Portal de Periódicos da CAPES

Detecting spatial economic clusters using kernel density and global and local Moran's I analysis in Ekurhuleni metropolitan municipality, South Africa

2022; Elsevier BV; Volume: 14; Issue: 2 Linguagem: Inglês

10.1111/rsp3.12526

1757-7802

Koech Cheruiyot,

Regional Economic and Spatial Analysis

Agglomeration economies are credited for providing the needed catalyst for economic growth and development. This paper uses approximately 14,000 firm-level data and employs several spatial data analysis approaches to examine evidence of types of spatial sectoral clusters and their footprints in Ekurhuleni Metropolitan Municipality, a major subregional economy in Gauteng metropolis, South Africa. The results of four selected industrial sectors show evidence of varying global and localized clustering. Localized clustering is statistically significant. This research suggests policies that ensure that regional economy's economic growth and development benefit from agglomeration economies. A las economías de aglomeración se les atribuye la función catalizadora necesaria para el crecimiento económico y el desarrollo. Este artículo utiliza aproximadamente 14.000 datos a nivel de empresa y emplea varios enfoques de análisis de datos espaciales para examinar las pruebas del tipo de agrupaciones sectoriales espaciales y su huella en el municipio metropolitano de Ekurhuleni, una importante economía subregional de la metrópolis de Gauteng y de Sudáfrica. Los resultados de cuatro sectores industriales seleccionados muestran evidencias de una agrupación global y localizada variable. Se estableció que la agrupación localizada también era estadísticamente significativa. Esta investigación sugiere políticas que garanticen que el crecimiento económico y el desarrollo de la economía regional se beneficien de las economías de aglomeración. 集積経済は、経済成長と発展に必要な触媒的役割を与えていると考えられている。本稿では、約14,000社レベルの企業データを使用し、いくつかの空間データ解析手法を用いて、ハウテン州の都市圏および南アフリカの主要な地方経済域であるエクルレニ都市圏における部門別の空間クラスターのタイプのエビデンス及びその足跡を検討する。選択した4つの産業部門の結果から、様々なグローバルおよび局地的なクラスター形成のエビデンスが示される。また、局地的なクラスター形成は統計的に有意であることも確認される。本研究から、経済成長と地域経済の発展が、集積経済から確実に利益を得られる政策が示唆される。 The study of the economics of agglomerations, either focusing on commercial/industrial districts within cities, industrial clusters at the regional level, or the imbalances between regions/countries, can be traced to Marshall's (1890) work relating to industrial districts in nineteenth-century England (Fujita & Thisse, 2002). Regardless of the geographical area under investigation, agglomeration economies—primarily in the form of knowledge spillovers among firms, labor market pooling, and sharing industry-specific non-traded inputs—provide the needed catalyst for economic growth and development. Witnessed as external economies, agglomeration economies comprise localization and urbanization economies that transmit positive externalities to either similar or dissimilar firms that geographically concentrate or colocate in particular areas. Localization economies accrue firms from similar industrial sectors (Aoyama et al., 2010; Hoover, 1948; Marshall, 1890). Urbanization economies—also known as Jacobian externalities, after Jacobs (1969)—refer to the advantage enjoyed by diverse firms when they colocate in a large urban area with large and heterogeneous markets. In the literature, several theoretical and empirical studies can be found focusing on developed and some developing countries. In South Africa, limited empirical studies have focused on the subnational regions—either provinces or cities (Fedderke & Wollnik, 2007; Krugell & Rankin, 2012; Naudé & Krugell, 2006; Pillay & Geyer, 2016; Pisa et al., 2015; Vom Hofe & Cheruiyot, 2018). The objective of this study was to extend the existing research by focusing on a large and detailed dataset (to the individual geocode level). The study achieved its objective by employing various spatial statistical analysis approaches to analyze the spatial location and economic sector data of about 14,000 firms obtained by triangulating various data sources in Ekurhuleni Metropolitan Municipality (EMM), a major subregional economy in Gauteng, South Africa. It answers the following questions: is there evidence of spatial business clusters; if so, what kind of spatial economic clusters are they, and what is the footprint of these spatial business clusters. Thus, this paper explores clustering using the number of firms per square kilometer, independent of firm size (Duranton & Overman, 2005; Pillay & Geyer, 2016). It suggests policies needed to grow and develop the regional economy emanating from agglomeration economies. The paper is structured as follows. After the introduction, the second section reviews the related literature, focusing on economics of agglomeration and measures of agglomeration economies. Section 3 sheds light on the economic significance of EMM to Gauteng City-Region and the country, while Section 4 focuses on the methods and data employed in this paper. Section 5 presents the results based on descriptive mapping, kernel density mapping, and global and local evidence of clustering. The last section concludes the paper. Fujita and Thisse (2002) note that economics of agglomerations' geographical focus is on exploring the formation of commercial districts within cities, industrial clusters at the regional level, and the existence of imbalances between regions. Firms—either from the same or different industrial sectors—geographically concentrate or colocate to take advantage of agglomeration economies, such as knowledge spillovers among firms, labor market pooling, and sharing industry-specific non-traded inputs (Marshall, 1890). The first two agglomeration economies are external to firms and are what Hoover (1948) called economies of localization and economies of urbanization, respectively. Hoover (1948) identified the third Marshall's (1890) agglomeration economy mentioned above as internal to firms and called it internal returns to scale. Internationally, there is a long list of theoretical and empirical studies on the agglomeration phenomenon; see Fujita and Thisse (2002) for a theoretical survey, and Duranton and Kerr (2015) and Rosenthal and Strange (2001) for empirical surveys. In South Africa, academic work on clusters or agglomeration is limited. Pisa et al. (2015) undertook cluster analysis in South Africa's North West Province, a rich platinum and gold mining region that is highly specialized and dependent on a few sectors. Employing structural path analysis and power-of-pull methods, they identified 10 industrial clusters that offer the greatest economy-wide benefits, while also creating opportunities for cross-sectoral collaboration. Vom Hofe and Cheruiyot (2018), while employing principal component analysis on Gauteng's social accounting matrix, showed evidence of a few but nonetheless critical masses of economic clusters in Gauteng's regional economy. Their analysis led to the identification of six distinctive industrial clusters: service and trade, food products, metal products, chemical products and petroleum, building and metal products, and light manufacturing products. Rogerson (1998), using University of South Africa's Bureau of Market Research data, found that Gauteng province dominated agglomeration activities in the high-technology clusters, with locations such as Johannesburg, Boksburg, and Kempton Park emerging as the largest foci of high-technology manufacturing (p. 889). Other studies have found evidence that different South African cities are agglomeration hot spots—places that offer urban diversity, industry specialization, dynamism and inclusivity, and opportunities for migrants as well as hosting growing industries, such as finance, business and consumer services, and high-level professional and technical occupations (Krugell & Rankin, 2012; Naudé & Krugell, 2006; Turok, 2011). As such, these cities (e.g., Johannesburg, Tshwane, and Cape Town) have experienced significant growth. All the aforementioned South African research focused on the subregional levels and were aspatial, meaning they did not spatially allocate the identified clusters or evidence of agglomeration economies in the respective study areas. One paper with a spatial focus, like the current one, is that of Pillay and Geyer (2016), who used aerial photography, zoning, and cadastral data as well as field surveys to show evidence of business clusters along one of the transport corridors in Gauteng. Pillay and Geyer (2016) employed a distributional directional analysis tool (part of ESRI's ArcGIS software's spatial statistical analysis; ESRI, 2022) to measure the geographic distribution of the data, thus producing a visual interpretation of how business areas along the M1–N3–N1 corridor between Johannesburg, Germiston, and Pretoria have densified between 2003 and 2012. Finally, a triangulation survey of the number of businesses revealed multiple and different business clustering across the A1–A15 business clustering zones (see Pillay and Geyer, 2016, pp. 349–352 for a complete description of their methodology and the different identified clusters). This paper complements existing research by analyzing spatial sectoral clustering in EMM. Beyond Pillay and Geyer's (2016) research, the present paper employs advanced spatial statistical techniques, including exploratory spatial data analysis (ESDA), kernel density analysis, global Moran's I, and Anselin's (1995) local indicators of spatial association (LISA) tests. These techniques were implemented in ArcGIS and GeoDa (Anselin et al., 2010). In doing so, the present paper not only identifies both global and local spatial sectoral clusters through spatial dependence but also their statistical significance. This was possible since point data (captured from individual firms' geocodes) used in the analysis were sufficiently large and detailed. Often spatial clustering or agglomeration is underestimated when data is only available for some defined discrete space that merely allows aspatial analysis (Guillain & Le Gallo, 2010). There has been a continuous search for better techniques for measuring the geographic concentration of economic activities or agglomeration (also known as cluster) over the years. Initially, aspatial techniques were employed, which essentially measured the geographic concentration of economic activities across some defined spatial scale and treated spatial units independently. Guillain and Le Gallo (2010) warn that such aspatial techniques, by ignoring spatial dependence across geographical units, potentially underestimate existing spatial agglomeration. They suggest that an appropriate empirical technique must capture two dimensions of agglomeration: "concentration in one spatial unit but also the spatial distribution of these units in the study area" (Guillain & Le Gallo, 2010, p. 3). Such techniques must measure the global and local spatial patterns of agglomeration, while allowing for spatial dependence across geographical units. Several global indices are available for measuring the spatial concentration of activities. These include the spatial concentration ratio, the spatial Hirschmann–Herfindhal index, the locational Gini coefficients (Krugman, 1991), and the Ellison and Glaeser (1997) concentration index. This paper does not attempt to review all the indices used to measure the spatial distribution of economic activities due to limited space. For a complete review of the various indices for measuring the spatial distribution of economic activities across the world, see Combes and Overman (2004), Fujita, Henderson, and Mori (2004), Guillain and Le Gallo (2010, pp. 5–7), Henderson and Mori (2004), Fujita, Holmes, and Stevens (2004); for details of the Ellison and Glaeser concentration index, see Lafourcade and Mion (2007, pp. 3–5). As suggested by Duranton and Overman (2005, p. 1079), an ideal test of localization should rely on a measure which "(i) is comparable across industries; (ii) controls for the overall agglomeration of manufacturing; (iii) controls for industrial concentration; (iv) is unbiased with respect to scale and aggregation … (v) give[s] an indication of the significance of the results." This paper focuses on global Moran's I and LISA (Anselin, 1995), which—while incorporating spatial distribution and dependence in the data within defined spatial units—allow for accurate spatial identification (i.e., where firms are located) and the statistical significance testing of agglomerations (Duranton & Overman, 2005; Guillain & Le Gallo, 2010; O'Donoghue & Gleave, 2004). As one of the three metropolitan municipalities located in South Africa's largest agglomeration (Gauteng), EMM covers 197,500 hectares, partitioned into six administrative (also economic) regions, namely, Regions A–F. With each region consisting of several industrial areas, Region A has the largest amount of industrial space (3,370.97 hectares), while Region C has the least (167.17 hectares). This is not surprising considering that Region A is comprises O.R. Tambo International Airport as well as industrial and logistical areas, such as Jet Park, Boksburg, and Germiston (see Figure 1). EMM is the industrial backbone of Gauteng province, and Gauteng province itself is South Africa's economic heartland, attributable for about a third of national gross domestic product (EasyData, 2021). Table 1 shows the size of EMM's contribution to the provincial and national economies. The table shows that EMM, as a key player in the province as well as nationally, contributes close to a quarter (23.7%) and close to a tenth (8.2%) of provincial and national gross value added (GVA) output, respectively. EMM's contributes more to the provincial and national economy in the secondary sector, followed by tertiary and primary sectors. In terms of employment and compensation, EMM's contributions mirror its contribution to GVA. For instance, it contributes 25.1% to the provincial total employment and 8.3% to the national total employment. Its share of provincial and national total real compensation is 22.9% and 9%, respectively. Table 2 shows the contribution of each of EMM's economic regions to the metro's economy. As expected, Region A—comprising Kempton Park, Germiston, and Boksburg—contributes more to the metro economy than the other regions (see Table 3 as well). Assuming that Germiston and Edenvale contribute equally, if we split their contributions, Region A's contribution—consisting of Kempton Park, Germiston, and Boksburg—increases to 42.7% of EMM's economy. Figure 2 shows different sectors' GVA contributions to EMM's economy. The table shows that manufacturing, wholesale and retail trade, and finance and business support services are important contributors to economic growth in the municipality. The mining and agricultural sectors contribute the least to EMM's economy. With the highest annual average growth rates (2010–2020), agriculture (9%), trade (3.1%), finance (7%), electricity, gas and water (5%), and community services (6.7%) are expected to drive EMM's economy going forward. The overall GVA growth rate averaged 3.2% between 2010 and 2020. Source: EasyData (2021) The data used in this paper comprise the number of firms obtained from EMM's geographical information system (GIS) corporate zoning and billing register. The billing register metafile consists of several zonal classifications with almost 600,000 records. After carefully cleaning the records of duplications, vacant lots and so on, and limiting the records to industrial and business classifications, a total of close to 14,000 firms was obtained. The firms' names have been anonymized in the research. Using the economic sectors that were part of the metafile, respective standard industrial classification (SIC) codes were assigned. At this point, the list contained key details regarding firms' names, physical addresses, suburbs, regions, economic sectors, and SIC codes. Subsequent work will focus on adding employment details to the database. Finally, using the above details, all firms were assigned geocodes in the ESRI website using an address geocoding tool (ESRI, 2022). Given that the point data collected had no measurable attribute for, for instance, the number of employees, the analysis had to proxy firm concentration by creating a one-kilometer-squared fishnet for the study area and merging it to point data on geocoded firms' locations. By doing this, it was possible to find firm density—the number of firms in each square kilometer. Since the study area is under the management of one municipal government, zoning and planning restrictions are ubiquitous; thus, we expected the firms' behavior vis-à-vis regulatory frameworks to be uniform. It was necessary to proceed using the number of firms per square kilometer because of the lack of firm data on measurable indicators, such as the number of employees, as is commonly used. We tested for scale and aggregation issues (using a one square kilometer grid and 0.5-square kilometer grid) and found broad consistency in the results (Duranton & Overman, 2005). The choice of one square kilometer was more appealing as it is commonly used and the results are easy to interpret. The resultant shape file with a density of firms was analyzed in ArcGIS (ESRI, 2022) and GeoDa (Anselin et al., 2006). These softwares' mapping cluster tools allow visualization of both global and local patterns of cluster locations and extent (ESRI, 2021a; Guillain & Le Gallo, 2010, p. 965). The analysis proceeded in three steps. In the first step, we used kernel density analysis in ArcGIS to explore hot and cold spots. Second, we ran global Moran's I to test for overall clustering, and third, we employed LISA to test for localized clustering. The latter also allowed for statistical significance testing. We strengthened the results of localized clustering with location analysis (using location quotient), albeit at a higher level of aggregation, based on the economic regions defined by EasyData (2021). Preliminary analysis focused on the distribution of firms based on their primary and secondary SIC codes. With most SIC codes represented, a few SIC codes stand out. For instance, of the 13,973 geocoded firms (see Table 3), SIC3 (manufacturing), SIC8 (finance and business services), and SIC61 (wholesale and retail trade) accounted for 37.6%, 24.4%, and 25.6%, respectively. At a far distance, transport and storage (SIC71) followed with 4.8%. The secondary SIC code in SIC3, SIC35 (the manufacture of metals, machinery, and equipment), singlehandedly accounted for 14.4% of the 13,973 geocoded firms. The spatial statistical analysis was based on SICs 3, 35, 8, and 71. Figure 3 shows the distribution of selected firms in EMM. It is observable that manufacturing is predominantly located in Region A, north of Region F, south of Region B, and west and south to southeast of Region D. Some manufacturing is also visible in the southeast of Region E. The further away from Region A, the less manufacturing activity. Overall, these results support the industrial role of the municipality both provincially (in Gauteng) and nationally. Furthermore, Figure 3b shows that the distributional pattern of metal, machinery, and equipment subsector firms is similar to Figure 3a (showing the distribution of manufacturing), except less dense. These results concur with the results displayed in Tables 2 and 3, showing the dominance of Region A in the regional economy. These results also confirm EMM as an "African workshop" (Corporate Governance & Traditional Affairs [COGTA], 2020). Moreover, Figure 3 shows that Region A has the greatest number of finance and business services firms, followed by Regions F and D. Away from these areas, the number of firms declines towards the outer parts of EMM. These results indicate that most of the finance and business firms are concentrated in the metro's core. These results are supported by sectoral analysis, as captured in Tables 2 and 3, which shows that Region A contributes 42.7% of EMM's regional economy. These results suggest, as expected, that finance and business-related firms, offering credit, liquidity, insurance, and business advisory services, will be in close proximity to other firms in the regional economy. The high number of finance and business-related firms are also likely to serve international clientele who visit or transit through the O.R. Tambo International Airport. The finance and business-related sector has witnessed the highest growth (2.95%) in EMM (COGTA, 2020). As expected, EMM is a major transport and logistics hub for Africa since it is where O.R. Tambo International Airport is located. As Figure 3 shows, the distribution of transport and storage firms is more concentrated around the airports, namely, O.R. Tambo International Airport, Rand Airport, and Brakpan and Springs Airfields and their surroundings. In recognition of its important role as a transport and logistics hub, O.R. Tambo International Airport has, as of 2019, been recognized four times as the leading handler of cargo volumes on the African continent (Airports Company South Africa, 2019). Kernel density mapping produced using the kernel density tool in ArcGIS (ESRI, 2022) aids in smoothing out the information represented by a collection of points in a way that is more visually pleasing and understandable. The kernel density tool calculates a magnitude per unit area from point or polyline features using a kernel function to fit a smoothly tapered surface to each point or polyline (ESRI, 2021b). This is necessary when the points cover large areas of the map and the mere representation of data as "blobs of points" may impair visual understanding of the phenomenon being investigated (ESRI, 2021c). Figure 4 shows that most of the manufacturing firms are concentrated evenly across Region A; in the south of Region B (near Kempton Park); and in several towns, including Boksburg, Alrode, Alberton, and Vosloorous. The concentration of metal, machinery, and equipment firms—while it follows the same pattern as aggregated manufacturing firms—has some particularities. For instance, while aggregated manufacturing firms show a dense concentration around Springs Airfield, the concentration of metal, machinery, and equipment firms is barely visible around the same area. The dense concentration of aggregated manufacturing firms in the north of Region B is similarly barely visible regarding metal, machinery, and equipment firms. Finance and business services firms are predominantly in Kempton Park and located along EMM's boundary with the City of Johannesburg (EMM's western boundary). The cores of the various regions seem to have a concentrated number of finance and business services firms. Transport and storage firms are predominantly in Kempton Park, Boksburg, near Rand Airport, and Brakpan and Springfield Airfields. Figures 5a–5d show the global spatial autocorrelation of the selected industrial sectors. As inferential statistics, the values of Moran's I cannot be interpreted directly; rather, they can only be interpreted within the context of the null hypothesis of no spatial autocorrelation. The results show that all the Moran's I scatter plots show statistically significant global spatial autocorrelation, with manufacturing (SIC3) more autocorrelated (Moran's I = 0.196802), followed by the manufacture of metal, machinery and equipment (SIC35, Moran's I = 0.177295), finance and business services (SIC8, Moran's I = 0.157276), and transport and storage (SIC71, Moran's I = 0.0925195). While the results in this paper are informative as they show evidence of clustering of similar values, they do not tell us where they are located and whether the identified similar values are high or low. To uncover this, we turned to an analysis of local spatial autocorrelation in the next section. Figures 6-9 show local spatial associations of the selected industrial sectors. These figures show local spatial association, where different pockets of associations show either spatial clustering (i.e., a high number of firms surrounded by a high number of firms and a low number of firms surrounded by a low number of firms) or spatial outliers (i.e., a high number of firms surrounded by a low number of firms, and a low number of firms surrounded by high). The statistical significance of these spatial associations are shown in Figures 6-9, respectively. Figure 6a shows that spatial clustering of a higher number of manufacturing firms is found in several locations, including areas around Edenvale, Kempton Park, Alrode, and Alberton. Spatial outliers in Figure 6a border the spatial clusters of the higher number of manufacturing firms. The closeness of spatial clusters and spatial outliers imply that the distribution of manufacturing firms (SIC3) depicts spikes and the subsequent decline of the number of manufacturing firms as one moves away from the evident spatial clusters. Figure 6b shows that the preceding spatial associations are statistically significant, with p-values ranging between 0.0001 and 0.05. The local evidence of clustering in this paper is sufficiently supported by the calculation of simple location quotients to reflect a measure of relative concentration. In this paper, location quotient is used to quantify the concentration of a particular industry in a given region (i.e., EMM's economic Regions A–F) compared to the Gauteng province and EMM. In this manner, location quotient captures localization economies (part of agglomeration economies, the other being urbanization economies) associated with local specialization. In terms of broad manufacturing (SIC 3), all EMM's economic regions have location quotients exceeding one, except two (Germiston/Edenvale and Springs). The same picture is reflected in the manufacture of metal, machinery, and equipment (SIC35). The major difference is the higher location quotient of 1.59 (Benoni) for the manufacture of metal, machinery, and equipment, compared to the highest location quotient of 1.38 (Kempton Park) for broad manufacturing. At the metro level, Kempton Park (1.19) and Benoni (1.08) have location quotients exceeding one, meaning that there is a higher concentration of manufacturing firms (SIC3) in Kempton Park compared to the other regions. Still referring to the manufacture of metal, machinery, and equipment (SIC35), Benoni (1.35) and Kempton Park (1.16) are the only regions with location quotients exceeding one. These results coincide with Rogerson's (1998), who used both the number of establishments and estimated total employment and found that a higher cluster of high-technology production, including the manufacture of electrical and industrial machinery, was found in Boksburg, Kempton Park, Germiston (part of EMM) compared to the rest of the country. Figure 7 shows the spatial clustering of the number of firms manufacturing metals, machinery, and equipment (SIC35). Compared to Figure 6a, Figure 7a shows a close similarity of spatial association. A notable difference in Figure 7a compared to Figure 6a is the location of spatial outliers (high–low) in the northern parts of Region B and parts of Regions C, D, and E. Figure 7b shows that the preceding spatial associations are statistically significant, with p-values ranging between 0.0001 and 0.05. Figure 8a shows the spatial association in the finance and business services sector (SIC8). It shows a more widespread distribution of firms compared to Figures 6a and 7a. While locations running south from Kempton Park to Alrode on the western side of the metro show spatial clustering, new areas—such as Geluksdal in Region E, and areas near Springs and Brakpan Airfields (the latter two areas are in Region D)—show spatial clustering as well. The statistical significance of these spatial associations (p-values) range from 0.0001 and 0.05 (Figure 8b). These results seem to support the results of the location analysis. The finance and business services sector in EMM is less concentrated at the provincial level: only one region (Alberton) has a location quotient of 1.07. A regional-level (in-metro) comparison shows that in terms of finance and business services, the four regions with location quotients exceeding one are Alberton, Germiston/Edenvale, Nigel, and Brakpan. These regions have a higher concentration of finance and business services firms than the other regions at the metro level. Figures 9a and b show the spatial association and respective statistical significance of spatial associations for transport and storage firms. Figure 9a shows a limited spatial clustering of a high number of transport and storage firms in Kempton Park, Boksburg, and near Brakpan Airfield. A spread of spatial outliers (high–low) is visible in the outer parts of Regions B, D, and E. These results are broadly supported by the location quotient analysis, where five out of eight of EMM's economic regions have location quotients exceeding one. These include Kempton Park (in Region A), with the highest location quotient of 2.11, followed at a distance by Region C's Benoni (1.33), Region D's Brakpan (1.04), Region A's Boksburg (1.04), and Germiston/Edenvale (1.03). These location quotient results show the dominance of Region A's hold on transport and storage size of employment in the province. Within EMM, Kempton Park (part of Region A) is the only economic region with a location quotient exceeding one; thus, it has the greatest concentration of transport and storage firms in the metro. One of the key motivations for the work reported in this paper was to learn about the nature of clustering and potentially the agglomerative forces that operate in EMM, particularly to assist policymakers considering proposals to encourage further agglomeration economies or to develop new agglomeration sites. This would allow firms to benefit from agglomeration economies by sharing knowledge, labor, and other industry-specific non-traded inputs. The paper focused on the following questions: is there evidence of spatial sectoral clusters; if present, what kind of spatial economic clusters are they, and what is the footprint of these spatial business clusters. This paper's results (based on geocoded point firm data) have shown the presence of a statistically significant concentration of firms in EMM. The three techniques employed in the paper were complementary and assisted in adequately addressing this paper's research questions. Descriptive analysis in the form of the regional distribution of firm count (Table 3) and choropleth mapping shows that Region A (comprising Kempton Park, Germiston, and Boksburg) dominates EMM's industrial economy. Here, choropleth maps provide an easy way to visualize firms' distribution, and thus the level of variability across the different regions of EMM. Kernel density mapping removed "blobs of points" representation in firm count, and thus was complementary in improving visualization of where the different selected firms in EMM are concentrated. Statistical evidence of overall and local spatial clustering was evident from global and local spatial autocorrelations. As expected, broad manufacturing (SIC3), with a higher global Moran's I of 0.1977, was more clustered, followed by the manufacture of metal, machinery and equipment (SIC35), finance and business services (SIC8), and transport and storage (SIC71). Regarding inferential statistics, the values of Moran's I cannot be interpreted directly; rather, they can only be interpreted within the context of the null hypothesis of no spatial autocorrelation or that the distribution of firm count is randomly disbursed. All the Moran's I indices were statistically significant, supporting the alternative hypothesis that the number of firms are more spatially clustered than would be expected by chance alone. The above results were further complemented by employing Anselin's LISA maps to identify specific locations of clustering of similar high or low values (Anselin, 1995). Anselin's LISA maps (i.e., cluster and significance maps) showed (localized) footprints and statistical significance of the identified spatial clusters. As observed in Figures 6a/b–9a/b, the localized clustering of the selected industrial firms varies. Nonetheless, Region A and its surroundings dominate other regions. The footprint of local clustering was partly supported by calculated location quotients of unemployment data, newly calculated by EasyData (2021) as well as Rogerson (1998). Pilot results for a survey support the results in this paper. In the pilot survey, firms were asked whether there were clustering or agglomeration advantages where they are located. The firms were presented with several proxies for clustering or agglomeration advantages as described in the existing literature. With ranking scores calculated as the average ranking for each answer choice by the various firms, it was possible to identify which of the answer choices was preferred overall. The results showed that the most preferred choices, given their higher scores, were access to markets (4.63), quality infrastructure (electricity, water, public transport, etc.) (4.30), and proximity to suppliers (4.25). Others had lower scores as follows: availability of land to expand (3.29) and inexpensive rent (3.29). Given that the present paper is a pioneering work, employing spatial statistical analysis in EMM and South Africa, this paper has not benefitted from reference to existing literature on the subject. These results are useful in helping EMM know the spatial footprint of different cluster types so as to devise policies to encourage further agglomeration in areas where (statistically significant) agglomeration exists or to encourage potential agglomeration economies in other areas. Policies that enhance existing clustering, and potentially agglomeration economies, or develop new agglomeration economies could either be cluster-specific or cluster-informed strategies. The former is important in encouraging the emergence or development of a distinct cluster, while the latter is important in improving the implementation of individual (or isolated) development initiatives (Feser, 1998). From the literature there is a plethora of cluster policies that EMM may implement. These include provision of general business assistance, network brokering, technology transfer, information provision, training opportunities, and hard (e.g., roads) and social infrastructure (e.g., employee transport) subsidies. However, since firms and the clusters they belong to vary significantly, as evident in this paper, EMM should tailor such cluster policies depending on need assessment. It is important that this is based on a detailed investigation of each of the needs of the specific clusters identified in this paper. Such detailed study or studies can mirror the pilot study described above, where different select firms expressed various preferences as to what they perceive as the cluster or agglomerations economies they witness and enjoy. This work had the following limitations. It used the number of firms per square kilometer obtained from the use of the fishnet tool in ArcMap. This was necessitated by a lack of firm data on measurable indicators, such as the number of employees, as is commonly used. The use of location quotient calculations to reflect and further support relative firms' concentration was limited since the available data was at a lower resolution (i.e., main places across the country, including aggregated economic regions in EMM). As such, there is a need to develop firm-level data to include variables such as the number of employees (possibly from the Department of Labour's unemployed insurance fund data) in future work. These high-resolution data might offer more nuances that the present paper has not shown. The use of built area as contained in erfs' (also plot's) description in valuation roll could also easily be used as a proxy for density as suggested by one of the reviewers. Other proxies, such the amount of energy used and nighttime light data, have been proposed as well to capture the density of firms' economic activities This will foster a deeper understanding of spatial clustering and agglomeration in EMM and beyond. There is also a need to explore the relevance of economic corridor development policies that are being advanced by various spheres of government given the evidence put forward by this research. All these suggestions are fertile fodder for future research.