campuses, and port areas.

12 8 India Policy Forum 2017 geographic area could well lead some rural areas adjacent to cities to be defined as urban Population and Other Characteristics of Cities and Towns from Census Data The Census of India provides town directories which include detailed tabulated information on not only demographic variables, but also on geographic features (including land area), climatic amenities, infrastructure provision, social and educational services, and government revenues and expenditures at the city/town (and village) level. We limited our attention to towns with a population of at least 10,000 in 2001 and used town IDs to merge the town directories of the 2001 and 2011 censuses. For those with inconsistent IDs, we made use of the town names, district names and state names to verify and match. We treated each urban agglomeration (UA) as an integrated city, aggregating variables of the constituent towns and areas to reach the final values for the UA. By matching cities and towns across the 2001 and 2011 censuses, we are able to generate both our main dependent variable (growth in a city/town s population) and various town-level characteristics as explanatory variables, such as city amenities, infrastructure, and access to state headquarters, among others. We encountered towns/cities with extreme population and/or area growth between 2001 and 2011 in the data. After diligent validation of the data, we cut our samples only to those towns/cities with population growth greater than -50% and less than 500%; and area growth greater than -10% and less than 500%. Our final dataset allows us to work with about 2,427 Indian cities with a population of at least 10,000 in 2001 and belonging to 502 districts across 21 states and 4 union territories that are well integrated from the perspective of transport connectivity. 8 It is important to note that our sample does not capture the urban growth that results from the transformation of a rural area in 2001 into a standalone urban center such as a census town in We also do not capture towns in 2001 that get classified as a new urban agglomeration in This is because we have no way to split a new urban agglomeration into its 2001 constituents. Nevertheless, our sample should capture key features of urbanization s and its growth between 2001 and The Census of India reports that the majority (86.1%) of cities in India had a population of less than 100,000 in There were only 73 cities in India, or 2.6% of the total, at the other end of the spectrum with a population of 500,000 or more. In our final sample, 85.5% of cities had a population of less than 100,000 in 2001 and 2.8% had a population of 500,000 or more 2.8% Error! Reference source not found.b presents the distribution of our sample cities by population and also compares it with the overall All-India numbers reported in the census. As may be seen, our sample mimics key features of the urban population and its distribution well. Figure 1 presents a scatter plot of urban growth rates over by 2001 population. Error! Reference source not found. reports some summary statistics for our 8 The included states and union territories are: Andhra Pradesh, Assam, Bihar, Chandigarh, Chhattisgarh, Dadra & Nagar Haveli, Daman & Diu, Delhi, Goa, Gujarat, Haryana, Himachal Pradesh, Jharkhand, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Orissa, Pondicherry, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, Uttaranchal, West Bengal.

13 Rana Hasan, Ji Yiang and Debolina Kundu 9 sample and by different city size groups. The upper panel is based on information provided in the population census tables, and in addition to population also cover measures of climate, infrastructure provision, and educational services. Although not shown in the table, the diversity of these characteristics across Indian cities is noteworthy. For instance, the 5 th percentile literacy rate across cities is around 54% while the 95 th percentile literacy rate is almost 89%. The 5 th percentile annual maximum temperature is 30 degrees, considerably lower than the 95 th percentile value of 46 degrees. There are towns with only 0.32 kilometers per square kilometer of city area and 356 electricity connections per 10,000 people (both at the 5 th percentile values of their respective distributions) versus 11.5 kilometers and 3,269 connections (at the respective 95 th percentile marks). Figure 1. Sample Cities and Towns Log Population, 2001 Several studies show that the basic infrastructure including urban amenities is higher in metropolitan cities as compared to small and medium towns (Kundu et al., 1999; Kundu, 2000, Bhagat, 2011, and NIUA, 2017). Town level data from both Population Census and NSSO as well as sample data used in this paper also confirm this trend. The sample data suggests that on an average there were 1705 electricity connections per 10,000 population in The number of connections increased with size class, registering a decline in the million plus cities. This could be attributed to the lesser number of industrial connections in the metropolitan cities as the existing byelaws prohibit setting up of industries within city limits. Road (concrete) density which recorded an average value of kms of road per sq.km area per town also increased systematically with size class, the million plus reporting the highest road density (Table 2). Importantly, the million plus cities recorded about double the road density of towns below 50,000 population. The number of degree and professional colleges and polytechnic institutes declined in higher order cities (Table 2). This may be attributed to the setting up of several of these colleges in non-metros to impart skill and vocational training to young people. However, literacy shows a reverse pattern,

14 10 India Policy Forum 2017 improving systematically in bigger cities. Workforce participation rates, on the other hand, also reported a declining trend in larger cities (Table 2) Measures of Economic Activity Using the Economic Census To capture the nature of economic activity at the city level, we use the Economic Census (EC), which is conducted by the Central Statistics Office of the Ministry of Statistics and Programme Implementation. The EC is a countrywide census of establishments engaged in all economic activities except crop production and plantations. 9 The key purpose of the EC is to provide a sampling frame for follow-up sample surveys intended to collect more detailed sector-specific information on the nonagricultural economy. In this study, we employ the fourth edition of the EC carried out in The data allows for the geographic location of establishments to be identified at the town and city level. 10 The Economic Census of 1998 contains information on an establishment s number of employees, major economic activity, registration status with government authorities, type of ownership, and age of the enterprise (as captured by information on the number of years of operation of the enterprise). Overall, the dataset covers 12.6 million establishments for the urban India, and 17.7 million establishments for rural India. In order to include Economic Census variables into our analysis, we had to match administrative geographic boundaries across census years. Since the Economic Census of 1998 follows spatial boundaries as of August 1997, significant manual effort was spent matching city-level units to their 2001 (population) Census equivalent. Secondary sources such as the administrative atlas of India, pin code database, digital maps and other online resources were utilized to track geographical variabilities between both years. This effort resulted in a one-to-one mapping of urban areas in the Economic Census of 1998 with those of 2001 Census. In constructing the various measures of the nature of economic activity across cities, we limit our attention to 22 manufacturing and 22 service industries. This means we omit one manufacturing industry (recycling) and several service industries (public administration and defense; other services; activities of domestic households as employers of domestic staff; and extraterritorial organizations and bodies). Given that economic census data has rarely been used to examine the structure of economic activity (certainly at the city or town level), we compared sectoral totals and composition with round 55 of the Employment-Unemployment Survey (EUS) of the National Sample Survey Organisation carried out in for the geographical locations considered in the study. Though level differences in employment across the two sources of data are apparent, correlations of employment distribution across the 44 two-digit industries considered here are very strong (correlation of 0.94) across the NSS EUS and EC. 11 We also find the manufacturing to services sector employment ratios from EC and EUS to closely mirror each other. The only main difference between the EC and NSS EUS can be found in estimates of the share of total employment accounted by 9 We use the terms establishment, enterprise, and firms interchangeably. 10 A village is the rural counterpart to a town. 11 Since the EC uses the 1987 four-digit National Industrial Classification (NIC 1987) to record an establishment s economic activity, we first transformed the 4-digit 1987 industry codes to their 2-digit NIC 1998 equivalent.

15 Rana Hasan, Ji Yiang and Debolina Kundu 11 formal/large firms (i.e., firms with more than 10 workers). Overall, the Economic Census appears to be a good source on the sectoral distribution of firms and employment at the city-level. The middle panel of Table 2 presents the variables of interest from the Economic Census, including total employment in the 44 manufacturing and service industries we consider, sectoral employment share from large (10+) and very large (100+) firms, and share of young firms (i.e., enterprises reporting 5 years or less of operations). 12 We also constructed diversity and specialization indices to introduce measures pertaining to the composition of industries. These indices are defined as Div c = 1 j s jc s j, and Spec c = max j (s jc s j ), respectively, wherein s jc refers to the share of industry j in city c's total employment, and s j is industry j's share in national total employment. The values taken by these variables can vary widely across cities. Moreover, they also vary by city size. For examples, larger cities are more diversified in terms of industrial activity, they tend to have a larger share of manufacturing relative to employment in manufacturing and the restricted set of services we consider here, and a larger share of younger firms Road Connectivity and Market Access We obtain a road network vector data from ML Infomap, which contains geographically referenced state highways, national highways and expressways of India since late 1990s. With aid of GIS software, we compute the straight-line distance from a city s centroid to the nearest state highway and to the nearest national highway or expressway by 2001 as two measures of a city s road connectivity. For market access, we consider each city s access to the largest 74 cities with population above 500,000 in Specially, market access is calculated with the formula: 74 1 MAc POPi d i 1 ci where d ci is the distance from city c to one of the 74 large cities travelled through the available road network, and set to 1 if c i. 13 The MA variables is expected to capture the demand for outputs of a city from major markets within the country. The bottom panel of Table 2 shows that the average distance to state highway and national highway or expressway is 6.0 km and 12.5 km respectively. The smallest 12 Note that for a household establishment carrying out a business for generations, the years of operation refers to the number of years since the current owner or operator took over. This is not ideal for our purposes but unavoidable. 13 The road network data from ML Infomap have many disconnected or isolated roads, which may be because it does not include small local roads. This causes the GIS software which uses an algorithm that accounts for junctions to generate the shortest path to fail to calculate road network distance for many cities. As an alternative, we use the GRoads road data to resolve the issue. The concern is that GRoads only contains 2010 information without data on historical network. We compare the distance computed with ML Infomap and GRoads data, respectively, for cities that are possible with the former. The two sets of distance are highly correlated.

16 12 India Policy Forum 2017 cities (population below 50k) are farthest to the state highways. But the distance to the state highways does not decrease with city size monotonically. Cities with population between half to one million are also located far from the state highways on average. The distance to the national highways or expressway decreases monotonically with city size. Moreover, cities with population above half million are considerably closer to national highways or expressways than the rest cities. As far as market access is concerned, the sample average is 194,820. Means of different size groups deviate from the sample average moderately whereas larger cities do not necessarily have better access to the country s major markets. 4. City-size across India: Some Patterns Given the novelty of our data set, it is interesting to examine whether and how city size varies with economically relevant variables. In Table 3, we report regressions of city size on measures of human capital. Unfortunately, some of these measures such as the share of the population having at least a certain level or type of education are only available at the district level (and are labeled with the prefix D in the table). Others, such as the literacy rate, and the presence of educational institutions such as schools and colleges are available at the city level. The key takeaways from the regressions are as follows. First, larger cities tend to have more literate and educated populations. Second, they are also more likely to have a greater number of colleges though not necessarily as a proportion of city population. Thus, our measure of college density which is defined as number of Arts, Science, Commerce Colleges per 100,000 population, 2001 is negatively signed. Table 4 reports regressions of city size on various measures of city level infrastructure, it s including that related to market access. Surprisingly, larger cities are not systematically associated with either higher road density or electricity connections for reasons cited in Section 3.3. Larger cities, are, however, better connected to state and national highways (or expressways) as the negative and significant coefficient on distance to state and national highways/expressways shows. Interestingly, market access is not necessarily better for larger cities. The next set of tables (Tables 5 a-c) reports regressions of city size on various measures of economic activity in order to examine whether the nature of economic activity carried out across different types of cities varies by city size. In addition to regressions based on our full sample of cities, we also consider different subsamples based on city size. We find that larger cities tend to have higher diversity and specialization indexes. Interestingly, this result tends to be driven by the larger cities i.e., those with population of hundred thousand or more especially in the case of the diversity index. Larger cities also tend to have a higher share of employment coming from manufacturing, formal sources of employment (as proxied by firms with 10 or more employees), and large firms (as proxied by firms with hundred or more employees). As in the case of the diversity Index, much of this relationship is driven by the so-called Class 1 cities with population of a hundred thousand or more.

17 Rana Hasan, Ji Yiang and Debolina Kundu City Growth We now turn to examining the empirical importance of the different factors that may be driving urban growth by estimating various versions of equation (1) above. As noted in Section 3.1, this formulation, whereby city growth is regressed on the initial levels of the determinants of city growth (rather than growth of the determinants), is justified when labor mobility is imperfect, so that city populations adjust slowly. We, thus, first show that based on available data, migration from rural to urban areas or from one urban area to another is on the lower side in India though not trivial and that some amount of city growth also takes place through an expansion of the area encompassed by cities and towns (passive migration). We then move on to analysis of other drivers of city growth Migration and City Growth Data limitations prevent us from examining the relationship between city growth and demographic factors births, deaths, and migration at the city level. However, sufficient data does exist at the district level to construct proxies for the birth rate and migration and examine the relationship between these and city level growth. The reason it is important for us to try and examine this relationship, even with imperfect data, is that our assumption that the type of economic activity of the city has a bearing on its subsequent growth is based on the idea that city growth takes place not only due to natural population growth (i.e., the difference between births and deaths), but also migration. Table 6 presents summary statistics on three proxies for the birth/fertility rate based on population census data and various measures of migration from the NSS Round 55 survey carried out in that provided information on migration and a breakdown based on whether it was driven by within district migration, inter-district migration, or inter-state migration. Unfortunately, the data are only available at the district level and pertain to the urban sectors of the districts covered by our sample cities. 14 What they show is that urban growth is very much driven by migration within districts and within states. Table 7 uses these numbers in regressions of city growth. In all columns presented, initial city size as captured by the 2001 city population is included as a regressor. 15 Overall, the results indicate that both natural growth (as captured by our various proxies for birthrates/fertility rates) and migration contribute to city growth. At the same time, it must be noted that the R-squares of all the regressions are relatively low. This is likely a reflection of the fact that our explanatory variables are the district level while our dependent variable is at the city level. In addition, it also probably reflects the fact that we do not capture the passive migration that results from the physical expansions of cities beyond their original borders. As noted earlier, if we split 14 Strictly speaking, only numbers for the full sample should be considered when examining Table 7. This is because most districts contain a mix of cities belonging to different size groups. 15 The first column its does not include any of the demographic and migration related measures. The positive and statistically significant coefficient suggests a failure of Gibrat s law which holds that the population growth of cities is orthogonal to their initial level. Further exploration revealed that this result tends to be driven by the relationship between city population and initial population levels in cities with less than hundred thousand persons in the initial year.

18 14 India Policy Forum 2017 our sample of cities into those that experienced growth of area versus those that did not, the former account for 70% of the total increase in urban population as estimated by our sample. This increase in city area must surely have been associated with some nontrivial expansion in the population reclassified as urban from previous rural status Human Capital Research on the determinants of city growth in developed countries often finds an important role for human capital. Corroborative evidence from developing countries is much more scarce, but studies based on Colombian and Brazilian data indicate its relevance to the developing country context as well (see Duranton 2016a and de Mata et al 2007, respectively). Tables 8 a-c report our findings on the importance of human capital on city growth in India. As noted earlier, a number of our measures for human capital are less than ideal, given that they pertain to the district level (albeit its urban component) rather than the city level. Nevertheless, the overall picture we get is one of weak association between human capital and city growth. In fact, literacy rate appears with a negative and statistically significant sign. Interestingly, this relationship is driven by cities with smaller cities i.e., those with less than 100,000 people in Infrastructure and Connectivity Tables 9a-c describe the relationship between city growth and our various measures of infrastructure and connectivity. Regardless of which sample we consider, our measure of market access enters the various city growth regressions with a positive and statistically significant coefficient. Consistent with this, a smaller distance from the city to a highway, especially with a national highway or expressway, is associated with faster city growth in smaller cities. In contrast, a higher percentage of households with electricity connections and greater density of paved roads either has no relationship with subsequent city growth (the case in larger cities) or has a negative relationship with city growth (the case in the smaller cities) Characteristics of Economic Activity We finally turn to the relationship between city growth and the nature of economic activities in cities as described in Tables 10 a-c. Our key findings are as follows. First, a larger share of 2001 employment in manufacturing relative to the services we consider is associated with faster city growth subsequently (i.e., from ). Interestingly, this effect is stronger in cities with hundred thousand persons or more in our initial year, 2001, but present in the smaller cities as well. Second, in the smaller cities the presence of a formal manufacturing enterprise (as proxied by an enterprise with 10 or more employees) is associated with faster city growth. Third, a larger share of young firms i.e., of age 5 years or less is associated with faster city growth. Finally, the diversity and specialization indices, which tended to be larger in larger cities as shown in the results of the previous section, are not associated with faster city growth. However, when we omit initial population, diversity in manufacturing has a positive and significant effect on city growth in the smaller cities.

19 Rana Hasan, Ji Yiang and Debolina Kundu 15 A robustness check involving the consideration of the above relationships at the urban district level yield similar results to those of our full sample of cities (Table 11) A Policy Experiment: How New Entrepreneurship Can Boost City Growth The above results show that some aspects of economic activity, such as the share of firms which are young (5 years or less) or the share of employment in 10+ worker firms in the initial year, can be highly correlated with subsequent city growth. Moreover, the correlations vary greatly across city size class. For instance, the coefficient estimates for the share of young manufacturing is for small cities with population below 100,000 and for the large cities with population above 100,000. Similarly, a higher share of manufacturing employment in formal firms (i.e., firms with 10 or more workers) is significantly associated with faster city growth in cities with population of 100,000 or more but not smaller cities. It is interesting to consider what these estimated coefficients suggest for the impact of hypothetical policies that foster entrepreneurship and formalization on city growth on the assumption that our regression estimates can be given a causal interpretation. Table 12 illustrates two examples involving the share of young firms in manufacturing and both manufacturing and services firms, respectively. Suppose a policy to promote entrepreneurship leads to a 10 percentage point increase in the share of young firms in small cities. We can then calculate how many new young firms this increase translates into and how many more urban residents it would lead to across small cities. A related question of interest is what additional urbanization would be generated if we were to proportionally relocate the same number of young firms to the big cities. Take case of the manufacturing firms (left panel). A 10 percentage point increase in the share of young firms in 2001 means 62 additional young manufacturing firm on average and 129,329 in total for the small cities. Applying these numbers to the estimated regressions, the urban population of small cities would increase by 334,341 in total by With the same number of young firms distributed to the large cities, the average share of young manufacturing firms would increase from 38.9% to 47.7% in The predicted urban population by 2011 would increase by 28,508 per city on average and 10 million in total for the large cities. The results are remarkable in that the same number of young manufacturing firms could lead to an urban population that is 30 times higher if the increase were concentrated in large cities rather than in small cities. The gap is mainly driven by the different elasticities of urban population to the share of young firms across small and large cities. When the services sector is taken into account (right panel), the differential effects across small and big cities are moderated but remain significant (20 times). For policymakers interested in promoting urbanization by fostering entrepreneurship in cities, the simulation exercise suggests that focusing on large cities may be more cost effective (assuming similar costs of enterprise promotion policies across city size).