POLICY RESEARCH WORKING PAPER 3 073 Metropolitan Industrial Clusters Patterns and Processes Sanjoy Cbakravorty Jun Koo Somik V. Lall The World Bank Development Research Group Infrastructure and Environment June 2003 I POLICY RESEARCH WORKING PAPER 3073 Abstract Where do industries locate within a metropolitan area? textiles, leather, printing/publishing, chemicals, metals, Do different industrial sectors have different patterns of machinery, and electricaVelectronics) for evidence of location/clustering? Can these patterns be understood global and local clustering, and distinguish between and with reference to industry characteristics? What is the test for co-clustering and co-location of industries. The geographical relationship between clusters of different results suggest an evolutionary model of industry types of industry? To what extent do localization location in mixed rather than specialized industrial economies influence the clustering process? These districts. There is little evidence of localization questions are investigated with geographically economies from labor markets or buyer-supplier disaggregated industry location and size data from networks. The authors suggest that land use policy is the Mumbai, Kolkata, and Chennai. Chakravorty, Koo, and key variable influencing the intra-metropolitan spatial Lall analyze eight industrial sectors (food/beverages, distribution of industry. This paper-a product of Infrastructure and Environment, Development Research Group-is part of a larger effort in the group to examine the role of spatial externalities in the concentration and performance of economic activities. This paper has been funded by World Bank Research Grant 77960, "The Economic Geography and Political Economy of Industrial Location." Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Viktor Soukhanov, room MC2-523, telephone 202-473-5721, fax 202-522-3230, email address vsoukhanov@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at sanjoy@temple.edu or slalll @worldbank.org. June 2003. (35 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by Partnerships, Capacity Building, and Outreach METROPOLITAN INDUSTRIAL CLUSTERS: PATTERNS AND PROCESSES Sanjoy Chakravorty Department of Geography and Urban Studies Temple University, Philadelphia, PA, USA sanjoy@temple.edu Jun Koo Development Research Group The World Bank, Washington, DC, USA kkyojun@worldbank. org Somik V. Lall Development Research Group The World Bank, Washington, DC, USA slalll@worldbank.org The findings reported in this paper are those of the authors alone, and should not be attributed to the World Bank, its executive directors, or the countries they represent. This paper is part of a larger research effort on understanding the spatial dimensions of industry location. This paper has been funded by a World Bank Research Grant 77960, "The Economic Geography and Political Economy of Industrial Location." We are grateful to the Central Statistical Organization (CSO), Government of India, for making the firm level data available for this study. We would like to thank participants of the 2003 UNU/WIDER Tokyo conference on "Spatial Inequality in Asia" for helpful comments and suggestions. I METROPOLITAN INDUSTRIAL CLUSTERS: PATTERNS AND PROCESSES Industrial clusters have reemerged as important objects of research and policy analysis. The benefits of industry clustering were identified early by Alfred Marshall (1919) who suggested that these arise from localization economies: namely, the availability of common buyers and suppliers, the formation of a specialized/skilled labor pool, and the informal transfer of knowledge (on trade secrets, production processes, market agents etc.). Krugman's work in economic geography (Krugman, 1991, 1996) and Porter's work in business economics (Porter, 1990, 1996) have drawn the interest of economists to the idea of "increasing returns" to proximity in the form of clusters (see Fujita, Krugman, and Venables, 1999). Meanwhile, geographers have long been interested in industrial location and clustering, and theories of globalization and flexible accumulation using clusters of small, networked firms have been widely discussed following the pioneering work of Piore and Sabel (1984).1 This journal participated in this discourse by publishing a special issue containing a number of narratives on individual clusters (see Nadvi and Schmitz, 1999). This paper contributes to this growing literature on industrial clustering in developing countries. We focus on one geographical scale-the metropolis-and identify patterns and discuss processes of industrial clustering at that scale. The symbiotic relations between industrialization and metropolitanization are known well enough to be in the realm of common knowledge. It is understood that industrialization takes place in cities, some of which, usually important cities from the pre-industrial or colonial periods, Note that the questions on industry location and their implications are not new. Examining the locational aspects of economic activity has been of interest to geographers, planners, and regional scientists for some become so successful at attracting industry and migrants that they become very large cities or metropolitan regions. It is also understood that industrial success-in these cities is a cumulative causation process where increasing returns are derived from industrial clustering. However, there are several specific questions that remain unanswered. Where do industries locate within a metropolitan area? Do different industrial sectors have different pattems of location/clustering? Can these patterns be understood with reference to industry characteristics? What is the geographical relationship between clusters of different types of industry? To what extent do localization economies influence the clustering process? In this paper, we present the first estimates of industrial clustering within metropolitan regions in developing countries using data from three Indian metropolises- Mumbai, Kolkata, and Chennai. We first provide the necessary background - the theory of industrial clustering, cluster measurement methods, the data used for the analysis, and a brief introduction to the study areas. Next, we test eight industrial sectors (food/beverages, textiles, leather, printing/publishing, chemicals, metals, machinery, electrical/electronics) for evidence of global and local clustering, and distinguish between and test for co-clustering and co-location of industries. The results suggest an evolutionary model of industry location in mixed rather than specialized industrial districts. There is little evidence of localization economies from labor markets or buyer- supplier networks. We suggest that the key variable influencing the intra-metropolitan spatial distribution of industry is land use policy. time (Weber, 1929; Losch, 1956; Hotelling, 1929; Greenhut and Greenhut, 1975, Isard 1956). The more 2 BACKGROUND Why do industries cluster? Clustering is a term describing a phenomenon in which events or artifacts are not randomly distributed over space, but tend to be organized into proximate groups. Industrial clustering is a process that has been observed from the beginning of industrialization. From the cotton mills of Lancashire and automobile manufacturing in Detroit, to the textile mills of Ahmadabad and Bombay and the tanneries of Calcutta and Arcot, even even the casual observer can visually discern the evidence on industrial .clustering by industry type. Why should this happen? It seems obvious that competing firms in the same industry derive some benefit from locating in proximity to each other. The benefits that are external to the firm and accruing to similar firms in proximity are called the economies of localization. Now, these typically are not the only firns in the immediate region. There are usually other factories, producing other goods, distributed through other channels, sometimes for different markets. These other firms, and their employees, and the service workers who provide food, education, and health care for all these employees and their families, comprise, typically an urban area. All the firms that benefit from being in the urban area, regardless of whether or not there are other similar firms in the area, derive economies of urbanization from their location choice. In other words, there are productivity gains from industrial clustering. To put it in another way, at thefirn level, it is expected that the size and number of firms (i.e., the competitive structure) will influence internal returns to scale. In particular, as demand for a firm's goods and services increases (say, due to improved access to consumer markets), the entrepreneur has an incentive to increase scale of production by restructuring the production process through the use of specialized workers recent geographical approaches are summarized in Walker (2000). 3 and investing in cost reducing technologies (Lall et al., forthcoming). At the industry level, we expect to see quantifiable localized benefits of clustering which accrue to all firms in a given industry or in a set of inter-related industries. Productivity is likely to be higher in regions where an industry is more spatially concentrated due to increased potential of knowledge spillovers and buyer-supplier networks, access to a specialized labor pool and opportunities for efficient subcontracting. Finally at the metropolitan area level, economies of scale result not from the size of a specific industry or market but from the overall size, diversity, and spatial configuration of the metropolitan area. These economies of urbanization include access to specialized financial and professional services, availability of a large labor pool with multiple specializations, inter-industry information transfers, and the availability of less costly general infrastructure. At the interregional scale these gains are expected to lead to industry concentration in metropolitan and other leading urban regions (as a result of urbanization economies); at the metropolitan scale the gains from localization economies are expected to lead to the creation of local industrial clusters. These typically un-quantified agglomeration economies are one set of inputs into the location decision of a firm. There are other significant factors that a firm facing a location decision must consider. The two most important of these additional factors (especially in developing countries) are the availability of infrastructure, and the regulatory framework, both arenas where the state is the key player. The state not only sets the rules of market entry and participation, but is also the primary, often the sole provider of physical and social infrastructure (highways, airports, ports, export processing zones, etc. are examples of the former, while schools and hospitals are examples of the 4 latter), and is often directly active in the production process.2 The state's regulatory role at the local level goes beyond setting the rules of market participation; by being the single largest owner of land, by having the police and taking powers to acquire necessary land, and by being the final arbitrator on land use decisions, the state, as we shall show later, has a very strong influence on industrial location decisions within metropolitan areas. The measurement of clustering The ability to describe spatial pattems is the first step to understand spatial processes. Spatial statistics are the most widely used tools for describing and analyzing spatial pattems (see Getis and Ord, 1992 and Anselin, 1995 for excellent discussions on the subject). In classifying spatial pattems, researchers are often interested in determining whether the distribution of economic activity is clustered, dispersed, or random. Such spatial association of distribution pattems is also called spatial autocorrelation meaning that the attribute values being studied are correlated according to the geographic ordering of the objects. When the location of firms is spatially autocorrelated, that implies the geographic distribution of economic activity is not random and is likely to be determined by some underlying political and economic factors attributable to each geographical unit. Spatial associations are often measured for their strength. Strong spatial autocorrelations mean that the attribute values of adjacent geographical units are closely related (either positively or negatively). One of the most popular measures for spatial autocorrelation is Moran's I. There are two types of Moran's I depending upon geographic scales. The Global Moran is a measure describing overall spatial relationship across all geographical units. Therefore, only one value is derived for the entire study area. On the 2 In India, in its efforts to capture the "commanding heights of the economy," the state has invested heavily in capital-intensive industry, e.g., integrated steel and power plants. It was so successful in its efforts that as recently as the late 1990's, nine of the top ten and twenty of the top twenty-five corporations in India were public sector units (Nayar 1998). 5 other hand, The Local Moran (often called Local Indicator of Spatial Association or LISA) is a measure that is designed to describe the heterogeneity of spatial association across different geographical units. Global Moran's I can be defined as n z ,wt (x, - x)(x) - x) W (x1- x)2 where xi is the value of a variable of interest in areal unit i, wx, is a weight derived from a spatial weight matrix, and W is the sum of all cell values of the weight matrix. For the calculation of Moran's I, both a binary and a stochastic weight matrix can be used. A binary weight matrix defines the connectivity of pairs of regions with 0's and I's. When two regions are adjacent, the corresponding cell value is 1, but otherwise, it is 0. On the other hand, a stochastic weight matrix takes into account the number of immediate neighbors. Rather than assigning 1 to every neighbor, '1/total number of neighbor' is used as a weight for areal unit i. Equation (1) suggests that the calculation of global Moran's I is based on a comparison of values in neighboring geographical units. Note that the numerator is basically the covariance of neighbors, and the denominator is the sum of the squared deviation scaled by the total weight of the matrix. Therefore, if neighboring units have similar values over the entire study area, the statistic will show a strong positive spatial association. However, if dissimilar values are observed among neighboring units, the statistic should indicate a strong negative spatial association. The value of Global Moran ranges from -1 (negative spatial autocorrelation) to 1 (positive spatial autocorrelation). The statistical significance of Moran's I, which can be an important basis to determine whether the observed spatial association pattern arises from a random or from a systematic process, can be tested by-comparing 6 calculated Moran's I from equation (1) and the expected value of Moran's I from E(I) = -I I(n - 1) - The magnitude of spatial association is not necessarily uniform over the space. It is more likely to be heterogeneous according to local characteristics that influence the formation of spatial structure. Local Moran's I can be used as an indicator of heterogeneity in spatial association over geographical units and is defined as I, =zi Xw z, where;z, =(x,-x)/E where, zi is deviations from mean, and 8 is the standard deviation of xi. Similar to global Moran, a high value of local Moran means the association of similar values whereas a low value means the association of dissimilar values. Deriving local Moran statistics does not mean much unless one can determine observed spatial associations occur by chance or by a systematic process. By comparing calculated local Moran with its expected value based on E(I,) = -w, /(n - 1), the significance of local Moran statistics can be tested. Data issues In order to undertake sub-metropolitan level analysis it is necessary to have spatially disaggregated data. In India industrial data are collected by the Central Statistical Organization (CSO) and disseminated as the Annual Survey of Industries (ASI). In the late 1 990s the ASI data were first released at the district level and then at the firm level. These data are based on a survey carried out by CSO on a sample taken from an industry sampling frame which includes every registered (or legal) industrial unit with at least ten workers. This sampling frame contains one record for each industrial unit and includes three critical pieces of information: the National Industrial Classification (NIC) code, the number of workers in the unit, and the street address, with, sometimes, a pin code (equivalent to U.S. zip codes). 7 The last piece of information is the key to disaggregating the district data down to smaller enumeration units. We are fortunate to have access to the sampling frame for the whole country for the enumeration period 1998-99. On further examination we found that while street addresses were generally available for all metropolitan areas, there were no base maps of streets to which these addresses could be matched. Hence we had to rely on the pin code information, which, however, turned out to be erratically available. For some cities the pin codes were generally available or imputable, for other cities they simply were not available. We identified Mumbai, Kolkata, and Chennai as the three metropolises with enough information to begin geocoding the industry location data to the pin code level. The pin code maps were acquired from a private sector firm in New Delhi (ML Infomap, 1998). These maps have somewhat variable coverage. For Kolkata and Chennai the pin code maps cover the largest definition of their metropolitan areas. For Mumbai the pin codes cover the district of Greater Bombay only; that is, the far northern and eastern suburban reaches of the Mumbai metropolitan area (in Thane and Raigad districts) are not covered. Even in the covered area there appear to be some situations where adjoining pin codes have been merged. As a result some data that are known to be in the metropolitan area could not be geocoded to pin codes. We have been able to achieve the following "hit" rates, that is successfully geocoded factory records: In Mumbai 99.9 percent, in Kolkata 97.1 percent, in Chennai 97.5 percent. Following Ratcliffe (2002), who argues that a "hit rate" of 85 percent is acceptable for most map-based analysis, we believe that these are acceptable levels of address matching. Finally we aggregated the firms into eight distinct and internally consistent sectors. These are (with NIC codes in parenthesis): 1. Food Processing (151, 152, 153, 154, 155) 2. Textiles and Textile products, including wearing apparel (171, 172, 173, 181) 8 3. Leather and leather products (191, 192) 4. Paper products, printing and publishing (210, 221, 222) 5. Chemical, chemical products, rubber and plastic products (241, 242, 243, 251, 252) 6. Basic Metals and Metal Products (271, 272, 273, 281, 289) 7. Mechanical Machinery and Equipment (291, 292) 8. Electrical and Electronics (including computer) Equipment (292, 300, 31, 32) The study areas Mumbai (Bombay), Kolkata (Calcutta), and Chennai (Madras) are three of India's big four metropolitan areas. All are colonial cities, created by the British primarily for administrative purposes (see Kosambi and Brush, 1988 on their structure and morphology). Our study areas cover the city of Mumbai (with 94 pin codes), metropolitan Kolkata (with 133 pm codes), and metropolitan Chennai (with 108 pin codes). Mumbai is the sub-continental leviathan. The population of the metropolitan area is estimated to be over 18 million, making it, with Mexico City, the second largest urban agglomeration in the world. The metropolitan population has grown by around 50 percent in each of the two preceding decades. Mumbai is the center of the financial sector in India, home of the Reserve Bank and the majority of the state-owned and international banks and financial institutions. It is a major industrial hub, one end of India's most dominant industrial region stretching north up to Ahmedabad in Gujarat state; this region has done particularly well in attracting new industrial investments after the 1991 structural reforms (Chakravorty 2000). Kolkata was the most important colonial city in India, the seat of the British empire and its political center till 1911, and the industrial center of the nation till the early 1960s. 9 The Kolkata metropolitan region is the second largest in the country, after Mumbai, with around 12.5 million people; till the 1991 census it had been India's largest metropolis. The city and the region have seen a dramatic decline in industry in general; its specializations are in "sunset" industries, specifically jute (textiles) and iron and steel. The incomes of the city and the state have declined from an all-India high in 1960 to around the national average now. The state of West Bengal has been ruled by the Communist Party of India (Marxist), popularly known as the CPM, which resisted the 1991 structural reforms and continues to be seen to be deeply suspicious of capitalism and globalization. Chennai is the fourth largest metropolitan area in India (behind the other two and Delhi). It was the preeminent city of south India for most of the twentieth century, till Bangalore, in the neighboring state of Kamataka, emerged as a serious regional rival, primarily as a result of the growth of the information technology sector. Chennai continues to be a stronger industrial center than Bangalore. It is also part of a large industrial region in the state of Tamil Nadu with the cities of Madurai and Coimbatore. ANALYSIS Global clustering The measures of global clustering for the eight industry sectors for the three metropolises are reported in Table 1. The results are reported separately for number of factories and number of workers. This method of presentation, which is used throughout this paper, is based on the assumption that factory size has some bearing on the location decision. Large factories can be expected to rely more on intemal economies of scale; small factories may rely on extemal or localization economies. Here the variable "factories" suggests small- scale units; when factories are clustered it suggests that a large number of small firms are 10 clustered. Conversely, when workers are clustered we can assume that a small number of large firms are clustered. In general, clustering is most consistent in Kolkata (other than textile factories and metals and electrical/electronic workers everything else is clustered); Chennai and Mumbai show inconsistent evidence of clustering with more consistent clustering in factories than workers. Let us discuss the results by industry sector. In the food/beverages sector, factories are clustered in all three metropolises (most weakly in Chennai), and workers are clustered in Kolkata and Mumbai, but not in Chennai. The textiles sector, which is by far the largest in terms of both factories and workers, has strongly clustered factories only in Chennai; there is evidence of weak clustering in Mumbai and no clustering in Kolkata. Textile sector workers, however, are strongly clustered in Kolkata and Mumbai, but not clustered in Chennai. In the leather sector, in Kolkata both factories and workers are strongly clustered, whereas in Chennai and Mumbai neither factories nor workers are clustered. In the printing/publishing sector, both factories and workers are very strongly clustered in Kolkata, strongly clustered in Chennai, and not clustered in Mumbai. Factories in the chemicals sector are clustered in all three metropolises, while workers are clustered only in Kolkata and Chennai. Factories, but not workers, are clustered in Kolkata and Chennai; in Mumbai neither factories nor workers are clustered. Factories in the Machinery sector are clustered in all three metropolises (weakly in Mumbai and strongly in the other two); workers are very strongly clustered in Kolkata, but not clustered in Mumbai or Chennai. Finally, in the appliances sector, factories are clustered in all three metropolises, but workers are clustered only in Mumbai. In summary, three possible combinations are seen. As expected, clustering among factories and workers in the same industry and city is the most common pattern. However, 11 combinations with only one of them (factories or workers) being clustered are also common. Only factories are clustered in metals and electrical/electronics (in both Kolkata and Chennai). Only workers are clustered in the textiles sector in Kolkata. At this point in the analysis the picture is unclear. Local clustering Next we take the analysis one step beyond the simple confirmation of the existence of clusters. In this section we identify the locations and other characteristics of the clusters. Recall from our earlier discussion on the measurement of clustering that local clusters may exist even when the system as a whole is not clustered (hence the distinction between "global" and "local" Moran indices). These results are reported in Table 2 and in a series of maps (Figures 1, 2, and 3). It is useful to explain what is reported in these figures and the table. We began the analysis by calculating local Morans for each of our eight industry categories, for each of our three cities, for factories and workers separately. We used two methods to calculate local Morans: first, neighbors were identified on the basis of contiguity, and then on the basis of a distance cut-off. Only the contiguity-based local Morans are reported here, in the table and in the figures. The distance based measures yield about the same results. Now, consider Figure 1. The first pair of figures from the top left shows the distribution of local Morans for the Food and Beverage sector in Mumbai; the map on the left shows the local Morans for factories, the map on the right shows the distribution for workers in that industry. The strength of the clusters are determined by the statistical significance of each pin code's local Moran calculation.3 Next, the numbers of pin codes forming clusters and the numbers of 3 Z-values for local Morans are mapped. A negative cluster has Z values less than -1.65 (this is not shown). Z values between -1.65 and 1.65 are not clustered. Values of Z between 1.65 and 1.95 are weakly clustered, and values greater than 1.95 are stronigly clustered. Pin codes that show weak and strong 12 factories and workers in their respective clusters were summed. These data are also reported in Table 2. Note the great variation in the results between cities, between industries, and within cities and within industries. Let us identify some of the observed pattems. First, factories and workers in the same industry do not necessarily cluster in the same pin codes. In the general case there are some common pin codes and some unique pin codes for each industry. There are two types of exceptional cases: one where there is perfect overlap between factory and worker clusters (such as Printing/Publishing in Mumbai and Machinery in Chennai); the other in situations where there are no common pin codes (such as Food and Beverages in Mumbai, where there are five unique clustered pin codes each.for factories and for workers) or few common pin codes (such as Textiles or Machinery in Kolkata). This variation is seen in every city, with perhaps Chennai showing less evidence of the extreme situations. This is an important finding. It suggests that within the same industry small-scale operations tend to cluster together (this is where the factories are seen to cluster), often, but not always at separate locations from large-scale operations (where workers are seen to cluster). Later we discuss the implications of this finding. Second, and related to the first point, it is difficult to discem whether tactones are more clustered or workers in specific industries. Recall the argument that large factories rely on intemal economies of scale for productivity gains, whereas smaller factories rely on extemal economies, at least some of which are derived from localization or clustering. Hence, in our data, we can expect factories (small scale units) to cluster more and workers (large scale units) to cluster less. If we use the percentage of factories or workers within clusters as a measure of the extent of clustering, this expectation is correct across cities in only the Printing/Publishing and Metals sectors. The opposite is true in the Food/Beverages evidence of positive cl ustering are shown in the maps using different shades. In general, where local clustering does exist it does so at Z values greater than 1.95. 13 sector, and with the exception of Chennai (where the numbers are close) in the Textiles sector. This may not be a significant issue as it is unclear that the measure used here is an appropriate measure for comparing two very different types of units (factories and workers). Third, each city has one or two industries that appear to be more clustered than others-for instance, the Textiles and Electrical/Electronic sectors in Mumbai, Leather and Food and Beverages in Kolkata, and Leather in Mumbai. These are also the industries that generally have a significant overlap in terms of the locations of factory and workers clusters (with the exception of Textiles in Mumbai). Moreover, these are also the industries for which these cities have high location quotients (LQs) at the national level.4 The Electrical/Electronic sector in Mumbai has a LQ of 3.0, the Textiles sector's LQ is 1.7. The Leather sector's LQ in Chennai is 4.0, in Kolkata it is 1.5. Food and Beverages in Kolkata is an exception to this pattern. Another exception is the Machinery sector in Mumbai; its LQ is 2.5, yet it is the only industry in any of our study cities to have absolutely no local clustering among workers (and one of the lowest levels of clustering among factories in all cities). Therefore, though this pattern cannot be generalized, there may be a causal relationship between very high levels of clustering and industry dominance at the regionalnational level. At this point, however, it is difficult to determine the direction of the causal arrow. Fourth, the location of industry clusters generally appears to follow some widely held principles: one, polluting industries are located in fringe areas, and two, many industries are co-located in industrial districts. Consider the first principle. It seems obvious that any local regulatory agency will direct the location of polluting industry toward the urban fringe. The two most polluting industries considered here are Chemicals and Leather. In all three cities these industries are located in the fringe areas, and it appears from the maps that in the cases 4The location quotient (LQ) is simple measure of regional concentration used in regional science. It calculates the ratio of the share of a given variable to the share of population. Here, LQ = I indicates that the region's share of a particular sector is equal to its share of all industry. If LQ = 3 it indicates that the region's share of that sector is three times its share of all industry. 14 of Mumbai and Kolkata, there is some degree of co-location of these two industries. The Chennai case is particularly interesting for the Chemicals sector. Chemical factories are clustered in the southem extremity and in the far west, whereas workers (or large scale factories) are located in the northem fringes of the metropolitan area. Co-clustering or co-location The idea of the industrial district goes back to Marshall (1919) who suggested that small specialized firms would tend to cluster in space to derive extemal economies to offset the internal scale economies of large factories. Piore and Sabel (1984) argued that the late twentieth century had seen the arrival of a "second industrial divide" where the vertically integrated organization of production characterized by Fordist manufacturing was giving way to regional specialization organized around networks of small scale producers. Geographers see this in terms of the need for flexible specialization in globalized production systems geared toward rapid changes in technology and the need to respond to shifting patterns of demand (see Amin 2002). Economists have focused on the specific productivity advantages provided by proximity. In the simplest terms, these localization economies (to be distinguished from urbanization economies that accrue to all firms in an urban area) arise from two sources: local labor markets and knowledge spillovers. Labor markets: Do thick local labor markets create localization economies? We need to distinguish between industries based on unskilled labor (e.g., leather) from those based on skilled labor (printing/publishing). Consider labor markets with unlimited supplies of unskilled labor operating in land markets with minimal land use / zoning regulations but where land rents decline with proximity to industry. In other words, unskilled labor is available in plenty and can locate anywhere, but is likely to locate near factories where rents are lowest. Therefore, unskilled labor is likely to co-locate with industrial clusters. Does this 15 generate localization economies for those industries that are reliant on unskilled labor? If it does we are likely to see unskilled labor-based industries more oriented toward locating in industrial clusters with other unskilled labor-based industry that do not share buyer-supplier networks. These clusters will not include industries based on skilled labor. If these latter (skill-based) industries do cluster, they do not do so for labor market localizafion economies, as their critical workers, the high-skill high-wage labor, will not co-locate with industry. They do so for other reasons (discussed below). In summary, labor market localization economies, if any exist, are likely to be industry specific and inversely related to the proportion of skilled labor in a given industry. Knowledge spillovers: These are of two kinds: technology spillovers through informal interaction and information spillovers on market agents. Technology spillovers are irrelevant in low technology firms and industries in mature stages of the product cycle. The vast majority of manufacturing industry in India belongs in this category. Hence, for these industries we can argue that there are negligible localization economies from technology spillovers. Information spillovers on market agents such buyers and suppliers, however, are more likely to provide localization economies. Firms of all sizes (except perhaps the very largest vertically integrated firms) rely on dense buyer-supplier networks. Firms benefit from having to access to local buyers and suppliers, and knowledge pooling on buyer-supplier behavior is likely to eliminate inefficient agents. Hence, theory suggests the existence of two types of industrial districts: labor-sharing industrial districts that depend on the availability of low skill labor, and buyer-supplier-linked industrial districts where industries that have market interactions with each other benefit from co-location. We do not have spatially disaggregated wage data to test explicitly for the existence of labor-sharing districts. We are, however, able to establish what the expected buyer-supplier links are between our industry groups by assuming that the input-output links 16 at the national level are replicated at the local level. The national input-output tables are available. We assume that similar input-output linkages exist at the level of the metropolis and argue that industries with strong input-output linkages are likely to co-locate. It is useful at this point to distinguish between co-location and co-clustering. Co- location occurs when industries from two sectors are present in the same small region. Co- clustering occurs if both industries that are co-located are linked through input-output, innovation or labor market linkages. Before we examine the evidence on buyer-supplier linkages, let us first examine the data on co-location by industry for all sectors. The correlation coefficients for factories and workers for all eight industry groups are reported in Table 3. The data in this table are quite remarkable. There is strong evidence that industry groups co-locate at the pin code level, especially in the case of factories, or small scale units. In Mumbai, for instance, factories for every industry group are seen to have a statistically significant correlation with every other industry group. In general the correlation coefficients are very high: 0.93 between Machinery and Chemicals, 0 .91 between Machinery and Metals, etc. In the case of workers, however, the correlations are not as high, and fewer are statistically significant. In Chennai, the pattem is even more pronounced. With the exception of the Leather sector (which has a moderate but significant correlation with only the Textiles sector) the correlation coefficients of every other pair of industries is significant, and as in the case of Mumbai, often very high. The Leather sector is anomalous in Kolkata as well. It's only significant correlation in that metropolis, is with the Chemicals sector. But Kolkata itself is somewhat of an anomaly compared to Mumbai and Chennai, at least in terms of the co-location of workers. Factories in Kolkata, with the exception of the Leather sector, are generally co-located; however, workers in at least four industries-Food/Beverages, 17 Textiles, Leather, and Printing/Publishing-are generally not correlated with workers in other industries. This is an intriguing finding whose implications we will discuss later. It is one thing for industries to be co-located, it is another for them to be co-clustered. If the reasoning on buyer-supplier clusters outlined above is correct we can expect: (a) that industries that have strong input-output links will co-cluster; that is, they will form clusters in the same or proximate pin codes, and (b) industries that do not share buyer-supplier linkages will not co-cluster; if they do it will be for labor-sharing, and therefore they will share similar labor profiles (low-skill with low-skill, or high-skill with high-skill). In order to test this hypothesis we conducted co-clustering tests on four industry pairs; two of these pairs have strong input-output linkages (the two highest among our eight industry groups); Metals and Machinery with an 1-0 factor of 34.95, and Metals and Electrical/Electronic with an 1-0 factor of 25.75. The two other pairs have virtually no input-output linkages; Food/Beverages and Electrical/Electronic have an I-0 factor of 0.07 and very different labor profiles (the former is low skill, the latter is high skill); Textiles and Metals have an I-0 factor of 0.41 and potentially similar labor profiles. We combine the proportion (rather than raw)5 data on factories and workers for these four industry pairs by pairs, and conduct tests of global clustering (Moran's I) and local clustering. The results are mixed. Consider the results of the global clustering test first (in Table 4). In Mumbai, among the strong I-0 pairs, only factories in the Metals-Electrical combination are clustered, whereas the Food-Electrical combination, which shares neither I- 0 links nor similar labor profiles, shows clustering for factories and workers. In Kolkata, all four pairs are clustered, for factories and workers (more for the former than the latter). In Chennai, as in Mumbai, factories in the Metals-Electrical combination are clustered; but factories in both weak 1-0 combinations are clustered. The strongest'l-0 pair-Metals and 5 If raw data are used, the dominant industry determines the outcome of paired calculations. 18 Machinery-is not only not clustered in Mumbai and Chennai, but the value of Moran's I is negative half the time. Next we look at a final set of data: the share of industry in the top pin codes. Table 5 lists these data by metropolis, for all industry and by sector. (Note that in a given metropolis the top 10 or top 20 pin codes are not separately identified for each sector. What is reported is the share of each sector in the top overall pin codes.) The results are not surprising in the context of what we have presented earlier, but they are quite effective in making the point that industry is concentrated in a handful of pin codes in all three cities, and that all sectors are heavily represented in these top pin codes. In Mumbai the top ten pin codes include close to 55 percent of all factories and workers, and the top 20 pin codes include over 76 percent of all factories and about 80 percent of all workers. In Chennai and Kolkata the proportions are progressively smaller. This decline is probably a function of the fact that the total number pin codes in Kolkata (133) and Chennai (108) are higher than in Mumbai (94). In general, regardless of the number of pin codes in a metropolis, the top ten percent of the pin codes include close to 50 percent of all factories and workers. Summary of the findings (1) In Indian metropolises industry is generally clustered-the evidence for clustering is found at level of the metropolis (using the global Moran statistic), and certainly at the level of the pin code (using local Moran statistics and maps). On average about 50 percent of all workers and factory units are concentrated within ten percent of the most industrialized pin codes. At the sectoral level there are extremes: in one case there are no clusters at all; at the other end there are instances where 70 percent or more of workers/factories are concentrated within six to eight pin codes (see Table 2). 19 (2) The clusters are of two types: first, where factories and workers are both clustered in the same region within a metropolis; second, where factories are clustered at locations separate from where workers are clustered (i.e., separate clusters of small scale operations and large scale operations). The second pattern is more common. (3) A small number of pin codes account for a very large proportion of all industry, both factories and workers. As a result the extent of industry co-location is very high. However, the expected relationships between industrial sectors-whereby industries with strong input-output linkages are expected to co-locate, and industries using similar labor profiles are expected to co-locate-are not found. On the contrary, we see several examples of counterintuitive co-locations. (4) Some industries have distinct locational properties. For example, where the Leather industry is significant (as in Kolkata and Chennai) it is located on the urban fringe and is not co-located with other industries except Chemicals. The Printing/Publishing industry is located near the urban core in all three cities. Both industries are highly clustered, and are also examples of cases where workers and factories are clustered in the same pin codes. The Textile industry, on the other hand, the largest industry, is marked by separate factory and worker clusters, where the former is closer to the city center than the latter (except in Mumbai). (5) The extent of clustering is highest in Kolkata and lowest in Chennai. However, the extent of co-location follows the reverse pattern: highest in Chennai and lowest in Kolkata. This is indicative of the existence of different types of industrial districts, where the ones in Chennai have a greater mix of industries than the ones in Kolkata. These industrial districts appear as spatially separate bands or sectors. 20 An Explanatory Framework How can these pattems be explained? To begin with, we suggest that the default approach outlined earlier-whereby localization economies drive cluster formation-may be limited in its explanatory power. There is little evidence in support of the processes of localization, either via local labor markets or via buyer-supplier networks. It is possible to argue that these findings are artifacts of the method of local cluster identification. That is, had we used other methods that would have enlarged the definition of "local" beyond the one used here (a pin code and its contiguous neighbors), there would have been stronger evidence in support of localization economies. That, however, is unlikely. First, at a preliminary stage in our investigations we did indeed use larger definitions of "local", with virtually indistinguishable results. Second, it may be possible to enlarge the definition of local till it is meaningless or large enough to encompass a significant portion of the metropolitan area. At that point it becomes difficult to distinguish localization economies from urbanization economies. Perhaps localization economies arise as a result only of inter-firm technological interaction within an industrial sector (because following the national input-output data, inter- firm trade in most of these sectors is not very high). It is difficult to make that case for many of these sectors, which are dominated by old firms that are not close to the technological frontier (Lall and Rodrigo, 2001). Therefore, it is possible that the debate on the relative strength of localization and urbanization economies should be resolved firmly in favor of the latter. Looking beyond localization economies, we suggest that firm locations are guided by a complex set of factors which often rule out most spaces within metropolitan areas. These factors include the accidents of history, metropolitan expansion, state regulations (especially ones that affect the land market), and industry characteristics. As a result we see the usually unplanned evolution of mixed industrial districts, which include a variety of related and 21 unrelated industry sectors, it leap-frogging locations within metropolitan areas. We suggest that an evolutionary framework may explain the observed patterns of industry location in Indian metropolises. This framework is speculative (we do not have the temporal information needed to make a stronger claim), but it is able to account for many of the expected and unexpected findings. First, at some specific historic point the pioneer industrial unit in a specific industrial sector makes a location decision within the metropolis. The driving force behind that decision (not the location decision, but the decision to start a new industry) may have one or several motivations: nationalism (as in the case of the first textile mill in 1854 in Mumbai), war (as in the case of the first leather units in Kolkata, created to outfit saddles for the, Imperial army during the first world war), bureaucratization and the spread of literacy (creating the need for printed matter), etc. This first factory did not rely on localization economies to boost productivity, but probably relied on urbanization economies, at least in terms of providing market access and pool of labor. This unit located in what was then the urban fringe, beyond dense population settlements, but close enough and connected enough for workers to reach the plant. Following the convention of the time, this unit was large, relying on internal economies of scale to reduce costs. A cluster of firms in the same industry began forming around this original location. It is difficult to determine whether these subsequent location decisions were the result of localization advantages arising from labor pooling, advantages derived from shared infrastructure such as railheads, or state regulations that directed new firms to this location. It will be necessary to conduct archival research for more concrete judgments. It is very likely that at some point industries in this cluster began to derive benefits from labor pooling. Two factors should be considered here: First, these were not high technology industries of the time; they were industries in the late stages of the product cycle (Vernon, 1966), reliant on 22 unskilled and semi skilled labor. Second, the location of these industries began influencing the land market around them; because of the local environmental impacts of these industries, the only bidders for the proximate land were other industries or low wage labor. In other words, this sub-region became an industrial district, with large-scale factories and slums/tenements for unskilled workers. As the city continued to grow beyond this industrial district most new industry was directed to this district by state regulations. The single purpose industrial district became a mixed industrial district. Where was even newer industry to locate? With all the land in the original industrial district in use, and with state regulations that forbade the conversion of any land with housing (slums, tenements, or middle class residences) to industrial use, new industries now sought new locations on what was then the urban fringe. The role of the state in discouraging land use change turns out to be a critical influence on industrial location (as we shall see below). The cycle of industrial district formation began again, this time with more active involvement of the post-colonial state which assisted with land acquisition (often the most difficult part of urban industrial location) and provided physical infrastructure. When this cycle was complete (i.e., the point was reached where there was no more land for new factories) new industrial units leap-frogged over the residential communities that had grown in the interim to new locations at the urban fringe. This is the current stage. Now there is even more active involvement of the state, which sets up export processing zones, free trade zones, technology parks, industrial parks etc. to entice new industrial units. (Consider the names of some of the most active pin codes: in Mumbai, Chakala and SEEPZ, both Maharashtra Industrial Development Corporation centers; in Chennai, Ambattur Industrial Estate and SIDCO Industrial Estate, both set up by the state of Tamil Nadu.) This stylized framework explains a number of observations-some regularities, some irregularities-listed in the preceding section. In this framework, the exact location of 23 specific industries has to be understood in terms of the functions of these industries. For instance, the Printing/Publishing industry remains located close to the CBD, which is its principal market area. The Leather industry, which pollutes both the air (with the smell of animal hides) and the groundwater, and therefore cannot even co-locate with other industries (not to mention residential areas), remains at the urban fringe-when the fringe moves beyond or envelopes the Leather cluster, the state compels the cluster to move' further out. This is exactly what has happened in Chennai and is happening in Kolkata. This framework also explains some anomalies, such as the location of a large textile cluster near Mumbai CBD. This cluster should not exist, for the land there is too valuable to remain devoted to an industry with outmoded technology and very low value addition. Yet it remains a textile cluster, with virtually closed factories, state take-over (or nationalizing) of "sick" units, and job losses that mount by the year, because the state will not permit the conversion of this industrial land to commercial or residential use (D'Monte, 2000). The other significant anomaly comes from the relative locations of factory and worker clusters. It is not unusual to see these clusters form separately, but the expectation is that worker clusters (which are clusters of large factories), which require more undivided land; will locate on the least expensive land, furthest from the center. Yet when there is no exit policy, i.e., factories are not allowed to close and factory land cannot easily be transferred, it is possible to see clusters of large scale units near the center of the city. And most important, this framework explains why co-location is so common, but theoretical expectations on co-clustering are not realized in practice. We conclude by highlighting two implications of these findings. First, mixed industrial districts are the norm in Indian (and perhaps most developing nation) metropolises. This study is unable to determine whether the past industrial successes of these metropolises arose from this fact; but it suggests that single sector industrial districts, especially ones that 24 cannot benefit from urbanization economies, are unlikely to succeed. Second, land use policies deeply influence industry location. A policy regime that does not allow land use change creates significant inefficiencies in industrial location and, by extension, commercial and residential location. We recognize that land is the most precious and emotive commodity in urban areas, and the political economy of land ownership involves every metropolitan resident. However, the distributional and welfare implications of current policies are unclear. A careful analysis of the issue would be very useful to metropolitan governments in developing nations. 25 References: Amin, A. 2002. Industrial Districts. In A Companion to Economic Geography. E. Sheppard and T. J. Barnes Eds. Oxford: Blackwell. Anselin, L. 1995. Local Indicators of Spatial Association: LISA. Geographical Analysis 27:93-115. Chakravorty, S. 2000. How Does Structural Reform Affect Regional Development? Resolving Contradictory Theory with Evidence from India. Economic Geography 76:367-394. D'Monte, D. 2000. Ripping the Fabric: The Decline of Mumbai and its Mills. New Delhi: Oxford University Press. Fujita, M., & Thisse, J.-F. 1996. Economics of Agglomeration. Centerfor Economic Policy Research Discussion Paper, No 1344. Fujita, M., Krugman, P., & Venables, A. 1999. The Spatial Economy: Cities, Regions, and International Trade. Cambridge, MA: MIT Press. Getis, A., and J. K. Ord, 1992. The Analysis of Spatial Association by Use of Distance Statistics. Geographical Analysis 24:189-206. Greenhut, J., and Greenhut, M. L. 1975. Spatial Price Discrimination, Competition and Locational Effects. Economica, 42, 401-419. Henderson, V. 1988. Urban Development: Theory, Fact, and Illusion. New York: Oxford Univ. Press. Hotelling, H. 1929. Stability in Competition. Economic Journal, 39, 41-57. Isard, W. 1956. Location and Space-Economy. Cambridge: MIT Press. Kosambi, M. and Brush, J. E. 1988. Three Colonial Port Cities in India. Geographical Review 78:32-47. Krugman, P. 1991. Geography and Trade. Cambridge, MA: MIT Press. Krugman, P. 1996. Urban Concentration: The Role of Increasing Returns and Transport Costs. International Regional Science Review 19:5-30. Lall, S. V. and Rodrigo, C. (2001). Perspectives on the Sources of Heterogeneity in Indian Industry. World Development 29, 12: 2127-2143. Lall, S.V., Shalizi, Z., and Deichmann, U. (forthcoming). Agglomeration economies and productivity in Indian industry. Journal of Development Economics. Also, World Bank Policy Research Working Paper 2663. Losch, A. 1956. The Economics of Location. New Haven, CT: Yale Univ. Press. 26 Marshall, A. 1919. Industry and Trade. London: Macmillan. MLInfomap. 1998. IndiaMap. Delhi: ML Infomap Pvt. Ltd. Nadvi, K and H. Schmitz (Eds.). 1999. Industrial Clusters in Developing Countries (Special Topic). World Development 27: 1503-1704. Nayar, B. R. 1998. Business and India's Economic Policy Reforms. Economic and Political Weekly 33: 2453-68. Piore, M. J., and Sabel, C. F. 1984. The Second Industrial Divide: Possibilities for Prosperity. New York: Basic Books. Porter, M. 1990. The Competitive Advantage of Nations. London: Macmillan. Porter, M. 1996. Competitive Advantage, Agglomeration Economies, and Regional Policy. International Regional Science Review 19: 160-76. Ratcliffe, J. 2002. Geocoding Crime and a First Estimate of a Minimum Acceptable Hit Rate, (mimeo). Vemon, R. 1966. International Investment and International Trade in the Product Cycle. Quarterly Journal of Economics, 80:90-207. Walker, R. A. 2000. The Geography of Production. In A Companion to Economic Geography. E. Sheppard and T. J. Barnes Eds. Oxford: Blackwell. Weber, A. 1929. Theory of the Location of Industries. Chicago: Univ. of Chicago Press. 27 Table 1. Indices of global clustering Total Moran's I Total Moran's I Factories for factories Z of I Workers forworkers Z of I Mumbal Food/Beverages 274 0.172 2.951 16,992 0.156 2.699 Textiles 1,988 0.100 1.797 131,318 0.191 3.259 Leather 115 0.074. 1.374 2,629 0.018 0.474 Printing/Publishing 882 0.060 1.139 17,001 0.048 0.959 Chemicals 1,146 0.149 2.591 41,470 0.008 0.031 Metals 894 0.069 1.290 25,056 -0.019 0.145 Machinery 767 0.104 1.867 40,808 -0.045 0.560 Electrical/Electronic 802 0.244 4.133 58,871 0.141 2.455 Kolkata Food/Beverages 355 0.165 3.332 9,573 0.170 3.429 Textiles 355 0.063 1.354 90,621 0.168 3.380 Leather 283 0.211 4.216 6,149 0.316 6.229 Printing/Publishing 428 0.366 7.203 10,574 0.255 5.071 Chemicals 671 0.214 4.274 24,871 0.147 2.992 Metals 1,023 0.178 3.582 38,440 0.047 1.067 Machinery 451 0.130 2.655 13,322 0.459 9.003 ElectricalElectronic 511 0.168 3.400 18,320 0.051 1.129 Chennal Food/Beverages 166 0.094 1.798 8,459 0.069 1.353 Textiles 1,446 0.135 2.495 107,266 0.039 0.694 Leather 406 0.058 1.161 17,919 0.087 1.666 Printing/Publishing 402 0.230 4.135 9,361 0.141 2.595 Chemicals 512 0.126 2.329 17,920 0.132 2.443 Metals 695 0.184 3.337 29,076 0.006 0.266 Machinery 416 0.153 2.799 18,772 0.039 0.829 Electrical/Electronic 359 0.127 2.361 21,018 -0.003 0.098 Data Source: CSO Government of India, ASI sampling frame, 1998-99 28 Table 2. Concentration in clusters Percent Percent # of Pin # of Pin factories Workers workers codes in codes in Total Factories in Total in in factory worker Common Factories in Clusters Clusters Workers Clusters Clusters clusters clusters Pin codes Mumbal Food/Beverages 274 79 28.8% 16,992 7,791 45.9% 5 5 0 Textiles 1,988 703 35.4% 131,318 62,649 47.7% 5 6 2 Leather 115 45 39.1% 2,629 691 26.3% 4 3 1 Printing/Publishing 882 203 23.0% 17,001 3,567 21.0% 2 2 2 Chemicals 1,146 508 44.3% 41,470 5,623 13.6% 7 3 2 Metals 894 307 34.3% 25,056 1,347 5.4% 4 1 1 Machinery 767 185 24.1% 40,808 0 0.0% 3 0 0 Electrical/Electronic 802 318 39.7% 58,871 34,575 58.7% 4 3 3 Kolkata Food/Beverages 355 122 34.4% 9,573 6,034 63.0% 5 9 4 Textiles 355 120 33.8% 90,621 38,271 42.2% 6 7 1 Leather 283 209 73.9% 6,149 4,393 71.4% 6 8 6 Printing/Publishing 428 209 48.8% 10,574 4,155 39.3% 10 9 9 Chemicals 671 224 33.4% 24,871 10,053 40.4% 7 9 7 Metals 1,023 357 34.9% 38,440 8,166 21.2% 4 3 2 Machinery 451 101 22.4% 13,322 4,217 31.7% 6 6 1 Electrical/Electronic 511 170 33.3% 18,320 5,283 28.8% 9 6 4 Chennal Food/Beverages 166 45 27.1% 8,459 3,065 36.2% 4 4 2 Textiles 1,446 313 21.6% 107,266 22,447 20.9% 7 3 2 Leather 406 204 50.2% 17,919 9,932 55.4% 4 6 4 Printing/Publishing 402 161 40.0% 9,361 2,534 27.1% 8 5 5 Chemicals 512 137 26.8% 17,920 5,333 29.8% 6 5 2 Metals 695 240 34.5% 29,076 5,755 19.8% 4 3 2 Machinery 416 167 40.1% 18,772 5,222 27.8% 2 2 2 ElectricalElectronic 359 157 43.7% 21,018 1,065 5.1% 4 1 1 Data Source: CSO Government of India, ASI sampling frame, 1998-99 29 Table 3. Correlation coefficients for Industry pairs Foodl/ Prnting/ ElectricaV Beverages Textiles Leather Publishing Chemicals Metals Machinery Electronic MUMBAi Food/Beverages 1.00 0.39 0.31 0.39 0.34 0.40 0.40 0.25 Textiles 0.10 1.00 0.58 0.79 0.77 0.73 0.77 0.63 Leather 0.20 0.23 1.00 OA3 0.77 0.63 0.75 0.63 Printing/Publishing 0.16 0.74 0.35 1.00 0.49 0.44 0.51 OA5 Chemicals 0.25 0.37 0.42 0.33 1.00 0.91 0.93 0.81 Metals 0.16 0.26 0.18 0.25 0.56 1.00 0.87 0.70 Machinery 0.09 0.17 0.15 0.07 0.34 0.15 1.00 0.78 ElectricaVElectronic 0.27 0.07 0.16 0.10 0.21 0.25 0.06 1.00 KOLKATA Food/Beverages 1.00 0.31 0.09 OA6 OA3 0.33 0.46 0.40 Textiles 0.07 1.00 0.04 0.36 0.64 0.33 0.51 0.51 Leather 0.02 -0.05 1.00 0.03 0.49 0.07 0.12 0.10 Printing/Publishing 0.12 0.03 0.08 1.00 0.34 0.15 0.37 0.32 Chemicals 0.17 0.19 0.33 0.10 1.00 0.36 0.60 0.72 Metals 0.14 0.09 0.14 0.07 0.25 1.00 0.76 0.33 Machinery 0.21 -0.03 0.04 0.26 0.22 0.41 1.00 0.68 Electrical/Electronic 0.25 0.06 0.02 0.08 0.37 0.08 0.30 1.00 CHENNAI Food/Beverages 1.00 0.30 0.02 OA7 0.69 0.59 0.52 0.62 Textiles 0.15 1.00 0.29 0.61 0.51 0.40 0.41 OAO Leather -0.01 0.35 1.00 0.10 0.16 0.11 0.14 0.15 Printing/Publishing 0.25 0.32 0.17 1.00 0.58 0.56 0.57 OA6 Chemicals 0.27 0.34 0.23 0.33 1.00 0.87 0.84 0.79 Metals 0.22 0.31 0.13 0.63 0.33 1.00 0.95 0.71 Machinery 0.41 0.39 0.09 0.31 OA1 0.34 1.00 0.76 Electrical/Electronic 0.23 0.50 0.27 0.26 0.40 0.23 0.27 1.00 Notes: Factory data above diagonal, Worker data below diagonal Figures in bold are significant at 0.01. Data Source: CSO Government of India, ASI sampling frame, 1998-99 30 Table 4. Indices of co-clustering In selected Industry pairs Factories Workers National Total Total 1-0 factor Factories Moran's I Z of I Workers Moran's I Z of I MUMBAI Strong 1-0 links Metals and Machinery 34.95 1,661 0.082 1.496 65,864 -0.052 0.668 Metals and Electrical/Electronic 25.75 1,696 0.161 2.774 83,927 0.105 1.876 Weak 1-0 links Food/Beverages and Electrical/Electronic 0.07 1,076 0.184 3.145 75,863 0.294 4.936 Textiles and Metals 0.41 2,882 0.080 1.473 156,374 0.045 0.904 KOLKATA Strong 1-0 links Metals and Machinery 34.95 1,474 0.222 4.429 51,762 0.121 2.49 Metals and ElectricalElectronic 25.75 1,534 0.211 4.222 56,760 0.109 2.249 Weak 1-0 links Food/Beverages and ElectricalElectronic 0.07 866 0.178 3.579 27,893 0.185 3.730 Textiles and Metals 0.41 1,378 0.270 5.349 129,061 0.181 3.644 CHENNAI Strong 1-0 links Metals and Machinery 34.95 1,111 -0.079 1.060 47,848 0.029 0.673 Metals and Electrical/Electronic 25.75 1,054 0.149 2.738 50,094 -0.004 0.087 Weak 1-0 links Food/Beverages and ElectricalElectronic 0.07 525 0.126 2.332 29,477 0.041 0.873 Textiles and Metals 0.41 2,141 0.161 2.948 136,342 0.026 0.623 Data Source: CSO Government of India, ASI sampling frame, 1998-99 31 Table 5. Industry concentration In top districts Number of Share of Food/ Textiles Leather Printing/ Chemicals Metals Machinery Electrical/ units metropolitan Beverages (%) (%) Publishing (%) (%) (%) Electronics total(%) (%) Mumbal Factories Top 10 3,744 55.20 23.73 57.75 56.52 45.69 57.77 60.40 56.32 53.49 Next 10 1,447 21.34 25.18 22.69 19.13 16.78 21.73 20.69 23.99 17.33 Top 20 Total 5,191 76.54 48.91 80.43 75.65 62.47 79.49 81.10 80.31 70.82 Mumbal Workers Top 10 176,701 53.19 28.05 51.18 20.27 35.45 34.17 50.00 62.75 77.91 Next 10 86,411 26.01 15.70 32.58 51.16 23.27 29.80 24.15 25.16 11.85 Top 20 Total 263,112 79.21 43.75 83.76 71.44 58.72 63.97 74.15 87.91 89.76 Kolkata Factories Top 10 1,610 36.38 24.35 38.64 59.25 26.44 29.70 46.22 35.54 26.37 Next 10 787 17.79 25.13 13.05 21.57 16.78 18.08 13.72 18.88 22.41 Top 20 Total 2,397 54.17 49.48 51.70 80.82 43.22 47.78 59.93 54.42 48.78 Kolkata Workers Top 10 77,588 33.73 11.19 57.66 0.42 7.00 19.81 19.02 9.68 25.57 Next 10 49,117 21.35 30.47 23.14 42.00 15.09 13.98 23.82 13.33 13.40 Top 20 Total 126,705 55.09 41.66 80.80 42.42 22.09 33.79 42.84 23.01 38.97 Chennai Factories Top 10 1,872 42.53 33.74 27.25 53.20 28.36 40.62 51.22 69.95 66.02 Next 10 792 17.99 21.08 23.03 6.16 33.83 18.16 17.12 5.53 7.80 Top 20 Total 2,664 60.52 54.82 50.28 59.36 62.19 58.79 68.35 75.48 73.82 Chennal Workers Top 10 104,902 45.65 23.34 40.60 53.58 35.54 42.91 51.12 44.82 73.68 Next 10 45,364 19.74 29.32 20.44 14.09 14.56 15.17 16.37 38.04 11.66 Top 20 Total 150,266 65.39 52.66 61.04 67.66 50.10 58.08 67.49 82.85 85.34 Data Source: CSO Government of India, ASI sampling frame, 1998-99 32 Food F Tle Textile Factories Workers Factories Workers Leather lt rLeather P nnting/ Piinting/ Factories Workers Publishing Publishing Factories Workers Chemical Chemical Metal Metal Factories Workers FactorTes Workers Electrcal/ Electrical/ MFchanical Mehanical Electronic Electronic Factories Workers Factories Workers Figure 1. Industrial Clusters in Mumbai 33. Food Food Textile Textile Factories Workers Factones Workers Leather Leather Pnnting/ Pisting/ Factones ~~~Workers Publishing Pbihn Factones Workers Chemical Chemical Metal Metal Factones Workers Factories Workers Mechanical ancal lecincal/ Electrcal/ Factories Workers Electronic Electronic Factories Workers Figure 2. Industrial Clusters in Kolkata 34 Food/Beverage Food/Beverage Textile Textile Factores Workers Factores Workers Printing/ Prlnting/ Leather Leather Publishing Publishing Factones Worker's Factre okr Chemicals Chemicals Metal Metal Factores Workers Factories Workers Machinery Machinery Electrical/ Electrical/ Factones Workers Electronic El Workers Figure 3. Industrial Clusters in Chennai 35 Policy Research Working Paper Series Contact Title Author Date for paper WPS3056 Avoiding the Pitfalls in Taxing Patrick Honohan May 2003 A. Yaptenco Financial Intermediation 31823 WPS3057 Teaching Adults to Read Better and Helen Abadzi May 2003 H. Abadzi Faster: Results from an Experiment 80375 in Burkina Faso WPS3058 Working for God? Evaluating Service Ritva Reinikka May 2003 H. Sladovich Delivery of Religious Not-for-Profit Jakob Svensson 37698 Health Care Providers in Uganda WPS3059 Do Capital Flows Respond to Risk Cesar Calder6n May 2003 T. Tourougui and Return? Norman Loayza 87431 Luis Serv6n WPS3060 World Market Integration through the Rui Albuquerque May 2003 T. Tourougui Lens of Foreign Direct Investors Norman Loayza 87431 Luis Serven WPS3061 Reciprocity in Free Trade Caroline Freund May 2003 P. Flewitt Agreements 32724 WPS3062 Labor Effects of Adult Mortality in Kathleen Beegle May 2003 E. de Castro Tanzanian Households 89121 WPS3063 Trade Liberalization, Firm Paolo Epifani May 2003 P. Flewitt Performance, and Labor Market 32724 Outcomes in the Developing World: What Can We Learn from' Micro-Level Data? WPS3064 Trade Policy, Trade Volumes, and Ana M. Fernandes May 2003 A. Fernandes Plant-Level Productivity in Colombian 33983 Manufacturing Industries WPS3065 Ghost Doctors: Absenteeism in Nazmul Chaudhury May 2003 H. Sladovich Bangladeshi Health Facilities Jeffrey S. Hammer 37698 WPS3066 The Potential Demand for and Shreelata Rao Seshadri May 2003 H. Sladovich Strategic Use of an HIV-1 Vaccine P. Subramaniyam 37698 in Southern India Prabhat Jha WPS3067 The Mini-integrated Macroeconomic Pierre-Richard Agenor May 2003 M. Gosiengfiao Model for Poverty Analysis: A Framework 33363 for Analyzing the Unemployment and Poverty Effects of Fiscal and Labor Market Reforms WPS3068 Poverty and Economic Growth in Heba El-laithy June 2003 P. Sader Egypt, 1995-2000 Michael Lokshin 33902 Arup Banerji WPS3069 International Migration, Remittances, Richard H. Adams, Jr. June 2003 N. Obias and the Brain Drain: A Study of 24 31986 Labor-Exporting Countries * WPS3070 Are You Satisfied? Citizen Feedback Uwe Deichmann June 2003 Y. D'Souza and Delivery of Urban Services Somik V. Lall 31449 Policy Research Working Paper Series Contact Title Author Date for paper WPS3071 Survey Techniques to Measure and Ritva Reinikka June 2003 H. Sladovich Explain Corruption Jakob Svensson 37698 WPS3072 Diversity Matters: The Economic Somik V. Lall June 2003 V. Soukhanov Geography of Industry Location in Jun Koo 35721 India Sanjoy Chakravorty