Spatial Distribution of Employment in Hermosillo,1999

Abstract

While the suburbanisation process has been well documented in some large cities of several developed countries, much less attention has been devoted to the case of small and middle-sized cities in developing countries. This article focuses on an exploratory spatial data analysis to investigate the location of the central business district (CBD) and other employment centres in Hermosillo, Mexico. The results reveal the significant presence of spatial dependence and spatial heterogeneity, although their extent varies with the sector under study. These spatial effects take the form of a persistent cluster of high values of employment around the historical district of the city shaping a huge CBD, although a sub-centre of high values emerges to the south and to the north-west of the CBD in 2004. Overall, Hermosillo is still characterised by a traditional monocentric model, but the role of its CBD has changed.

1. Introduction

In recent decades, one of the most documented phenomena in the urban structure of many developed countries has been the process of suburbanisation of both the population and economic activities. This phenomenon has had an impact on the traditional monocentric urban structure according to which cities are organised around a central business district (CBD) and employment density gradually decreases with distance from it (von Thünen, 1826/1966; Alonso, 1964; Muth, 1969; and Mills, 1972). Nowadays, cities are increasingly experiencing a polycentric structure. As a consequence, the CBD counts for smaller portions of employment than it did in the past (Griffith, 1981; Griffith and Wong, 2007).

In developed metropolitan areas, the CBD has been able to maintain its traditional economic role and importance (see, among others, the work of Shearmur and Coffey, 2002, and Coffey and Shearmur, 2002, for the Montreal metropolitan region, and Guillain et al., 2006, for the Paris region). In other studies (see as an example, McMillen and McDonald, 1998, for metropolitan Chicago), growth is now shared between the CBD and suburban agglomerations, while in other cases the CBD is losing ground to edge cities (see Lang, 2003, who has conceptualised this generalised dispersion based on 13 US metropolitan areas, as well as Gordon and Richardson, 1996, for the case of Los Angeles). However, the degree to which agglomeration economies in sub-centres are great enough to attract employment is still an open question (Gordon and Richardson, 1996).

Empirical work on medium-sized cities is scarce. A notable exception is the work of Baumont et al. (2004) who focused on Dijon, France, and found emergent sub-centres outside the CBD, although they did not have a significant impact on employment distribution. However, to our knowledge, no prior contribution has focused on the case of a medium-sized city located in a developing country. The goal of this paper is therefore to fill this gap and provide evidence for comparison. We investigate whether Hermosillo, the largest city in the northern Mexican state of Sonora, is experiencing a trend towards employment decentralisation or whether the distribution of employment still follows the traditional monocentric hypothesis.

Officially established in 1741, Hermosillo is a middle-sized city according to Mexican standards (between 500 000 and 1 million inhabitants). The current urban plan (2006–09), elaborated by the city’s Planning Institute (IMPLAN, 2006), claims that Hermosillo had had, until the late 1970s, a monocentric structure. This is because most of its employment density is registered close to the CBD, the oldest part of the city that still holds the commercial centre, the civic centre, the government centre and the university centre based on the University of Sonora. However, IMPLAN also affirms that a form of polycentrism has characterised the city over the past three decades. The veracity of this finding is questionable since the Planning Institute does not provide the methodology it uses to reach its conclusion. In addition, it does not define the specific boundaries of the CBD, nor dos it pay any attention to the potential presence of spatial dependence across observations.

A polycentric form implies the presence of agglomeration economies outside the traditional CBD and, as consequence of it, should be reflected in the spatial patterns associated to the distribution of the employment data. For that reason, our second contribution consists in identifying the employment centre and sub-centres based on the tools of exploratory spatial data analysis (ESDA). The first spatial effect that this technique allows us to uncover and measure is spatial heterogeneity which comes from the fact that some geographical clusters of high or low employment densities may be present in the city because of differences in the quality of local amenities, local labour or real estate markets. Spatial autocorrelation, the second spatial effect, reflects the fact that nearby observations tend to display similar characteristics (Anselin, 1995). Moreover, these provide the necessary statistical tests to indicate whether global and local spatial associations are significant. In the present case study, another advantage of ESDA lies in its capacity to identify the location and extent of employment centres without defining a priori and arbitrary employment cut-offs (Guillain et al., 2006).

In order to get more insights into the recent evolution of the spatial distribution of employment density across the 364 districts that compose Hermosillo, this paper will be organised as follows. Section 2 provides a review of the theoretical literature and related empirical studies that highlight how employment sub-centres emerge as well as different procedures to identify them. Section 3 describes the study area and the data. Section 4 uses an exploratory spatial data analysis where we first describe the spatial weight matrix and then perform the appropriate measurements of both global and local spatial autocorrelations. Finally, the conclusion summarises our results and points out similarities or differences with other medium-sized cities.

2. Spatial Distribution of Employment and Sub-centre Identification

The spatial concentration of economic activities and jobs in the CBD is explained by the history of the city (in the case of Hermosillo, the historical CBD is also the current one) and the persistent presence of agglomeration economies (see Fujita, 1988; Parr, 2002). On the other hand, the CBD may become less attractive when increasing agglomeration leads to higher land prices and wages, and creates congestion problems. As a result, agglomeration can occur in other areas because of economies of scales due to information spillovers, better accessibility to local inputs and a specialised local labour pool (Parr, 2002; Coffey and Shearmur, 2002). In addition, an employment sub-centre can emerge because of improved infrastructures that reduce transport costs. Following Redfearn et al. (2008), the emergence and growth of employment centres is also explained by exogenous factors such as planning decisions by the local government or the decision of large firms to locate outside the city’s core.

On the other hand, according to Gary (1990), the urban employment structure can be classified into two main categories: locally centralised employment and dispersed employment. In the first structure, firms are clustered in the CBD to exploit agglomeration economies, while in the second structure employment is clustered in some sub-centres or along major transport corridors. With regard to the latter structure, the urban literature discusses two forms of employment decentralisation: the edge city (also called the ‘suburban downtown’ phenomenon) and the scatteration process (Shearmur et al., 2007). Both exhibit a flat employment density gradient outside the CBD and a lack of spatial urban pattern (absence of or little spatial dependence with high agglomeration diseconomies pushing towards employment decentralisation). However, the ‘scatteration’ process is a generalised dispersion of employment at relatively low densities, rather than the dispersed agglomeration structure proposed by Gary (1990).

Previous contributions define an employment centre as a cluster of activity which must have: a significantly larger employment density than nearby locations; and, a significant effect on the overall employment density function (McMillen and McDonald, 1998; McMillen, 2001a). Hence, identifying a single employment centre is trivial (the zone with the highest employment density), while identifying employment sub-centres is more challenging. However, the literature offers different options in order to detect both. In one of the earliest works on centre identification, Giuliano and Small (1991) identified an employment centre as a cluster of contiguous zones for which total employment exceeds a predetermined cut-off level (10 jobs per acre and 10 000 jobs for its adjacent zone). Later, variations in the extent of the cut-off were used (see McMillen and McDonald, 1998; Giuliano et al., 2007), but without any accurate criterion such as the size of the urban area under study or any knowledge of local conditions that would help to establish more appropriate cut-offs.

Another set of studies uses non-parametric procedures to identify potential employment sub-centres. They use locally weighted regressions (LWR) (McMillen and McDonald, 1997), a two-stage approach combining LWR and a semi-parametric procedure (McMillen, 2001a), or even a combination of McMillen (2001a) and Giuliano and Small’s (1991) methods (McMillen and Smith, 2003). These procedures have been applied to a variety of large American cities and some empirical regularities are evident: large cities have more sub-centres than smaller cities, and sub-centres tend to develop near freeway intersections and in old satellite suburbs (McMillen and Smith, 2003).

More recently, a set of studies has relied on recent advances in the fields of spatial statistics and spatial econometrics to account formally for spatial effects in the identification of employment sub-centres. Baumont et al. (2004) focus on the case of the city of Dijon, France, and Guillain et al. (2006) on the Ile-de-France region. Both studies use local indicators of spatial associations (LISA) to identify potential sub-centres. This is the methodological approach we decide to rely on in our work because we feel it gives us more flexibility than previous procedures.

3. Study Area and Data

Hermosillo is both the largest and the capital city of the North-western Mexican state of Sonora. It is a middle-sized city located 271 kilometres south of the US border (see Figure 1) and was home to 641 791 inhabitants in 2005 (26.8 per cent of the state’s population) (see Table 1). In 2005, the city spread over an area of 15 480 hectares divided in 364 areas or AGEBs (área geoestadística básica), the smallest spatial unit in the census. Because the population growth has outpaced the city’s sprawl, Hermosillo’s density has actually increased from 40.1 to 41.4 inhabitants per hectare over our study period (see Table 1).

Figure 1.

Location of the City of Hermosillo, Mexico.

Table 1.

Physical, social and economic characteristics of the city of Hermosillo

Variables	1999/2000	2004/05	Growth (percentage)
Surface characteristics ^a
Areas (number of AGEBs)	254	364	43.3
City size (hectares)	13 619.30	15 511.70	13.9
Demographic characteristics ^a
Population^a	545 928	641 791	17.6
Households	133 283	168 204	26.2
Inhabitants per house	4.1	3.8	−6.8
Population density (hectares)	40.1	41.4	3.2
Economic characteristics ^b
Number of firms	16 538	17 082	3.3
Firm size (average)	6.9	6.4	−6.9
Total employment^c	113 956	109 628	−3.8
Forestry, fishing, hunting and agriculture support	29	213	634.5
Mining and oil extraction	136	365	168.4
Manufacturing^d	30 624	23 244	−24.1
Water and electricity production	756	884	16.9
Retail and wholesale	32 366	42 957	32.7
Services^e	50 045	41 965	−16.1
Total employment density (jobs per hectare)	8.4	7.1	−15.5

Data are based on population census and population survey for 2000 and 2005 respectively.

The information corresponds to the 1999 and 2004 Economic Census elaborated by INEGI.

The information is disaggregated by sector (two digits) according to the North America Industrial Classification System (NAICS).

As a consequence of a confidentiality agreement, the data exclude employment in construction.

Includes professional services, but excludes services related to construction, transport and warehousing, finance, insurance and real estate, as well as government services.

Source: based on data from INEGI.

The population and the size of the city grew in several steps. During the 1980s, the size of the city grew by 98.3 per cent and then by 61.0 per cent over the next decade. It was the most important growth since the 1950s when the city tripled its size because of increasing agricultural activity in the coastal valley (IMPLAN, 2006). IMPLAN’s projections indicate that the city will experience a 60 per cent increase and an 84 per cent increase in its area and population respectively over 2000–30. Local authorities have understood the consequences in terms of urban development and planning, creating in 2000 the Planning Institute of Hermosillo whose goal is to provide expertise on urban planning to local decision-makers. It marks a significant change compared with previous practices when the lack of planning led to an anarchic and explosive city growth.

Hermosillo represents roughly one-third of the state’s total employment, value added and number of firms. In Mexico, the employment data can come from the population census (where each respondent indicates his/her place of work) or from the economic census (where each firm reports where the employees work). In this study, we use the latter which come from the Mexican Bureau of Statistics (Instituto Nacional de Estadística, Geografía e Informática—INEGI) and have been collected every five years since 1980. However, only the 1999 and 2004 results are available at the AGEB level in a harmonised dataset.¹

Following the implementation of the North American Free Trade Agreement in 1994, the northern cities of the country have experienced an increase in foreign investments, in employment as well as the consolidation of an industrial strategy based on maquiladora (Rodríguez, 2003).² According to the Economic Census, 109 628 people worked in the city in 2004 (see Table 1), they specialised mostly in services and in retail and wholesale. Recently, the share of manufacturing (mostly maquiladora) in Hermosillo’s economy has decreased. Hermosillo’s economy is closely tied to that of the US and the latter has experienced a 24.1 per cent fall in its manufacturing production in 2000/01 in conjunction with a decrease in services (-16.1 per cent). In addition, in Hermosillo, professional services and new national and foreign investments in retail and wholesale activities have surged (Lara et al., 2007). However, it is the primary sector (forestry, fishing, hunting and agriculture support) which experienced the highest increase in the number of workers over 1999–2004.

Our variable of interest to analyse the employment distribution across the urban areas of Hermosillo is gross employment density (the average number of jobs per unit of area). It would be tempting to use the employment-to-population ratio as Guillain et al. (2006) did, but we cannot. Indeed, employment and population data come from two different datasets based on different collection methodologies and different years. Other indicators, such as net employment density (employment in sector i divided by land used by sector i), cannot be used either because the information needed is not available. Because so few people live in the historical CBD, we feel that net and gross employment densities are very similar there. However, this is not the case for the peripheral areas where people live, where gross employment density would be lower than net employment density.

4. Identification of Employment Sub-centres by ESDA

Following the contributions of Baumont et al. (2004) and Guillain et al. (2006), we also use the area (AGEB) with the highest density of workers for our definition of the CBD. In the case of Hermosillo, as in Guillain et al. (2006), the CBD is also the historical centre. In 1999, the CBD registered 116.5 employees per hectare, but this decreased to 101.7 in 2004 (see Figure 2). This fall in employment density is explained, partially, by the increase in unemployment in the city as a whole as well as by an increase in employment density around the CBD.

Figure 2.

Spatial distribution of total employment in Hermosillo, 1999 (left) and 2004 (right).

For exploratory purposes, we choose 25 employees per hectare (approximately 10 jobs per acre suggested by Giuliano and Small, 1991) as a cut-off to define a sub-centre. From Figure 2, we note that the sub-centres are located along the major streets and intersections of the city, and some of them are located outside the CBD. Also, we can see how, in a developed and bigger area such as Ile-de-France region (Guillain et al., 2006), high employment areas follow the main highways or streets. Based on this, in 2004, just 26 AGEBs were identified as potential sub-centres, while in 1999 we found 29 areas. It appears that these sub-centres lost their employees whereas the centre of the city still experienced a high density of workers (see Figure 2).

However, it is difficult to determine whether nearby AGEBs are sub-centres or if they are part of a sub-centre’s adjacent area. In fact, we cannot assume the cut-offs established for the Los Angeles case (Giuliano and Small, 1991), because Los Angeles has a unique polycentrism that Hermosillo does not have. Therefore, we define Hermosillo’s CBD boundaries, analyse if their fringe has changed over time and test for the presence of spatial autocorrelation by means of exploratory spatial data analysis (ESDA). It is a collection of techniques that describe and visualise spatial distributions, discover patterns of spatial association, clusters or hot spots, and suggest spatial regimes (Anselin et al., 2007).

4.1 Spatial Weight Matrix

The previous section provides us with a description of the distribution of employment across AGEBs and its evolution over time, but it does not account for the eventual presence of spatial effects that several previous studies have highlighted at the urban level (McMillen, 2001a, 2004; Baumont et al., 2004; Guillain et al., 2006; Guillain and Le Gallo, 2009). In order to investigate both spatial autocorrelation and spatial heterogeneity, the starting-point consists of defining a weight matrix (W) to define the spatial connectivity between our observations. In this matrix, each observation is connected to a set of neighbouring observations according to a spatial pattern defined exogenously (Baumont et al., 2004). As usual in the spatial statistics literature, the diagonal elements of the weight matrix are set to zero whereas the off-diagonal elements indicate the way locality i is spatially connected to locality j (Cliff and Ord, 1981; Anselin, 1995; Anselin et al., 2007). These elements are non-stochastic, non-negative and finite. In order to normalise the outside influence upon each unit, the weight matrix is standardised such that the elements of a row sum up to one.

While there is very little formal guidance on the choice of the ‘correct’ spatial weights in any given application, we decided to adopt a k-nearest neighbour’s weight matrix which implies that each spatial unit is connected to the same number k of neighbours, wherever it is localised. This approach avoids arbitrarily defining a distance cut-off and it is particularly indicated when the spatial distribution of points or areas exhibits a high degree of heterogeneity (Anselin, 2002), which is the case with Hermosillo. Another advantage of a k-nearest weight matrix is its capacity to ensure that each observation has the same number of neighbours no matter the size of its territory. A similar matrix has been used by Baumont et al. (2004) and Guillain et al. (2006) while contiguity or great circle distance based matrices have been used in Nijkamp et al. (2009) and Helsel (2008). The general form of a k-nearest neighbour’s weight matrix w(k) is defined as follows

{\begin{matrix} w_{ij}^{*} (k) = 0 i = j, \forall k \\ w_{ij}^{*} (k) = 1 if d_{ij} \leq d_{i} (k) and w_{ij} = \frac{w_{ij}^{*} (k)}{\sum_{j} w_{ij}^{*} (k)} \\ w_{ij}^{*} (k) = 0 d_{ij} < d_{i} (k) \end{matrix}

(1)

where, $w_{ij} (k)$ is an element of the standardised weight matrix; and $d_{i} (k)$ is a critical cut-off distance defined for each unit i. More precisely, $w_{ij} (k)$ is the kth order smallest distance between unit $i$ and all the other units such that each unit i has exactly k neighbours.

Based on contiguity criteria, the average number of neighbours in Hermosillo in 1999 was 4.12 and 4.5 in 2004. As a result, we choose to build several weight matrices (k2, k4, k5, k10) to perform our ESDA and test the sensitivity of our results to the specification of the matrix.

4.2 Global Spatial Autocorrelation

There are a number of ESDA techniques that can be used to study spatial autocorrelation in a geo-referenced dataset. The most widely used statistics to test for the presence of global spatial dependence are the Geary’ C and Moran’s I.³ Given its simplicity and popularity, we will use the latter for our study. It measures the degree of linear association between observed values and its spatially lagged values (Moran, 1948; Hongfei et al., 2007; Anselin et al., 2007). Values of Moran’s I larger (smaller) than the expected value $[E (I) = - 1 / (N - 1)]$ indicate positive (negative) spatial autocorrelation and/or neighbourhood similarity (neighborhood dissimilarity). A value close to 1 indicates neighbourhood similarity, while −1 indicates neighbourhood dissimilarity. A coefficient close to 0 indicates spatial randomness or independence. Formally, Moran’s I is defined as follows

I = \frac{N \sum_{i} \sum_{j} w_{ij} (x_{i} - \bar{x}) (x_{j} - \bar{x})}{(\sum_{i} \sum_{j} w_{ij}) \sum_{i} {(x_{i} - \bar{x})}^{2}}

(2)

where, N is the total number of areas; $w_{ij}$ is the spatial weight measure of contiguity; $x_{i}$ and $x_{j}$ denote the observed values for areas i and j respectively; and $\bar{x}$ is the average of the attribute values. Based on the k4 weight matrix, the Moran’s I values for Hermosillo in 1999 and 2004 are 0.549 and 0.661 respectively and are significant (p-value = 0.001), which confirms the presence of spatial autocorrelation in the distribution of employment density (see Table 2).⁴ This supports our previous results: in Hermosillo, spatially adjacent regions tend to display similar levels of employment density. The 20.35 per cent increase in the Moran’s I values over time suggests that, on average, the employment density in each AGEB has become more similar to that of its neighbours. Previous contributions complement the ESDA approach with an outlier analysis;, however, in our case it is not a very informative analysis. Figure 2 shows us that employment-rich areas are in the centre of the city (upper outliers), while employment-poor areas are localised towards the outskirt of the city (lower outliers).

Table 2.

Moran’s I statistics (standardised values) in Hermosillo, 1999 and 2004 (with weight matrix k4)

Economic sector	1999		2004
	Moran’s I	p-value	Moran’s I	p-value
Forestry, fishing, hunting and agriculture support	−0.006	0.640	0.118	0.006
Mining and oil extraction	−0.010	0.632	0.070	0.026
Manufacturing	0.196	0.001	0.210	0.001
Water and electricity production	−0.002	0.812	−0.001	0.952
Retail and wholesale	0.339	0.001	0.403	0.001
Services	0.646	0.001	0.718	0.001
Total employment	0.549	0.001	0.661	0.001

Table 2 also indicates that positive spatial autocorrelation is not present in all the sectors. The results are not significant in 1999 for the following sectors: forestry, fishing, hunting and agriculture support; mining and oil extraction; water and electricity production. In 2004, only the statistic for water and electricity production was not significant. That year, all the other economic activities registered a positive and significant global spatial autocorrelation, which indicates that areas with similar values of employment density tended to be spatially clustered in Hermosillo. In terms of location choice, spatial dependence means that the city exhibits a homogeneous behaviour about location choice, which can be observed in the most important sectors and over the two time-periods.

The statistics for manufacturing (+6.8 per cent over 1999–2004), retail and wholesale (+18.8 per cent) and services (+11.1 per cent) increase over the study period, showing that spatial dependence is an increasingly important element for these sectors. Retail and wholesale activities have become more concentrated in the CBD and central areas. In the case of Dijon, its CBD is also a group of areas which are centrally located (see Baumont et al., 2004). This is because they provide the most accessible location for workers as well as customers. This may also be because these sectors need to take advantage of economies of scope (such as in the Phoenix metropolitan area; see O’Huallachain and Leslie, 2007), while economies of agglomeration are more important for services (for instance Ile-de-France is specialised in professional and financial services, see Guillain et al., 2006) and manufacturing activities (see McMillen, 2001b, for the case of the industrial city of Milwaukee).

4.3 Local Spatial Autocorrelation

Since the Moran’s I statistic does not allow us to identify employment sub-centres, we switch to a local approach which has been increasingly used to analyse the heterogeneity present in spatial processes. The local indicators of spatial autocorrelation (LISA statistics) allow us to uncover if the concentration of high or low employment density is significantly greater in some contiguous AGEBs than predicted in a spatial homogeneous distribution (global autocorrelation). LISA statistics are defined as follows

I_{i} = \frac{(x_{i} - \bar{x})}{s_{x}^{2}} \sum_{j} [w_{ij} (x_{i} - \bar{x})]

(3)

where, $s_{x}^{2} = \sum_{j}^{n} {(x_{i} - \bar{x})}^{2} / n$ is the variance and other notations are the same as in equation (2).

One way to explore the autocorrelation in space is by means of Moran’s scatterplot. The scatterplot displays the distribution of local spatial autocorrelation according to four quadrants, and the global Moran’s I statistic corresponds to the value of the slope in a Moran scatterplot (Anselin, 1995). For instance, observations in the lower left have low values surrounded by low values (LL) and the upper right quadrant contains all the observations with a high value surrounded by high values (HH), thus representing potential spatial clusters (values surrounded by similar neighbours). On the other hand, observations in the upper left quadrant have a low value and are surrounded by observations with high values (LH) while the lower right quadrant (HL) shows high values surrounded by low values (HL). These last two options suggest potential spatial outliers (values surrounded by dissimilar neighbours).

Figure 3 shows the Moran’s scatterplot which provides additional information on the spatial structure of the data. It plots the standardised employment density in each AGEB against its spatial lag for 1999 (Figure 3, left) and 2004 (Figure 3, right). Both scatterplots confirm a positive spatial autocorrelation. This spatial pattern characterises many areas in Hermosillo, even if many areas have a value close to the average of the sample. The Moran scatterplot can also help us identify the AGEBs that deviate from the global pattern of positive autocorrelation (LH and HL observations).

Figure 3.

Moran’s scatterplot of employment density in 1999 (left) and 2004 (right), based on the weight matrix k4.

The LISA statistics can be classified according to the four categories of the Moran scatterplot (Anselin et al., 2002) and mapped in a LISA cluster map (Figure 4). The results are all significant at the 5 per cent level (based on a permutation approach with 9999 permutations) and are consistent with those obtained earlier.⁵ Therefore, the HH cluster as defined by the results of LISA is the tool we use to identify the employment centre and sub-centres. A similar approach is proposed in Guillain et al. (2006) and Baumont et al. (2004). According to them, a sub-centre is defined by two attributes: it is an AGEB (or a set of neighbouring areas) for which the employment per hectare is significantly higher than the average employment density in Hermosillo; and, it is an AGEB (or a set of neighbouring areas) surrounded by AGEBs for which the average employment density is significantly lower.

Figure 4.

LISA maps of employment densities in 1999 (left) and 2004 (right), based on the k4 matrix.

The number of areas in the cluster of high employment density values (HH) was 27 in 1999 vs 39 in 2004 (see Table 3). The biggest HH cluster (integrated by 25 AGEBs) is located in the centre of the city—around the CBD in 1999, but it is more spread in 2004 (conformed by 34 AGEBs), which indicates that the CBD is spreading (see Figure 4). It was also the case in Dijon, where the CBD was identified as a HH cluster of areas centrally located (Baumont et al., 2004). Also, we note that the incipient sub-centre identified north-west of the CBD in 1999 (see Figure 4) has moved northward in 2004 and is now composed of three AGEBs.

Table 3.

Summarising LISA results, 1999–2004

Cluster	1999			2004			Change in the number
	Number	Percentage	Percentage E	Number	Percentage	Percentage E	Change in the number
High–High	27	10.6	32.6	39	10.7	44.4	12
High–Low	1	0.4	6.4	0	0.0	0.0	−1
Low–Low	56	22.0	3.8	104	28.6	2.5	48
Low–High	7	2.8	1.8	3	0.8	0.6	−4
Not significant	163	64.2	55.4	218	59.9	52.5	55
Total AGEBs	254	100.0	100.0	364	100.0	100.0	110
	Areas	Average	S.D.	Areas	Average	S.D.	Cluster
HH based on 1999	27	3.500	4.355	39	2.542	3.842	HH
HH Based on 2004	27	6.226	6.363	39	4.689	5.872
LL based on 1999	56	0.349	0.051	104	0.331	0.068	LLL
LL based on 2004	56	0.172	0.075	104	0.258	0.033

Note: Elaborated base on local Moran’s I results.

In addition, we note that a new cluster made up of two HH AGEBs emerged in the southern part of the usual CBD at the final period. This can be interpreted as an employment pole or sub-centre (see Figure 4). In 2003, IMPLAN identified several sub-centres which should have led to an agglomeration of activities. They are included in Figure 4 (right) for comparison purposes. Five are localised along Solidaridad, the main north–south corridor of the city—i.e. on the western boundary of the extended CBD. While our results seem to confirm the predictions of IMPLAN for these sub-centres, we do not find any evidence of high employment density AGEBs around the other sub-centres anticipated by IMPLAN. Worse, we actually discover that some of them are surrounded by LL-type AGEBs. It seems that their location along one of the city’s main corridors was not a sufficient condition to support their development.

Other forms of local spatial association include a LL cluster located in the periphery of the city for both years, seven LH-type AGEBs located to the east, south, north-west and west sides of the CBD in 1999, even though only three (located on the east and south sides) keep their significance in 2004. Finally, only one HL-type AGEB appears in 1999 and it is located in the south-east of the city (the industrial area with specialisations in various manufacturing sectors, but principally in the production of automobile and automobile parts for export). However, it lost its statistical significance in 2004 (as well as a great percentage of employment), as a consequence of the fall in US manufacturing in 2000/01 (see Figure 4, left). Hence, it is not a local competition effect that drove the changes in this AGEB. No other AGEB specialises in automobile production, or can provide enough space for its plants.

Table 3 summarises the four different patterns of local spatial autocorrelation. In 1999, 32.6 per cent of the observations were characterised by significant positive spatial association (22 per cent in LL and 10.6 per cent in HH clusters) and concentrated 36.4 per cent of employment. In 2004, significant positive spatial autocorrelation characterised 39.3 per cent of all areas (28.6 per cent in LL and 10.7 per cent in HH clusters) and concentrated 46.9 per cent of total employment. These results indicate that the AGEBs have become more similar to their neighbours over time. This is confirmed in Table 3 which shows that the average level of local spatial autocorrelation among the significant results has increased between 1999 and 2004. As a consequence, negative spatial autocorrelation has decreased both in terms of the number of significant results and intensity.

Now that the spatial locations of the CBD and sub-centres have been identified, we can rely on the value of a location quotient and the regional diversity index to uncover their degree of specialisation and diversification with regard to the city itself.⁶ This methodology has also been used in an urban context by Carroll et al. (2008) and Guillain et al. (2006), as well as Duranton and Puga, (2000). In essence, one area is considered specialised in one sector if its location quotient for that sector is above one. From Table 4, we can observe that the historical CBD is specialised in retailing and wholesale, and the CBD (HH cluster) is specialised in services. Over time, we can observe a link between the diversification of Hermosillo’s CBD and its expansion: a spreading CBD is associated with high values of the regional diversity index. This relationship is the opposite to the relationship found in Ile-de-France (see Guillain et al., 2006): its centre extends while specialising. The north-western sub-centre, specialised in manufacturing as well as in retailing and wholesale, kept its degree of diversity over time. In contrast, the southern sub-centre was in 2004 more diversified, while its trend in specialisation was moving from manufacturing to services.

Table 4.

Locational quotients of employment in Hermosillo, by economic sector

Economic sector	Historical CBD		CBD		Sub-centre 1 (north-west)		Sub-centre 2 (south)
	1999	2004	1999	2004	1999	2004	1999	2004
Forestry, fishing, hunting and agriculture support	—	—	1.976	0.910	—	—	—	—
Mining and oil extraction	—	—	1.592	1.271	—	—	—	—
Manufacturing^a	0.557	0.036	0.288	0.457	1.549	10.418	2.152	0.758
Water and electricity production	—	—	—	—	—	—	—	—
Retail and wholesale	1.539	1.825	1.029	0.925	1.430	3.156	0.612	1.377
Services^a	0.941	0.719	1.430	1.394	0.404	0.571	0.563	0.779
Regional diversity index^c	3.266	1.543	2.527	3.291	1.854	1.843	1.613	3.377

As a consequence of a confidentiality agreement, the data exclude employment in construction.

Includes professional services, but excludes services related to construction, transport, warehousing, finance, insurance and real estate, as well as government services.

Calculated based on the Duranton–Puga index; high index values represent a high degree of diversification in an area and inversely.

Note: A high locational quotient (above 1) indicates that a region is relatively specialised in a particular sector.

Source: based on data from INEGI.

5. Concluding Remarks

In this paper, we contribute to the urban economics literature by focusing on the spatial distribution of employment density in Hermosillo, a middle-sized city in Mexico, under the lens of spatial statistics. Our results confirm our expectations about the dynamics of the city’s employment distribution: while the CBD has remained the densest area in terms of employment, a process of suburbanisation in conjunction with increasing spatial dependence between neighbouring spatial entities has been taking place between 1999 and 2004, the only two years for which data are available.

The results of the Moran’s I and Moran’s scatterplot reveal a significant presence of global spatial autocorrelation and that employment is significantly clustered around the CBD for both periods, which indicates the presence of spatial heterogeneity in the city. This paper shows that ESDA is a useful tool for the identification of centres and sub-centres. First, it has allowed us to detect the CBD and its north-westward extension rather than the supremacy of the historical CBD. Secondly, it helped us to identify the emerging sub-centres located in the southern as well as the north-western part of the city.

Whether the middle-sized city is developed or developing, its CBD can be identified as a HH cluster centrally located, highlighting its monocentric structure; nevertheless, a developed city can be more monocentric, holding all other factors constant, compared with a developing city (Dijon vs Hermosillo). Over time the trend towards the CBD’s sprawl, observed in Hermosillo as well as in Ile-de-France, was throwing back and the number of employment sub-centres has increased in developed as well as in developing cities. However, deep differences in the specialisation of employment centres are a consequence of the city’s degree of development.

Therefore, even though a recent employment decentralisation process has been taking place, we can conclude that Hermosillo is still a monocentric city. Undoubtedly, this result is in contradiction with the idea of polycentrism that Hermosillo’s Planning Institute supports. We believe their misconception comes from the absence of consideration of spatial effects in the methodology they rely on.

Future work aims at updating our results with the data of the 2009 economic census which will be released at the end of 2011. Our goal will be to verify if employment decentralisation in Hermosillo is a lasting phenomenon or whether it only reflects the economic crisis that took place in 2000/01. Finally, we intend to use spatial econometric techniques to estimate the density gradient which reflects how much employment density decreases with distance from the CBD (as in McMillen, 2001a and 2004; and Guillain and Le Gallo, 2009). This should complement our current results about the influence of the CBD and sub-centres on the spatial distribution of employment in Hermosillo.

Footnotes

Notes

Funding Statement

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

References

Alonso

(1964) Location and Land Use. Cambridge, MA: Harvard University Press.

Anselin

(1995) Local indicators of spatial association—LISA, Geographical Analysis, 27, pp. 93–115.

Anselin

(2002) Under the hood: issues in the specification and interpretation of spatial regression models, Agricultural Economics, 27, pp. 247–267.

Anselin

Sridharan

Gholston

(2007) Using exploratory spatial data analysis to leverage social indicator databases: the discovery of interesting patterns, Social Indicators Research, 82, pp. 287–309.

Anselin

Syabri

Smirnov

(2002) Visualizing multivariate spatial correlation with dynamically linked windows. University of Illinois, Regional Economics Applications Laboratory (REAL), Urbana-Champaign, IL.

Barajas

Grijalva

Lara

Velázquez

. (2009) Cuatro décadas del modelo maquilador en el norte de México. Hermosillo: El Colegio de la Frontera Norte y El Colegio de Sonora.

Baumont

Ertur

Le Gallo

(2004) Spatial analysis of employment and population density: the case of the agglomeration of Dijon 1999, Geographical Analysis, 36(2), pp. 146–176.

Carroll

Reid

Smith

(2008) Location quotients versus spatial autocorrelation in identifying potential cluster regions, Annals of Regional Science, 42, pp. 449–463.

Cliff

A. D.

Ord

J. K.

(1981) Spatial Processes: Models and Applications. London: Pion.

10.

Coffey

W. J.

Shearmur

(2002) Agglomeration and dispersion of high-order service employment in the Montreal metropolitan region, 1981–1996, Urban Studies, 39, pp. 359–378.

11.

Duranton

Puga

(2000) Diversity and specialization in cities: why, where and when does it matter?, Urban Studies, 37(3), pp. 533–555.

12.

Fujita

(1988) A monopolistic competition model of spatial agglomeration, Regional Science and Urban Economics, 18, pp. 87–124.

13.

Gary

(1990) The net of mixed beads: suburban office development in six metropolitan regions, Journal of the American Planning Association, 56(4), pp. 457–470.

14.

Giuliano

Small

(1991) Subcenters in the Los Angeles region, Regional Science and Urban Economics, 21, pp. 163–182.

15.

Giuliano

Redfearn

Agarwal

Zhuang

(2007) Employment concentrations in Los Angeles, 1980–2000, Environment and Planning A, 39, pp. 2935–2957.

16.

Gordon

Richardson

(1996) Beyond polycentricity: the dispersed metropolis, Los Angeles, 1970–1990, Journal of the American Planning Association, 62, pp. 289–295.

17.

Griffith

(1981) Evaluating the transformation from a monocentric to a polycentric city, The Professional Geographer, 33(2), pp. 189–196.

18.

Griffith

Wong

(2007) Modeling population density across major US cities: a polycentric spatial regression approach, Journal of Geographical Systems, 9, pp. 53–75.

19.

Guillain

Le Gallo

(2009) Employment density in Ile-de-France: evidence from local regressions, in: Páez

Le Gallo

Buliung

Dall’erba

(Eds) Progress in Spatial Analysis: Theory and Computation, and Thematic Applications, pp. 277–300. Berlin: Springer-Verlag.

20.

Guillain

Le Gallo

Boiteux-Orain

(2006) Changes in spatial and sectoral patterns of employment in Ile-de-France, 1978–97, Urban Studies, 43(11), pp. 2075–2098.

21.

Helsel

J. A.

(2008) Essays on the spatial analysis of manufacturing employment in the U.S. PhD thesis, Kent State University Graduate School of Management.

22.

Hongfei

Calder

Cressie

(2007) Beyond Moran’s I: testing for spatial dependence based on the spatial autoregressive model, Geographical Analysis, 39, pp. 357–375.

23.

IMPLAN (2006) Plan de Desarrollo Urbano (2006–2009). Gobierno Municipal de Hermosillo.

24.

INEGI (Instituto Nacioanal de Estadística Geografía e Informática) (n.d.) website (http://www.inegi.gob.mx; accessed 17 December 2008).

25.

Lang

R. E.

(2003) Edgeless Cities: Exploring the Elusive Metropolis. Washington, DC: The Brookings Institution Press.

26.

Lara

Velázquez

Rodríguez

L. I.

(2007) Especialización económica en Sonora: características y retos al inicio del nuevo milenio, Región y Sociedad, 19, pp. 27–49.

27.

McMillen

D. P.

(2001a) Nonparametric employment subcenter identification, Journal of Urban Economics, 50, pp. 448–473.

28.

McMillen

D. P.

(2001b) Polycentric urban structure: the case of Milwaukee, Economic Perspectives, QII, pp. 15–27.

29.

McMillen

D. P.

(2004) Employment densities, spatial autocorrelation, and subcenters in large metropolitan areas, Journal of Regional Science, 44, pp. 225–244.

30.

McMillen

D. P.

McDonald

J. F.

(1997) A nonparametric analysis of employment density in a polycentric city, Journal of Regional Science, 37, pp. 591–612.

31.

McMillen

D. P.

McDonald

J. F.

(1998) Suburban subcenters and employment density in metropolitan Chicago, Journal of Urban Economics, 43, pp. 157–180.

32.

McMillen

D. P.

Smith

S. C.

(2003) The number of subcenters in large urban areas, Journal of Urban Economics, 53, pp. 321–338.

33.

Mills

E. S.

(1972) Studies in the Structure of the Urban Economy. Baltimore, MD: Johns Hopkins University Press.

34.

Moran

(1948) The interpretation of statistical maps, Journal of the Royal Statistical Society B, 10, pp. 243–251.

35.

Muth

(1969) Cities and Housing: The Spatial Pattern of Urban Residential Land Use. Chicago, IL: University of Chicago Press.

36.

Nijkamp

Cracolici

Cuffaro

(2009) A spatial analysis on Italian unemployment differences, Statistical Methods & Application, 18(2), pp. 275–291.

37.

O’Huallachain

Leslie

T. F.

(2007) Producer services in the urban core and suburbs of Phoenix, Arizona, Urban Studies, 44(8), pp. 1581–1601.

38.

Parr

J. B.

(2002) Agglomeration economies: ambiguities and confusions, Environment and Planning A, 34, pp. 717–731.

39.

Redfearn

Giuliano

Agarwal

(2008) Network accessibility and the evolution of urban employment. Draft report, Project No. 06-16, METRANS.

40.

Rodríguez

(2003) Potencial de desarrollo económico regional en Sonora: Ventaja comparativa intermunicipal del acervo de infraestructura 1993–1998. Instituto Sonorense de Administración Pública A.C., Hermosillo.

41.

Shearmur

Coffey

(2002) Urban employment subcenters and sectoral clustering in Montreal: complementary approaches to the study of urban form, Urban Geography, 23(2), pp. 103–130.

42.

Shearmur

Coffey

Dubé

Barbonne

(2007) Intrametropolitan employment structure: polycentricity, scatteration, dispersal and chaos in Toronto, Montreal and Vancouver, 1996–2001, Urban Studies, 44(9), pp. 1713–1738.

43.

von Thünen

J. H.

(1826/1966) Der Isolierte Staaat in Beziehung auf Landtschaft und Nationalokonomie, Hamburg, transl. by Wartenburg

C. M.

Oxford: Pergamon Press.

Spatial Distribution of Employment in Hermosillo,1999–2004

Abstract

1. Introduction

2. Spatial Distribution of Employment and Sub-centre Identification

3. Study Area and Data

4. Identification of Employment Sub-centres by ESDA

4.1 Spatial Weight Matrix

4.2 Global Spatial Autocorrelation

4.3 Local Spatial Autocorrelation

5. Concluding Remarks

Footnotes

Notes

Funding Statement

References