ESTUDIOS SOBRE LA ECONOMÍA ESPAÑOLA

Does neighboring “industrial atmosphere” matter in industrial location?. Empirical evidence from Spanish municipalities.

Ángel Alañón Rafael Myro

EEE 199

Februay 2005

http://www.fedea.es/hojas/publicado.html

ISSN 1696-6384

Las opiniones contenidas en los Documentos de la Serie EEE, reflejan exclusivamente las de los autores y no necesariamente las de FEDEA.

The opinions in the EEE Series are the responsibility of the authors an therefore, do not necessarily coincide with those of the FEDEA.

Ángel Alañón Assistant Teacher in Applied Economics Facultad de Ciencias Económicas y Empresariales Dpto. Economía Aplicada I Universidad Complutense de Madrid (SPAIN)

Rafael Myro Chair in Applied Economics Facultad de Ciencias Económicas y Empresariales Dpto. Economía Aplicada II Universidad Complutense de Madrid (SPAIN)

Running title: Interurban agglomeration and location.

Corresponding author: Angel Alañón Pardo, Dpto. Economía Aplicada I, Fac. CC. Económicas, Universidad Complutense de Madrid, E-28223 Pozuelo de Alarcón; Fax: +34913942499; Telephone: +34913942470; e-mail: angel@ccee.ucm.es

Abstract

This paper focuses on the role of interurban agglomeration forces in industrial location, assuming that firms not only consider advantages and attraction forces linked to a certain town but also those coming from the nearby area. Exploratory analysis reflects the existence of spatial patterns on the creation of new manufacturing establishments and sheds light on the critical geographical dimensions over which local externalities operate in industrial location, which seem to be larger than shown in previous studies, in accordance with the importance of externalities supposed behind agglomeration forces. Bayesian Spatial Probit models and Standard non-spatial Probit models with spatially lagged explanatory variables are estimated to test if interurban agglomeration forces matter in industrial location. This forces help to explain the location of new manufacturing units in the Spanish municipalities (NUTS V) over the period 1991-1995, together with other better known factors reflecting urban and local externalities, such as manufacturing specialization and diversification or human capital endowment.

Key words: spatial location models, interurban agglomeration

JEL classification: L6; R3

1.INTRODUCTION

In order to choose a good location for their plants, decision-makers could not only be interested in factor endowments and externalities of a given city or town, such human capital or manufacturing specialization and diversity. They also might be interested in factor endowments and externalities in the neighboring area, looking beyond the borders of any given city. In fact, their decision may first pay attention to an big area and then look for a town inside it. So, it is interesting to know what is the role of city and its surrounding area in the decision of location, and what is the relevant neighboring area.

As Ottaviano and Puga [1] point out, economic geography literature identify economic agglomeration at different levels of aggregation, from the small scale (highly specialized industrial district as the city of Prato in Italy) to the large scale agglomerations that cuts across sates (as the US, “Manufacturing Belt” or the European “Hot Banana”). It also identify the forces of agglomeration inside large agglomerations, pointing mainly to pecuniary and technological externalities, underlying too some potential barriers against them, as immobile factor, restriction to the mobility of others (workers particularly), trade cost and economic integration (Krugman [2], Fujita and Thisse [3], Venables [4], Puga [5]). But we do not know enough about the spatial scope of the agglomeration forces involved, if they really cross the regions or are more local, or even about its concrete nature, perhaps because this must be mainly a matter of the empirical analyses. In this aspect, we do not have much more than the recognition of some authors as Ellison and Glaeser [6]: “In practice, we would expect that spillovers might provide benefits also to plants locating in nearby areas”.

This means that we really do not know properly the nature of the interactions between cities placed in a given territory or the role of interurban agglomeration forces. Therefore, we do not know well what is the spectrum of territorial advantages the manager takes into account to choose a determined location for a new plant.

In this paper, we want to approach to that question. Choosing deliberately the town as the spatial unit of reference to analyses the location of new plants in Spain, precisely because it is the smaller unit, and profiting from the fact that we have been able to build up a set of relevant economic information referred to it, we first try to find out what are the critical geographical dimensions over which the externalities of the neighboring town operate. Then, we estimate a simple location model to test if the interurban agglomeration forces previously detected matter in industrial location and if they do more or less than the location advantages operating inside the city.

The most remarkable points of this approach are not only the data set – the continental Spanish municipalities (NUTs V), within a two digits manufacturing context-, but also the inclusion of new explanatory variables, such as the product of Spanish municipalities and the interurban agglomeration forces, and above all, the use of Spatial Statistics, (BB Joint Counts statistics), and Spatial Econometrics techniques (spatial Probit models). These techniques allow us to tackle spatial autocorrelation problems and to include interurban agglomeration forces as one of the explanatory variables.

This paper is organized as follows. In section 2 the role of interurban agglomeration forces in location decision is analyzed and an exploratory analysis is carried out to find out if the creation of new manufacturing units reflects the existence of interurban agglomeration forces and to test the critical distance over which interurban agglomeration forces operate. In section 3 we deal with the Spatial Econometrics issues related to industrial location and introduce Spatial Probit models. In section 4 we present the location determinants (supply and demand factors, external economies and interurban agglomeration forces), the dataset and the specification of the models, which will be estimated in the following section. Then, in section 5, we estimate non spatial Probit models with spatially lagged explanatory variables and Bayesian spatial Probit models in order to explain the location of new manufacturing units in Spain over the period 1991-1995 and discuss the results. Finally, the main conclusions and limitations of this research are exposed in section 6.

2.INTERURBAN AGGLOMERATION FORCES: WHAT ARE THE CRITICAL GEOGRAPHICAL DIMENSIONS OVER WHICH LOCAL EXTERNALITIES OPERATE IN INDUSTRIAL LOCATION

As it has been pointed out above, the spatial economic analyses has mainly been focused in the detection of the agglomeration forces operating in a large scale and in the barriers erected against them that could explain industrial dissemination in the space and the maintain of differences in factor endowments and in industrialization levels among regions.

As agglomeration forces justifying industrial concentration, the accent has been put in the pecuniary and technological externalities, apart from the skill of workers, following the seminal ideas of Marshall (Krugman [2] and [7], Fujita and Thisse [3], Ottaviano and Puga, [1]). The main pecuniary externality comes from the higher efficiency arising from large scale of production reached inside a big market, with a lot of customers and large suppliers (Krugman [7], Venables [4], Martin and Ottaviano [8]). On the other side, the main technological externality stems from the knowledge accumulated inside great clusters of industrial specialization or even of industrial diversification, being crucial the usual distinction between MAR and

Jacobs externalities (Krugman [7], Glaeser at al [9]).

Some factors prevent firms to attend this agglomeration forces, mainly strongest

competition in the more populated areas that depress profits, difficulties to find

workers and rapid growth in their wages, derived from restrictions to the migrations

process, for example, or low economic integration and high trade and transport costs

(Krugman [2], Venables [4], Puga [5]).

Lower efforts have been addressed to know how the identified agglomeration forces

extend over the territory. If they slightly surpassed the borders of the city where the

firms are set up, we would find it difficulty to explain the creation of large scale

agglomerations. But if they extended largely beyond this borders we could distinguish

between the spectrum of towns involved, attending a different combination of local

and regional advantages of location. To choose a location for new plan, the managers

of a company will try to maximize the territorial advantage, attending local and

regional agglomeration forces.

The kind of activity should matter, as we would expect higher interurban

agglomeration forces in the sectors with large scale economies, more depending on

the surrounding territory, and less in that other with large technological spillovers,

supposed more locally concentrated, although the available evidence do not confirm

this patterns of concentration neither in U.S (Krugman [2], Kim [10]) or in Europe

(Fluviá and Gual [11]) or in Spain (Callejón and Costa [12]) .Similarly, we could

expect further territorial impacts of the town with higher industrial diversification.

Some few authors have approach theses questions. Following Audretsch and Feldman

[13], Ciccone and Hall [14], Ellison and Glaeser [6], Holl [15] and Viladecans-Marsal

[16], agglomeration economies operate within limited geographic scope and should be analyzed at a local scale better than in a interregional framework. However, information spillovers may flow between neighboring cities or towns; some non-trade local inputs may also be shared between the firms of the cities of a regional area; and, thanks to commuting, a local skilled-labor pool may be no restricted to a city. In fact, the nature of the technological and pecuniary externalities mentioned does not seem local. Examples of related concepts to interurban agglomeration forces can be found in Scott [17] and in Saxenian [18]. At a local level there is evidence of interurban agglomeration forces in Rosenthal and Strange [19] (2003), in Viladecans-Marsal [16] and in Alañón [20] (2004). Rosenthal and Strange [19] measure the geographic extent of agglomerative externalities using ZIP code data for the U.S. Viladecans-Marsal [16] analyses industrial location and externalities for the most crowded Spanish cities (over 15,000 inhabitants), which share a common fiscal system. Finally, in Alañón [20] the role of interurban agglomeration forces in per capita product generation is estimated for all the Spanish municipalities.

Anyhow, the effects of interurban agglomeration forces are supposed to be smoothly declining with distance (Ellison and Glaeser [6]). That is, these effects should be inversely proportional to the distance between these firms or between firms and natural advantages. Therefore, interurban proximity plays a key role in the development of interurban agglomeration forces.

In spite of the extensive literature which underlines the ubiquity of concentration and stresses the advantages of agglomeration¹, Simmie [24], Suarez-Villa and Alrod [25] or Arita and McCann [26] cast serious doubts about the spatial extent of agglomeration.

¹See Hoover [21], Carlton [22], Porter [23] or Krugman [2].

In this section we are testing if interurban agglomeration forces may play a role in the creation on new manufacturing establishments. If they matter the creation of new manufacturing units should exhibit positive spatial autocorrelation, that is, they should be clustered in space.

BB Joint Count Test for spatial autocorrelation or spatial dependence reflects if binary variables are clustered or randomly distributed in space. BB Joint Count Test is defined as follows²:

\[B B = (1 / 2) \sum_ {i} \sum_ {j} w _ {i j} L O C _ {i} L O C _ {j}\tag{1}\]

where is the element of a spatial weights matrix W, LOC is set to 1 for municipality i or j if new units of a given manufacturing activity have been created over the period , and is set to 0 otherwise. Spatial weights matrix W reflects the potential interaction between the observation pair i and j. A positive and significant z-value for this statistic indicates positive autocorrelation, i.e., similar values, either high values or low values, are more spatially clustered than could be caused purely by chance (Anselin [5]).

Figures 1 and 2 shows BB Joint Count test for the creation of new manufacturing units in the continental Spanish municipalities, (NUTS , over the period 1991- 1995. These test have been calculated for 11 manufacturing activities and 11 spatial binary weight matrices, Wd, based on distance thresholds in which interurban agglomeration forces are supposed to be active (5 kms, 10, kms, 15, kms, … 150 kms). The elements, of the matrix are set to 1 if the distance between municipalities i and m i is d or less and to 0 otherwise.

² See Anselin [27] or Cliff and Ord [28] for technical details.

³ This period of time it is a representative one since it includes years of positive and non positive economic growth (1991, 1992, 1994 , and 1995, on the one hand, and 1993 on the other). We restrict our exploratory analysis to this period for two reasons. Firstly, a larger sample could bias the statistics, since LOC is set to one no matter if one or fifty new establishments are set up in a given municipality. And, finally, we do not use a more recent dataset in order to be consistent with the estimations of next section, whose independent variables are referred to 1990 or to 1991.

4 We have taken into consideration 7906 municipalities. Balearic Islands and Canary Islands municipalities, Ceuta and Melilla –two cities located in the north of Africa – and some continental municipalities with no reliable data have been excluded from the dataset.

As expected, as distance increases spatial autocorrelation decreases, so spatial clustering or interurban agglomeration forces are weaker. Spatial dependence seems to vanish between 100 and 150 km and is strongest between 15 and 20 kilometers. Surprisingly, there are not very significant differences among manufacturing activities but as hoped, in sector with stronger scale economies as transport equipment or chemistry the spatial dependence is prolonged until 150 Km., being the opposite case for textiles and clothes, for example. In a general view, high- and medium-tech manufacturing activities exhibit higher spatial dependences than the less intensive sectors in technology, as food. textile, clothes, wood or products from non metallic minerals. This seems to indicate than not only large scale sectors but others with higher technological spillovers have larger spatial influence.

These results are in line with Rosenthal and Strange [19] and Viladecans-Marsal [29], [30] and [16]. In Rosenthal and Strange [19] agglomeration effects are significant for the location of manufacturing activities and these effects are limited to short distances, not far from 20 km. In Viladecans-Marshal [29] and [30] Spanish municipalities are considered neighbors, belonging to same economic area, if the distance between them is not far from 20-30 km. Finally, in Viladecans-Marshal [16], also devoted to the Spanish municipalities, according to the exploratory analysis – Moran’s I- the strongest distance threshold is 20 km.

Although spatial autocorrelation may be due to the characteristics of spatial data, the existence of spatial processes, such as interurban agglomeration forces, as outlined in the beginning of this section, may explain the results of BB Joint Count test.

FIGURE 1a BB Joint Count Statistics Significance

3.INDUSTRIAL LOCATION AND SPATIAL ECONOMETRIC ISSUES

Usually, location models are constructed considering the location decision problem as one of random profit maximization⁵ (Figueiredo et al [32]. Following McFadden [33] and Carlton [34] it is considered that if an entrepreneur, who previously decided to open a new establishment in manufacturing branch j, locates in municipality i it will produce a potential profit of . Formally,

\[\pi_ {i j} = U _ {i j} + \varepsilon_ {i j}\tag{2}\]

where stands for a random variable, which is expected to be distributed independently. So, this entrepreneur will locate in municipality i if the potential profit is greater than in other municipalities, say m, that is

\[\pi_ {i j} > \pi_ {m j}\tag{3}\]

where . This profit depends on a set of local characteristics, and it is usually expressed as a linear combination of these characteristics (Figueiredo et . Thus, in our case this profit would depend also on the characteristics of the neighboring area

\[\pi_ {i j} f (X _ {n}, W X _ {n})\tag{4}\]

⁵ See Guimarães et al. [31] for an extension of the random utility framework.

where the explanatory variables and account for the local characteristics which impact on profits and for the relevant characteristics of the neighboring municipalities respectively. So, WX could be substituted by

\[\pi_ {i j} f (X _ {n}, W \pi_ {i j})\tag{5}\]

As it is not possible to observe (Ellison and Glaeser, [6]) the dependent variable of location models is usually the number of new establishments or new firms created over a period of time –our LOC variable in the previous section. Then, expressing LOC as a linear combination of independent variables from equation (4)

\[L O C _ {i j} = \Sigma_ {n} \beta_ {n} X _ {n} + \Sigma_ {n} \rho_ {n} W X _ {n} + \varepsilon_ {i j}.\tag{6}\]

Location decision models are usually estimated using limited dependent variable models, i.e., Logit, Probit or Poisson specifications⁶. However, as in equations (4), (5) and (6) we are making depend on what happen in neighboring municipalities the assumptions of an independently distributed is too strong. As it was shown in the previous section the creation of new manufacturing establishments is autocorrelated in space. Although the existence of univariate spatial dependence in does not necessarily mean that the residuals of equation (6) are autocorrelated there will probably be spatial dependence in that model.

The existence of spatial autocorrelation invalidate the use of most of usual statistics and econometrics techniques, such us ordinary least squares⁷. So, to obtain reliable estimates spatial autocorrelation needs to be treated properly. However, as far as we know, there is no mention about spatial autocorrelation in industrial location models literature⁸.

⁶ See Arauzo [35], Holl [15] and [36] or Guimaraes et al [31].

⁷ See Anselin [38] for more information about spatial autocorrelation and Spatial Econometrics techniques.

Spatial Econometrics techniques are not commonly used by economists yet (Anselin and Florax, [39]).

Spatial autocorrelation in spatial data and processes may be treated in different ways. Basically, it may be removed from the dataset⁹ or included in the specification of the model. Two of the more common alternatives are the so called spatial autoregressive models, (SAR), and spatial error models, (SEM).

SAR models include a spatially lagged dependent variable, , as one of the explanatory variables, that is:

\[y = \rho W y + X \beta + \varepsilon\tag{7}\]

where y is a nx1 vector of observations on the dependent variable, Wy is a nx1 vector of spatial lags for the dependent variable, is the spatial autoregressive coefficient, X is a nxk matrix of observations on the (exogenous) explanatory variables with associated a kx1 vector of regression coefficients , and ε is a nx1 vector of normally distributed random error terms, with means 0 and constant (homoskedastic) variances

SEM models deal with spatial dependence through a spatially lagged error term, that is:

\[\begin{array}{l} y = X \beta + u \\ u = \lambda W u + \varepsilon \\ \varepsilon \approx N (0, \sigma^ {2} I _ {n}) \end{array}\tag{8}\]

Where λ is a coefficient on the spatially correlated errors.

Both spatially lagged dependent variables in SAR models and spatially lagged error terms in SEM models reflect the existence of spatial autocorrelation and both may be interpreted as a way to treat spatial dependence properly. However they may also have an economic meaning. It can be clearly shown in SAR models, equation (7) and its reduced form, equation (9), since it makes depend what happens in a given location –on what happens in the neighboring locations:

⁹ For example implementing robust estimation techniques, applying spatial filters or enlarging or improving the dataset.

\[y = (I - \rho W) ^ {- 1} X \beta + (I - \rho W) ^ {- 1} \varepsilon\tag{9}\]

In order to take into account spatial autocorrelation and to reflect properly the potential effect of neighboring locations in the creation of new manufacturing units a spatial limited dependent variable model is required¹⁰. In presence of spatial autocorrelation standard Logit, Poisson and Probit models are discarded since ε does not follow a normal distribution in limited dependent models so the resulting multivariate specification is intractable in Logit and in standard Poisson models, and standard Probit estimation is inconsistent (Anselin [42], p. 8). Thus spatial Probit models, both error and lag, become a feasible option of estimating location models. Following Anselin [43] and Fleming [44] there are several ways to implement spatial Probit models¹²: generalized methods of moments (GMM) estimation for error models (Pinkse and Slade [45]); EM (Expectation, maximization) approach for error models (McMillen [46]) or simulation estimators such as the Gibbs Sampler (LeSage [47] and [48]; and Smith and LeSage [49]) etc. In spite of some drawbacks¹³ we have chosen the Gibbs Sampling approach to estimate Bayesian Probit models proposed in LeSage [47], [48] and Smith and LeSage [49] because “it is the most flexible of the spatially dependent models because it can incorporate spatial lag dependence and spatial error dependence in addition to general heteroskedasticity¹⁴, of unknown form (Fleming [44], pp.166-167)”.

¹⁰ As we are interested in the economic meaning of spatial autocorrelation we are not trying to remove it from the dataset.

¹¹ However, Kaiser and Cressie [40] developed a Poisson auto-model which allows positive spatial dependencies in multivariate count data by specifying conditional distributions as truncated or Winsorized Poisson probability mass functions. See Kaiser and Cressie [40] or Arbia [41] for a more detailed discussion.

¹² See Fleming [44] for a more complete discussion on the advantages and disadvantages of different spatial probit estimation techniques.

¹³“…not clear full ‘simultaneity’ has been properly accounted for” (Anselin [43]).

4.LOCATION DETERMINANTS, DATA AND MODEL SPECIFICATION

Locations models try to explain how certain variables may influence location decisions. Most empirical works usually group these variables in categories such as supply factors, demand factors, external economies and diseconomies etc (Guimaraes et al. [31]). As we are mainly interested on the role of interurban agglomeration forces in industrial location we are not carrying out an extensive analysis of location determinants¹⁵,¹⁶. The location determinants we are taking into consideration are: human capital as supply factor; gross municipality product as demand factor; local external economies (location and urbanization); and interurban agglomeration forces. As dependent variable, we use , a binary variable which is set to 1 if at least one new unit of manufacturing activity j has been located in municipality i over the period 1991-1995 and to 0 otherwise.

Human capital index, is defined as the percentage of population over ten years old with at least a secondary school degree in municipality i in 1991. The expected sign is positive since it reflects labor market’s qualification.

Gross municipality product in 1991, reflects the volume of economic activity of the municipality, the potential market for new firms, so its expected sign is positive. It has been calculated regressing Spanish provinces (NUTs III) gross value added on some agglomeration and production variables, , using Spatial Econometrics techniques to overcome spatial autocorrelation problems. was obtained multiplying the estimated coefficients , and , by the municipal values of

¹⁴ (0, ), ( , ,...., ) _{1 2}² _nε ≈ N σ V V = diag v v v .

15 See Hayter [50], Guimarães et al. [51] , Figueiredo [32] or Guimarães et al. [31] for more information about locational determinants

¹⁶ Anyway, unless we use provinces (NUTs III) data there are not many homogeneous economic municipality data for the whole of Spain (Alañón [52]). However if we used provinces data we would

External economies are represented by the classic location quotient and by a diversity index. The location quotient, , represents the advantages of geographical specialization, traditional location economies, Marshallian externalities or MAR’s type (Marshall, Arrow and Romer) (Glaeser et al., [9]). It is defined as follows:

\[L Q _ {i, j} = \left(E _ {i j} / E _ {I}\right) / \left(E _ {J} / E _ {T}\right)\tag{10}\]

Where accounts for total employment in manufacturing activity j in municipality i, for total employment in municipality i, for national employment in manufacturing activity and total national employment in all manufacturing activities. Its expected sign is positive.

is a manufacturing diversification index for municipality i in 1990. The expected sign of this variable is positive since manufacturing diversity may reflect the existence of inter-industrial external economies, such as Jacobs type (Glaeser et al. [9]), and, also, because the creation of new plants is biased towards diverse cities (Duranton and Puga [54]). This index is based on the correction for differences in sectoral employment shares at the national level of the inverse of a Hirschman-Herfindahl index proposed in Duranton and Puga [54]):

\[D I _ {i} = 1 / \sum_ {j} / s _ {i j} - s _ {j} /\tag{11}\]

Where, is the share of manufacturing activity j in manufacturing employment in municipality i, and is the share of manufacturing activity j in total national manufacturing employment.

fall into ecological fallacy and Modifiable Areal Unit problems. See Anselin [38] or Arbia [53] for a more detailed discussion.

¹⁷ See Alañón [52] for more details.

Finally, we consider the potential role of interurban agglomeration forces, IAF_i. They may be measured by the spatially lagged independent variables in a standard (non spatial) Probit model, (WHC, WLQ, WDI and WMP), where W is a spatial weights matrix, or by the spatially lagged dependent variable in a spatial autoregressive Probit model¹⁸, (WLOC). As spatial weights matrix, W, we will use a binary contiguity matrix which elements, are set to 1 if municipalities i and m have a common border an 0 otherwise, and, in order to be consistent with the results of explanatory analysis we will also use a binary matrix which elements , are set to 1 if the distance between municipalities i and m i is 15 kilometers or less and to 0 otherwise. Since the average Spanish municipality diameter is around 10 km (Holl [15]) both spatial weights matrix should be equivalent.

Therefore, the spatial location of new manufacturing units could be explained as usual as a function of urban and interurban external economies, apart from the advantages in human capital and greater potential through the following expression:

\[L O C _ {i j} = f \left(H C _ {i}, L Q _ {i j}, D I _ {i}, M P _ {i}, I A F _ {i}\right)\tag{12}\]

The source for manufacturing unit’s creation is REI, Registro de Establecimientos Industriales (Industrial Establishments Register)¹⁹. Employment and capital human data sources are Censo de Locales 1990 and Censo de Personas 1991 (Spanish Census 1990-1991)²⁰.

¹⁸ The economic interpretation of SEM models is not so straightforward, so we will only consider SEM models as a way to deal properly with spatial autocorrelation.

¹⁹ REI belongs to Ministerio de Industria y Energia (Spanish Ministry of Industry and Energy).

²⁰ Although there are new census data for population (Censos de Población y Viviendas 2001), we can not obtain new data for employment at a municipality level since Censo de Locales 1990 is the last one available.

In this section three models are estimated²¹ for each manufacturing activity: a standard Probit with lagged explanatory variables, (PLEV), a Bayesian spatial autoregressive Probit, (SARP), and a Bayesian heteroskedastic spatial error model, (SEMP).

Results are summarized in table 1²². Surprisingly, Bayesian spatial Probit models estimates, both SARP and SEMP, are not very different from PLEV ones²³.

All non spatially lagged explanatory variables, HC, LQ, MP and DI, are always significant. So population qualification, manufacturing specialization (location economies), market potential, and diversity (urbanization or Jacobs external economies) play an important role in location processes.

²¹ Standard Probit models are estimated using Econometric Views 4. Bayesian spatial Probit models estimations are carried out using MATLAB R.12, via Markov Chain Monte Carlo methods (MCMC).

²² Extended results are reported in appendix I.

²³ Usually non spatial models produce larger coefficient estimates since they ignore spatial dependence and spatial effects are attributed to the rest of explanatory variables in these non spatial models (Smith and LeSage [49]).

These results differ in a certain extent from the evidence shown in previous studies. In Viladecans-Marsal [16] while urbanization economies influence location in most sectors specialization only plays a minor role.

Another striking result is the lack of significance of spatially lagged Human Capital indicator, WHC, in most manufacturing activities. It could be mean both commuting is not very important in Spain as a whole (excluding the biggest cities) or that commuters are not very qualified.

Non significance of spatially lagged Location Quotient, WLQ, in food and tobacco or in First transformation of metals does not seem too strange. In fact, non spatially lagged LQ indicator could be non significant for food and tobacco since it relies on basic human necessities which may be fulfilled at a local level. And first transformation of metals could be deeply rooted in the mineral sources due to transport costs. However WLQ is highly significant in the rest of activities, which could reflect the existence of industrial districts or interurban Marshallian external economies.

Spatially lagged market potential indicator, WMP, is not significant in most manufacturing activities, so decision makers could focus primarily in their internal market. This seems strange in sector exhibiting large scale in production.

The high significance of the spatially lagged diversity indicator, WDI, stresses the key role or inter-industrial linkages at an interurban level. As it was suggested at the beginning of this paper and in the comments on WLQ indicator, WDI also supports the evidence for interurban agglomeration economies.

General interurban agglomeration economies indicator, IAF in SARP models, is only non significant in computers and office equipment, which could be in line which the empirical evidence shown in Simmie [24], Suarez-Villa and Alrod [25] and Arita and

McCann [26]. However, IAF is always significant in the rest of manufacturing activities. Therefore, what happens in the neighboring municipalities, the interurban agglomeration forces, matter in location decision processes.

Although Viladecans-Marsal [16] provides empirical evidence of the existence of interurban agglomeration forces in Spanish biggest cities, its results differ slightly²⁴ since agglomeration effects only spillover beyond the administrative borders in three of the six sectors analyzed. Finally, as already quoted in section 2, Rosenthal and Strange [19] also found evidence for the effects of interurban agglomeration effects till a distance threshold of around 20 km.

6.CONCLUSIONS

This research has focused in something not enough explored in the empirical literature about location, the idea that not only urban agglomeration forces but interurban agglomeration forces influence the creation of new manufacturing units in a given municipality, making the manager of the companies to take into account a mix of urban and interurban forces in order to maximize the territorial advantage of the new plant. Our work has taken the Spanish municipalities as the unit of analysis. Exploratory study has shown that the creation of new manufacturing units at two digits level exhibits a spatially autocorrelated pattern, more important for high- and medium-tech sectors with large scale in production. This spatial behavior has two important implications. On the one hand, it stresses the key role of interurban agglomeration forces in the creation of new manufacturing units, which seems to be highly relevant between 15 and 20 kilometers and to vanish beyond 100 or 150 kilometers. This last distance is larger than that obtained in previous works, and more coherent with the main agglomeration forces stressed by economic geography literature, linked to scale economies and technological spillovers.

²⁴ We must have in mind that these works are not carried out using the same methodology and do not use exactly the same dataset, thus full comparison is not possible.

In the other hand, spatial autocorrelation should be taken into consideration when estimating location models, since spatial dependence invalidates the use of traditional estimation techniques.

In order to confirm the role of interurban agglomeration forces in the creation of new manufacturing units confirmatory analysis has been carried out. A simple location model has been outlined and estimated using spatial and non-spatial techniques. Although estimates do not differ broadly between spatial and non spatial models spatial variables are highly significant– spatial autoregressive coefficient for dependant variable in Bayesian spatial autoregressive Probit models and most of spatially lagged explanatory variables in standard Probit models. It could be reflecting the existence of interurban agglomeration forces since the significance of these variables means that what happens in a municipality depends on what happens in its neighboring area.

However, we must introduce some caveats about the statements and preliminary conclusions drawn and pointed out above, since could be highly dependant on the units of analysis. Thus further research is needed to improve our knowledge of location decision processes.

First, from a spatial point of view, municipalities could not be the ideal unit of analysis for location of new units of manufacturing industries. Further research is needed to find out what kind of spatial aggregation fits location analysis perfectly, since results could differ greatly²⁵, and it could be different for every kind of manufacturing activity, so standard metropolitan areas –which are not still devised for Spain- are discarded.

²⁵ It is due to the Modifiable Areal Unit Problem, MAUP. See Anselin [38] or Arbia [53].

Second, even if the level of municipality is confirmed as the more appropriate for the analysis of location, we have to distinguish highly industrialized municipalities from other categories We have also to separate the municipalities highly specialized in a given industrial activity from other municipalities which are very diversified.

Although we do not ignore spatial heterogeneity –Spatial Probit models are heteroskedastic-, estimations could be improved devising spatial regimes. However, as in the case of metropolitan areas, spatial regimes cannot be drawn intuitively, or rely only on administrative criteria.

Lastly, but not least, location decisions may differ according to firm size.

ACKNOWLEDGEMENTS

Earlier versions of this paper were presented at the Nethur School ‘The dinaymics of firm location’ -Groningen, February 5-6, 2004- , at VII Encuentro de Economía Aplicada –Vigo, June 3-5, 2004, and at First Seminar of Spatial Econometrics Jean Paelinck -Zaragoza, October 22-23, 2004. Comments and suggestions from Philip McCann, Enrique López-Bazo, Anna Matas, José Luis Roig and the rest of participants are acknowledged and appreciated. Josep María Arauzo provided part of the dataset and also made useful suggestions. Giuseppe Arbia and Elisabet Viladecans-Marsal recommend us relevant bibliographical references. Belén Rey helped us with earlier versions. The usual disclaimer applies.

REFERENCES

1. [1] G. I. P. Ottaviano, D. Puga,. Agglomeration in the Global Economy: A survey of the “New Economic Geography”, World Economy 21(6) (1998), pp. 707-731.
2. [2] P. Krugman, Geography and trade, MIT Press, Cambridge (Mass), 1991.
3. [3] M. Fujita, J.-F. Thisse, Economics of Agglomeration, Journal of the Japenese and International Economies 10 (1996), pp.339-78.
4. [4] A. J. Venables, Equilibrium Locations of Vertically Linked Industries, International Economic Review 37 (1996), pp. 341-59.
5. [5] D. Puga, The rise and Fall of Regional Inequalities, European Economic Review 43 (1999), pp.303-334.
6. [6] G. Ellison, E, L, Glaeser, Geographic Concentration in US Manufacturing Industries, A Dartboard Approach, Journal of Political Economy 105 (1997), pp. 889- 927.
7. [7] P. R. Krugman, Increasing returns and Economic Geography, Journal of Political Economy 99 (1991), pp. 484-99.
8. [8] P. Martin, G.I.P Ottaviano, Growing locations: Industry location in a Model of Endogenous Growth, European Economic Review 43 (1999), pp. 281-302.
9. [9] E. L. Glaeser, H. Kallal, J. Scheinkman, A, Shleifer, Growth in cities, Journal of Political Economy 100 (1992), pp. 1126-1152.
10. [10] S. Kim, Expansion of Markets and the Geographic Distribution of Economic Activities: The trends in U.S. Regional Manufacturing Structure, 1860-1987, Quarterly Journal of Economics (1995).
11. [11] M. Fluvia, J. Gual, Comercio Internacional y Desarrollo Regional en el Marco de la Integración Económica Europea, In J.M. Esteban and X. Vives (Eds.), Crecimiento y Convergencia Regional en España y Europa, Instituto de Análisis Económico, Consejo Superior de Investigaciones Científicas, Barcelona, 1994.
12. [12] M. Callejón, M. Costa, Economías Externas y Localización de las Actividades Industriales, Economía Industrial 305 (1995).
13. [13] D. B. Audretsch, M. P. Feldman, R&D spillovers and the geography of innovation and production, American Economic Review 86 (1996), pp. 630-640.
14. [14] A. Ciccone, R. Hall, Productivity and the density of economic activity, American Economic Review 86 (1996), pp. 329-355.
15. [15] A. Holl, Manufacturing location and impacts of road transport infrastructure: empirical evidence from Spain, Journal of Regional Science and Urban Economics 34 (2004), 341-364.
16. [16] E. Viladecans-Marsal, Agglomeration economies and industrial location: citylevel evidence, Journal of Economic Geography 4 (2004), pp. 562-582.
17. [17] A. J. Scott, New industrial spaces, Pion, London, 1988.
18. [18] A. Saxenian, Regional Advantage, Culture and Competition in Silicon Valley and Route 128, Harvard University Press, Cambridge (Mass.), 1994.
19. [19] S. S. Rosenthal, W. C. Strange, Geography, industrial organization and agglomeration, The review of Economics and Statistics (2003), pp. 377-393.
20. [20] A. Alañón, Un modelo espacial de renta per cápita regional: evidencias provincial, comarcal y municipal, Investigaciones Regionales 4 (2004), pp.
21. [21] E. M. Hoover, The location of Economic Activity, McGraw-Hill, New York, 1948.
22. [23] M. E. Porter, The Competitive Advantage of Nations, Free Press, New York, 1990.
23. [24] J. Simmie, Reasons for the Development of ‘Islands of Innovation’: Evidence from Hertfordshire, Urban Studies 35 (1998), pp. 1261-1289.
24. [25] L. Suárez-Villa, W. Walrod, Operational Strategy, R&D and Intrametropolitan Clustering in a Polycentric Structure, The advanced Electronics Industries of the Los Angeles Basin, Urban Studies 34 (1998), pp. 1343-80.
25. [26] T. Arita, P. McCann, Industrial alliances and firm location behaviour: some evidence from the US semiconductor industry, Applied Economics 32 (2000), pp. 1391-1403.
26. [27] L. Anselin,. SpaceStat Tutorial, a book for using SpaceStat in the analysis of spatial data, Urbana-Champaign: University of Illinois, 1992.
27. [28] A. D. Cliff, J. K. Ord, Spatial processes: models & applications, Pion, London, 1980.
28. [29] E. Viladecans-Marsal, Economías externas y localización del empleo industrial, Revista de Economía Aplicada 31 (2003), pp, 5-32.
29. [30] E. Viladecans-Marsal, La concentración territorial de las empresas industriales: un estudio sobre el tamaño de las empresas y su proximidad geográfica, Papeles de Economía Española 89-90 (2001), pp. 308-321.
30. [31] P. Guimarães, O. Figueiredo, D. Woodward, Industrial location modeling: extending the random utility framework, Journal of Regional Science 44 (2004), pp.1- 20.
31. [32] O. Figueiredo, P. Guimarães, D. Woodward, Home-field advantage: location decisions of Portuguese entrepreneurs, Journal of Urban Economics 52 (2002), pp. 341-361.
32. [33] D. McFadden, Conditional logit analysis of qualitative choice behaviour, in P. Zarembka, (Ed.), Frontiers in Econometrics, Academic Press, New York, 1974, pp, 105-142.
33. [34] D. W. Carlton, The Location and Employment Choices of New Firms: An Econometric Model with Discrete and Continuous Endogenous Variables, Review of Economics and Statistics 65 (1983), pp. 440–449.
34. [35] J. M. Arauzo,. Determinants of Industrial Location. An application for Catalan Municipalities, Working Paper Nº 138, Estudios sobre la Economía Española, FEDEA, Madrid, 2002.
35. [36] A. Holl, Transport infrastructure, agglomeration economies, and firm birth, Empirical evidence from Portugal, Journal of Regional Science 44 (2004), pp. 693- 712.
36. [38] L. Anselin, Spatial econometrics, Methods and models, Kluwer Academic, Dordrecht, 1988.
37. [39] L. Anselin, R. Florax, Introduction, in: L. Anselin, R. Florax (Eds), New Directions in Spatial Econometrics. Springer, Berlin:, 1994.
38. [40] M.S. Kaiser, N. Cressie, Modeling Poisson variables with positive spatial dependence, Statistics & Probability letters 35 (1997), pp. 423-432.
39. [41] G. Arbia, Introductory spatial econometrics with applications to regional convergence, Springer-Verlag, Berlin, 2005 (forthcoming).
40. [42] L. Anselin, Under the Hood. Issues in the Specification and Interpretation of Spatial Regression Models, Agricultural Economics 17 ( 2002), pp. 247-267.
41. [43] L. Anselin, Spatial Probit Models, lecture available at http://sal.agecon.uiuc.edu/courses/se/se2001/spProbit.html., 2001.
42. [44] M. M. Fleming, Techniques for estimating spatially dependent discrete choice models in L. Anselin, R. J. G. M. Florax. S. J., Rey (Eds.), Advances in Spatial Econometrics, Methodology, Tools and Applications, Springer, Berlin, 2004, pp, 145- 167.
43. [45] J. Pinkse, M. Slade, Contracting in Space: An Application of Spatial Statistics to Discrete-Choice Models, Journal of Econometrics 85 (1988), pp. 125-154.
44. [46] D. P. McMillen, Spatial effects in Probit models: a Monte Carlo investigation, in L, Anselin and R, Florax (Eds.), New Directions in Spatial Econometrics, Springer, Berlin, 1995, pp. 189–228.
45. [47] J. P. LeSage, Bayesian Estimation of Spatial Autoregressive Models, International Regional Science Review, 20 (1997), pp. 19-35.
46. [48] J. P. LeSage, Bayesian estimation of limited dependent variable spatial autoregressive models, Geographical Analysis 32 (2000), pp. 19–35.
47. [49] T. Smith, J. P. LeSage, A Bayesian Probit Model with Spatial Dependencies, available in www.spatial-econometrics.com, 2002.
48. [50] R. Hayter, The dynamics of industrial location, The factory, the firm and the production system, Wiley, New York, 1997.
49. [51] P. Guimarães, O. Figueiredo, D. Woodward, Agglomeration and the Location of Foreign Direct Investment in Portugal, Journal of Urban Economics 47 (2000), pp. 115-135.
50. [52] A. Alañón,. Estimación del valor añadido per cápita de los municipios españoles en 1991 mediante técnicas de econometría espacial, Ekonomiaz 51 (2002), pp. 172- 194.
51. [53] G. Arbia, Spatial data configuration in statistical analysis of regional economic and related problems, Kluwer Academic Publishers, Dordrecht, 1989.
52. [54] G. Duranton, D. Puga, Diversity and specialisation in cities, why, where and when does it matters?, Urban Studies 37 (2000), pp. 533-555.

APPENDIX I ESTIMATIONS OUTPUT

I.1 Food and tobacco

Non-spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.2 Clothes and leather

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.3 Wood and furniture

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.4 Printing and Paper

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.5 Chemistry

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.6 Other non metallic minerals

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.7 First transformation of metals

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.8 Machinery

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.9 Computer, office equipment etc.

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.10 Electric and electronic equipment

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model

I.11 Transport equipment

Non spatial Probit with spatially lagged explanatory variables

Bayesian spatial autoregressive Probit model

Bayesian heteroskedastic spatial error Probit model