IS PRODUCTIVITY HIGHER IN BRITISH CITIES?

JOURNAL OF REGIONAL SCIENCE, VOL. 00, NO. 0, 2012, pp. 1–25

IS PRODUCTIVITY HIGHER IN BRITISH CITIES?*

Richard HarrisThe Business School, University of Glasgow, Main Building, Glasgow, G12 8QQ.E-mail: [email protected]

John MoffatUniversity of Swansea, Richard Price Building, Singleton Park, Swansea, SA2 8PP.E-mail: [email protected]

ABSTRACT. This paper examines the determinants of total factor productivity (TFP) using a GBplant-level data set. The main findings relate to whether spatial spillovers and “place” effects areimportant: plants located in cities generally perform better than plants in the same region outside ofthese cities; but with the exception of Bristol, no city has significantly higher TFP levels than the SouthEast. This suggests that spatial externalities associated with city location are not as important as thebenefits of being situated in the South East region.

1. INTRODUCTION

It is widely accepted that cities have higher productivity. For example, in a recentreview of the literature on agglomeration economies, Puga (2010) writes: “ . . . firms andworkers are much more productive in large and dense urban environments than in otherlocations . . . over the past 30 years, urban economists have been successful at documentingand quantifying these advantages” (p. 203). He goes on to review the evidence and refersto the study by Rosenthal and Strange (2004) which reports that a doubling of city sizeincreases productivity by 3–8 percent.

Much of the literature to date has either focused on identifying the effect of aggregateagglomeration economies on productivity (e.g., Combes et al., 2008), or city size/densityon wages (e.g., Fingleton, 2003). The present study goes further by using business microdata to test whether agglomeration, along with other city-based advantages, translatesinto higher city-level total factor productivity (TFP) in Great Britain. Our main resultssuggest that (other than London) the core cities of Great Britain have on average lowerlevels of TFP than the South East of England (our “frontier” benchmark region), althoughplants in these cities do generally better (controlling for a wide range of characteristics)than if they had been (re-)located in the noncity hinterlands of the regions in which theyreside (especially in services). As expected, our results show that not all cities do equallywell (e.g., Leicester vs. Nottingham; Liverpool vs. Manchester; Edinburgh vs. Glasgow). Inshort, there is no overwhelming evidence from this study in support of British cities beingthe ideal location for encouraging growth, particularly in high-technology industries;

*This work contains statistical data from ONS, which is Crown copyright and reproduced with thepermission of the controller of HMSO and Queen’s Printer for Scotland. The use of the ONS statistical datain this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of thestatistical data. This work uses research data sets, which may not exactly reproduce National Statisticsaggregates. It was also carried out as part of an ESRC grant (RES-591–28-0001) awarded to the SpatialEconomics Research Centre, based at the London School of Economics.

Received: July 2011; revised: May 2012; accepted: May 2012.

C© 2012, Wiley Periodicals, Inc. DOI: 10.1111/j.1467-9787.2012.00778.x

1

2 JOURNAL OF REGIONAL SCIENCE, VOL. XX, NO. X, 2012

especially as diversification (or urbanization) economies were found to be largely negative(suggesting significant diseconomies associated with, inter alia, congestion).

We study differences in productivity (specifically the productivity of both labor andcapital inputs into the production process, i.e., total factor productivity) since this is widelyrecognized as a key driver of long-run economic growth (e.g., Baumol, 1984; Krugman,1997). To the best of our knowledge, this study is the most comprehensive and up-to-dateof its kind, using plant-level panel data for Great Britain covering 1997–2006 to estimateproduction functions for eight subsectors spanning marketable output in Great Britain(four in manufacturing, and four in marketable services). Our data set allows us to controlfor a wide range of determinants of TFP such as technical progress due to undertakingR&D in the plant, and exogenous gains over time; the age of the plant; access to bet-ter technology through belonging to a foreign-owned multinational firm (“greenfield” vs.“brownfield” impacts for different subsets of countries are also considered); the impact ofmultiplant economies of scale (particularly whether the firm operates in more than one GBregion) and external market-based competition effects; and lastly and most importantlythe impact of spatial spillover and “place” effects (as proxied by measures of industrialagglomeration and diversification, as well as the impact on productivity of being locatedin a particular region and/or city).1

The next section provides a brief overview of the literature on agglomeration exter-nalities (Section 2). This is followed in Section 3 by a discussion of the modeling approachand data used to obtain estimates of plant-level TFP. The results are then presented inSection 4, with an emphasis on spatial spillover and “place” effects. Finally, we summarizeour major findings and discuss some policy implications.

2. SPATIAL SPILLOVER AND “PLACE” EFFECTS

Spatial spillovers or agglomeration externalities are benefits that accrue to plantsfrom being located in the vicinity of large concentrations of other plants. Duranton andPuga (2004) describe the mechanisms that give rise to agglomeration externalities; assummarized by Overman et al. (2009) these are: “sharing,” “matching,” and “learning.”Agglomeration benefits arise through sharing when firms benefit from drawing on acommon pool of resources. There are four main assets which can be shared in areasof concentrated economy activity. First, indivisible goods or facilities such as ports orairports can be shared. Second, plants can share the benefits that arise from a widevariety of input suppliers. Such benefits exist when intermediate inputs are not perfectsubstitutes and the number of inputs that plants employ will therefore depend upon thenumber of available inputs. Benefits also accrue to plants through sharing the gains ofnarrower specialization. Essentially, a larger pool of labor allows greater specializationwhich in turn confers benefits upon the plants that employ that labor. Finally, the sharingof risk helps plants. In response to a positive (negative) shock, a plant may seek toincrease (reduce) employment. For plants operating in isolated areas, this will have alarge positive (negative) impact on the wage rate. This limits the ability of the plant toadapt its output in response to positive (negative) shocks. This will not be such an issuein areas of concentrated economic activity.

As to “matching,” locations with large numbers of firms and workers make it easierfor both parties to find a productive match. The probability, and therefore the speed,of matching is also improved in areas with many firms and workers. This reduces the

1 This list of determinants is not meant to imply that we do not consider the skill characteristics ofworkers employed in each plant to be unimportant as a determinant of TFP; but unfortunately the datawe use do not contain such information on human capital.

C© 2012, Wiley Periodicals, Inc.

HARRIS AND MOFFAT: IS PRODUCTIVITY HIGHER IN BRITISH CITIES 3

costs of recruitment. Finally, the diffusion and accumulation of knowledge is expectedto be better in areas with many people. In spite of improvements in communicationstechnology, distance continues to act as a barrier to learning. This is especially trueof tacit knowledge. By facilitating face-to-face contact, the concentration of people in aparticular area will facilitate the transfer of knowledge. In addition, workers will find iteasier to move from one firm to another. This process will cause the transfer of knowledge(i.e., “learning”) across firms. The same diffusion of knowledge will occur when plantsare better able to learn from their customers and suppliers when they are located inclose proximity. The accumulation of knowledge will also be enhanced in areas of highconcentrations of economic activity if, as a result of the aforementioned benefits that arisefrom agglomeration, there are greater resources available to devote to the generation ofnew knowledge. A rich empirical literature that tests for the existence of knowledgespillovers using spatial econometrics (see, e.g., Audretsch and Feldman, 1996; Anselinet al., 1997; Acs et al., 2002) has generally found them to be significant determinants ofinnovative activity.

Agglomeration externalities are usually distinguished in the literature accordingto whether they are an intra- or interindustry phenomenon. Intraindustry externalitiesare termed MAR (Arrow, 1962; Romer, 1986; Marshall, 1890) or localization externalitieswhile interindustry externalities are termed Jacobian (Jacobs, 1970, 1986) or urbanizationexternalities. The mechanisms that give rise to agglomeration externalities can supportboth localization and urbanization externalities. For instance, firms may learn from otherfirms in the same industry and from firms in another industry.

Many studies have sought to estimate the impact of agglomeration externalities onproductivity without distinguishing between urbanization and localization externalities.For instance, Aberg (1973) finds that population density has a positive impact on produc-tivity using Swedish regional data. According to his estimates, a doubling of populationdensity will lead to an increase in productivity of 2 percent. Similarly, using StandardMetropolitan Statistical Area (SMSA) data, both Sveikaukas (1975) and Segal (1976) findsthat productivity is higher in large U.S. cities, having controlled for other factors. UsingU.S. state level data, Ciccone and Hall (1996) find that a doubling of employment densityincreases labor productivity by 6 percent. For EU NUTS-3 data, Ciccone (2002) obtains acorresponding figure of 4.5 percent. Rice et al. (2006) use NUTS3 data for the U.K. andshow that their regional productivity index, which controls for occupational composition,is positively related to the population of working age close to the area. Doubling the popu-lation of working age close to a region is associated with an increase in their productivityindex of 3.5 percent. Using French area-level data, Combes et al. (2008) estimate theimpact of population density and market potential on TFP. They estimate a number ofspecifications but their preferred estimates are elasticities of 3.5 percent and 2.5 percentfor population density and market potential, respectively. Andersson and Loof (2011) usefirm level data from Sweden and find that firms located in larger regions with high em-ployment levels have higher TFP. Overall, there therefore appears to be relatively strongevidence in favor of some form of agglomeration externalities.

A further literature has sought to estimate the relative importance of urbanizationand localization externalities. Here, the empirical evidence on the existence of urbaniza-tion and localization externalities is split into studies using area level data and studiesusing firm or plant-level data. In relation to the former, Moomaw (1983) employs data forU.S. SMSAs derived from the Census of Manufacturing and finds evidence for an impacton productivity of both types of externalities. Nakamura (1985), using area level data forJapan, finds that firms in heavy industries benefit more from localization externalitieswhile firms in light industries experience more urbanization externalities. Across all in-dustries, the average elasticity of productivity with respect to industry size and city size



is 0.05 and 0.03, respectively, which leads him to conclude that localization externalitiesare relatively more important in determining productivity. Henderson (1986) finds littleevidence of urbanization economies but strong evidence of localization economies usingU.S. and Brazilian area level data, aggregated from firm level data. For the U.S., themean elasticity of productivity with respect to industry size was 0.19, while for Brazilthe corresponding figure was 0.11. Rigby and Essletzbichler (2002) regress labor produc-tivity on variables measuring the local input-output relationship between suppliers, thelocal labor mix, technological spillovers, and area size using US data. The first threecan be seen as measures of localization externalities while area size is a measure ofurbanization externalities. All four measures are significant which suggests that bothtypes of externalities may be important determinants of productivity. Using a panel ofEuropean regions, Brulhart and Mathys (2008) find, for most sectors, evidence of urban-ization economies but evidence of localization diseconomies. Graham et al. (2010) findevidence for the existence of both localization and urbanization economies using postcodelevel data aggregated from the U.K.’s Financial Analysis Made Easy (FAME) data set.The evidence from studies using aggregate data is therefore generally supportive of theexistence of both types of externality.

By contrast, the empirical evidence from micro-data studies generally suggests thatlocalization externalities are, in terms of their effect on productivity, more important.Henderson (2003), using plant-level data from the U.S. Longitudinal Research Database,finds evidence that localization externalities have a strong impact on TFP in high-techindustries but not in manufacturing industries. The average localization elasticity acrossindustries is 0.03. He is unable to find any evidence of productivity enhancing effects ofurbanization externalities. Capello (2002) also finds evidence that localization external-ities have a positive impact on TFP using data on high-tech firms in the metropolitanarea of Milan. The evidence for urbanization externalities is far weaker. Baldwin et al.(2010), using Canadian plant-level data, find that productivity growth is positively andsignificantly associated with the change in variables designed to capture Marshallianexternalities (these are the degree of labor market specialization, the local density ofupstream suppliers, and the number of plants from the same industry within 5 km). Incontrast, they find a negative relationship between productivity growth and the growthin local population which is included as a proxy for Jacobian externalities. Using Frenchplant-level data, Martin et al. (2011) find that productivity growth is positively associatedwith the change in their Marshallian externalities variable. However, they find little ev-idence that urbanization externalities are important determinants of TFP growth. Theysuggest the latter result may reflect congestion diseconomies. Van Der Panne (2004), us-ing a Dutch data set of firms that have announced new products, investigates the impactsof localization and urbanization externalities on innovation, and finds that the formerhas a positive impact on innovativeness but that the latter has no significant explanatorypower. Assuming a link between innovation and productivity (see, for example, Creponet al., 1998 for evidence in support of this assumption), this is further evidence in favorof the idea that Marshallian externalities have a stronger impact on productivity thanJacobian externalities.

Other studies have managed to find evidence that localization externalities have amajor role in improving productivity. Graham (2009), for instance, finds evidence thatboth types of externality have a positive impact on productivity using firm level datafrom the U.K.’s FAME data set on 27 industries. A positive and statistically significantimpact of localization externalities is found for 13 industries while the same is found forurbanization externalities in 14 industries. The weighted average localization elasticityis 0.03 for manufacturing and 0.01 for services while the corresponding urbanizationelasticities are 0.07 and 0.19. Overman et al. (2009), using establishment level data from



the ARD, also provide evidence for the existence of urbanization externalities. They alsofind that being situated near other establishments from the same industry has a negativeimpact on TFP. As we do below, they also include region and city dummies in their model.Their coefficients indicate that, with the exception of London, firms in regions and citiesoutwith the South East have lower TFP than firms inside the South East region. Ingeneral, firms in cities have higher productivity than firms in their rural hinterland.2

Another reason why it may be expected that plants in areas with heavy concentra-tions of economic activity have higher productivity relates to firm selection. Melitz andOttaviano (2008) provide a model in which such areas attract more firms because of theirlarger markets. This increases competition, and causes less productive firms to exit. Thisimplies that firms and plants operating in cities will have higher levels of productivity.Combes et al. (2009) attempt to distinguish between agglomeration and firm selectioneffects by exploiting their different implications for the distribution of firm productivity.Agglomeration effects will lead to a rightward shift of the productivity distribution for allfirms in large cities while selection effects will lead to a truncation of the left tail of theproductivity distribution in large cities. Using French firm level data, they find strongevidence for agglomeration effects in large cities but no systematic evidence of selectioneffects.

In summary, there have been a number of studies to date that suggest that “spillover”effects associated with location have a positive impact on productivity and growth. How-ever, many of these studies use aggregated data and/or do not control for a number ofother determinants of productivity that are likely to be important, especially the char-acteristics of plants themselves. Thus it is difficult to distinguish between the impact of“place” versus “firm” effects using the results from this earlier work. In this study wecontrol for both sets of determinants, in an attempt to provide more robust evidence onwhether productivity is higher in cities.

3. DATA AND MODEL ESTIMATED

Plant-level panel data from the Annual Respondents Database (ARD)3 are usedcovering 1997–2006 and all market-based sectors for Great Britain (although in thisstudy we have omitted the following industries: those areas of agriculture, fishing, andforestry covered in the ARD; mining & quarrying; utilities; construction, sales and motorvehicles repairs, wholesale and retail distribution; and financial services).4 These dataare collected by the U.K.’s Official for National Statistics (ONS) each year as part of theAnnual Business Inquiry, designed to obtain statistics for calculating the national incomeaccounts. In our econometric analysis we weight the data using sample weights to ensure

2 The results by Overman et al. (2009) have a similar order of ranking to those we obtain (seeSection 4), but their methodology is significantly different to ours. For instance, they use OLS regressionwhich does not take account of endogeneity or fixed effects; and their main establishment level controls(which take account of firm characteristics) only include labor and capital when compared to the 13 vari-ables used here. They also do not allow for any differences in technology between various manufacturingand service sectors (i.e., all firms are assumed to have access to the same technology). They also use un-weighted data and consequently do not deal with the problems posed by the stratified nature of the ARD.In addition, they perform their analysis at what the ONS refer to as the reporting unit (rather than theplant) level, which is an accounting rather than an economic unit. Harris (2005) provides a discussion ofmost of these issues.

3 For a detailed description of the ARD and discussion of several issues concerning its appropriateuse, see Oulton (1997), Griffith (1999), and Harris (2002, 2005).

4 For most of these industries we have no data on capital stocks, or they are only partially coveredby the ARD.



that the distribution of plants for which there is financial data are representative of thepopulation of plants operating in each year in Great Britain. Weighting is necessary bothto ensure that population parameters are estimated and because of the fact that oneof the endogenous variables in the model (employment) is used by the ONS as part ofthe stratified sampling approach to collect the ARD data; thus leading to the problem ofendogenous sampling or stratification (see the appendix in Harris, 2002).

The full set of available variables, and their definitions, is set out in Table 1. In-formation on intra- and extramural expenditure on R&D is available from the BusinessEnterprise R&D (BERD) database on enterprises that undertake this activity each year,and this is merged into the ARD. The other variables were created using the ARD, anddue to space considerations the reader is referred to Harris and Moffat (2011) for therationale for their inclusion.

Of particular importance here is our attempt to capture two particular types ofintraarea spatial spillover: localization externalities that arise due to industrial special-ization and are thus an intraindustry phenomenon (referred to as MAR externalities);and urbanization economies (or Jacobian externalities), that arise due to interindustryspillovers. For both types of spillovers, we are allowing for spatial productivity effectsamongst plants located in the same geographical area. We define the latter as the lo-cal authority district. Spillovers based on wider (and/or other types of) geographic areasare captured by a different set of (dummy) variables: variables are included measuringwhether the plant was located in an Assisted Area, and to which Government Office regionand city it belonged.5 Note, we expect all of these variables—industrial agglomeration anddiversification, as well as assisted-area status, and region and city location—to capturedifferent aspects of “spillovers,” with different locations (e.g., a single city) experiencing amix of potentially diverse impacts.

Spatial productivity spillovers are, by definition, limited in range but the extent towhich they are limited is likely to be an empirical issue (e.g., Gertler, 2003; Venables,2011). Despite the often-cited importance of face-to-face contact when transferring (in-tangible, tacit) knowledge, it might be argued that our choice of local authority district (ofwhich there are 381 in Great Britain) is too small a spatial area for capturing industrialagglomeration and diversification effects. We have therefore also calculated these indicesbased on travel-to-work areas (of which there are 243), where travel to work area (TTWA)is officially defined as an area in which at least 75 percent of the resident economicallyactive population actually work in the area, and also, in which of everyone working in thearea, at least 75 percent actually live in the area. Both sets of results are discussed below.

We have also experimented with different industrial agglomeration and diversifi-cation indices (but note that unlike the literature covered in Kominers, 2008, we arenot measuring whether an industry is agglomerated spatially by using an aggregatedindustrial agglomeration measure; rather we are trying to capture MAR-spillovers bymeasuring the percentage of output located in each local authority district for each five-digit industry). With regard to industrial agglomeration Devereux et al. (2007) used avariable measuring the number of plants in each industry in each county-year, whichis significantly correlated with our measure but which we believe to be inferior (as itignores plant size and thus the relative amount of output produced by an industry at aparticular location). For diversification, there are also several different approaches, from

5 The “core” cities we identify were either capitals (i.e., Cardiff and Edinburgh) or they met thecriteria of (in 2001) employing over 250,000 with a population density of 20 + persons per hectare; or theyhad employment of over 100,000 and densities of 30 + persons per hectare. They closely accord with thedefinition of “core” cities used in Great Britain (see http://www.corecities.com/). Thus “urban areas” thatincorporate large hinterlands (e.g., Leeds) are excluded on the population density criterion.



TABLE 1: Variable Definitions Used in BERD-ARD Panel Data Set for 1997–2006

Variable Definitions Source

Real gross output Plant level gross output data deflated by 2-digitONS producer price (output) indices. Data are in£’000 (2000 prices)

ARD

Real intermediate inputs Plant-level intermediate inputs (gross output minusGVA) deflated by 2-digit ONS producer price(input) indices (nonmanufacturing only has asingle PPI). Data are in £’000 (2000 prices)

ARD

Employment Number of employees in plant ARDCapital Plant & machinery capital stock (£m 1995 prices)

plus real value of plant and machinery hires(deflated by producer price index) in plant.Source: Harris and Drinkwater (2000, updated).

ARD

Age Number of years plant has been in operation basedon year of entry

ARD/IDBR

Single-plant Dummy coded one when plant comprises asingle-plant enterprise

ARD

>1 region multiplant Dummy coded one if plant belongs to multiplantenterprise operating in more than one U.K. region

ARD

Greenfield U.S.-owned Dummy coded one if U.S.-owned and newly openedduring 1997–2006

ARD

Brownfield U.S.-owned Dummy coded one if U.S.-owned and not newlyopened during 1997–2006

ARD

Greenfield EU-owned Dummy coded one if EU-owned and newly openedduring 1997–2006

ARD

Brownfield EU-owned Dummy coded one if EU-owned and not newlyopened during 1997–2006

ARD

Greenfield Other foreign-owned Dummy coded one if foreign-owned by anothercountry and newly opened during 1997–2006

ARD

Brownfield Other foreign-owned Dummy coded one if foreign-owned by anothercountry and not newly opened during 1997–2006

ARD

Herfindahl Herfindahl index of industry concentration (3-digitlevel).

ARD

Industry agglomeration % of industry output (at 5-digit SIC level) located inlocal authority district in which plant islocated—MAR-spillovers

ARD

Diversification % of 5-digit industries (from over 650) located inlocal authority district in which plant islocated—Jacobian spillovers

ARD

R&D undertakena Dummy coded one if plant had positive R&D stockbased on undertaking intramural and/orextramural R&D since 1997

BERD

Assisted Area Dummy coded one if plant located in assisted area ARDRegion Dummy coded one if plant located in particular

Government Office regionARD

City Dummy coded one if plant located in major GB city(defined by NUTS3 code)

ARD

Industry Dummy coded one depending on 1992 SIC of plant(used at 2-digit level).

ARD

aR&D stocks were computed using perpetual inventory method with 30 percent depreciation rate for thelargest components of R&D spending (intra-mural current spending and extramural R&D). See Harris et al.(2009) for details of methods used.



the simple measure used by Baldwin et al. (2010) of the population size of an area, to usinga locational Herfindahl index, calculated using employment shares for disaggregatedindustries for each area in each year, excluding a plant’s own industry (see, e.g., Devereuxet al., 2007). These two alternative measures of diversification were strongly correlatedwith the one used here; the correlation with population density (we prefer this to actualpopulation numbers to control for the spatial size of the district) is 0.55, and with thelocational Herfindahl index we had an overall correlation of 0.67 (it differs by year, butnever falls below 0.48). We also believe our diversification index is “better” since using five-digit industries and 381 local authorities, the mean of the locational H-index (subtractedfrom 1) was 0.98 with a standard deviation of 0.012 (i.e., most local authorities have littlediversification); our measure has a mean of 55.3 (standard deviation of 8.1)—see Table 2below.

Finally, there is also the possibility of spillovers across spatial boundaries (e.g., be-tween local authorities or TTWAs i and j, or even across larger spatial areas). The usualway of taking account of such spillovers is by using spatial econometric models whichinvolve the use of predefined spatial “weighting” or “W” matrices. These models offer thepossibility for considerable progress in the simulation (and therefore our understanding)of the effects of spatial shocks. However, we have not taken this approach here; first,because of the computational demands of estimating spatial econometric models withmicro-data which prior experience informs us are severe and would necessitate omittinga large part of our sample.6 Second, as far as we are aware there are no published papersthat have used spatial econometrics in conjunction with system GMM (generalized methodof moments; our preferred approach incorporating fixed effects and endogenous regres-sors). Additionally, we have not modeled interarea spatial effects because of uncertaintysurrounding the specification of the “W” matrix. Ex ante, we have little information onthe nature of spatial interactions (they may be a mix of global, national, and local effects7)and, despite recent developments in tests of the specification of spatial econometric mod-els (see, e.g., Kelegian, 2008), it is not possible to empirically test the theoretical validityof what could be a very large number of potentially relevant “W” matrices.8

While we do not estimate interarea spillovers using the spatial econometrics ap-proach, our measure of intra-area spillovers can be criticized regarding the way in whichlocalization and urbanization spillovers are measured. It is similar to a spatial economet-rics approach in which the area is redefined to be a spatial unit smaller than the localauthority (e.g., electoral wards) and the “W” matrix is calibrated so that spillovers arisebetween spatial units within local authorities but not between spatial units across localauthorities. We therefore agree with Corrado and Fingleton (2012) who argue that anyapproach that uses proxies for spatial spillovers “requires strong identifying assumptionsand therefore possesses no real advantage compared to employing a W approach (p. 224).”

6 In Harris et al. (2011), using 32 Gb of memory on a PC cluster, it was only possible to estimatea model using 2,373 observations (rather than the 4,559 observations that were available) because ofthe size of the matrices necessary to estimate such models. The smallest sector here (hi-tech manufac-turing) contains 12,906 observations while the largest sector (low KI market services) contains 351,721observations.

7 Harris et. al. op. cit. found that R&D spillovers were best modeled by a mix of national andcontiguous regional effects (global effects were not tested because of a lack of data).

8 Note, LeSage and Pace (2010) show that in the spatial econometrics model marginal effects compro-mise both direct and indirect effects; arguing that estimates of these “total” effects are not overly sensitiveto the choice of the “W” matrix. Given that direct and indirect impacts depend on the “W” chosen (seeEquations 22–25 in Corrado and Fingleton, 2012), we think the “jury is still out” on whether the “total”effects approach is moving towards a solution to the problems created by having to choose a particularvalue of “W.”



TABLE 2: Mean (Weighted) Values 1997–2006, by Sectora

Manufacturing Service

Sector (1) (2) (3) (4) (5) (6) (7) (8)Real gross output 22787.0 17634.2 12159.1 14773.9 9347.8 10561.8 1798.1 2028.6Real intermediate

inputs15243.4 12489.9 8173.8 9959.4 5008.5 5152.9 1023.4 1554.8

Capital 5.9 7.3 7.0 6.0 7.0 4.4 0.5 0.3Employment 141 114 112 118 64 98 32 22Age 4.4 4.7 6.1 6.2 5.7 11.7 7.8 8.2Single-plant 0.245 0.254 0.121 0.105 0.030 0.097 0.022 0.025>1 region multiplant 0.513 0.524 0.702 0.654 0.800 0.801 0.849 0.823Greenfield

U.S.-owned0.040 0.046 0.024 0.029 0.040 0.015 0.010 0.066

BrownfieldU.S.-owned

0.049 0.053 0.036 0.037 0.067 0.053 0.042 0.037

GreenfieldEU-owned

0.042 0.049 0.047 0.033 0.017 0.011 0.037 0.002

BrownfieldEU-owned

0.042 0.060 0.090 0.043 0.041 0.035 0.040 0.009

Greenfield Otherforeign-owned

0.022 0.021 0.016 0.015 0.007 0.003 0.002 0.001

Brownfield Otherforeign-owned

0.024 0.018 0.019 0.027 0.024 0.012 0.008 0.002

Herfindahl 0.023 0.018 0.028 0.015 0.083 0.011 0.026 0.064Industry

agglomeration1.00 1.40 3.26 2.35 0.83 1.78 0.60 0.65

Diversification 52.3 54.6 57.5 55.5 53.9 70.2 58.9 57.5R&D undertaken 0.242 0.206 0.171 0.117 0.048 0.035 0.002 0.002Assisted Area 0.196 0.228 0.205 0.200 0.156 0.174 0.165 0.171

Notes: a(1) High-tech: Pharmaceuticals (SIC244); office machinery & computers (SIC30); radio, TV & communications equip-ment (SIC32); medical & precision instruments (SIC33); aircraft & spacecraft (SIC353);

(2) Medium high-tech: Chemicals (SIC24 exc. pharmaceuticals, SIC244); machinery & equipment (SIC29); electrical machin-ery (SIC31); motor vehicles (SIC34); other transport equipment (SIC 35 exc. ships & boats, SIC351, and aircraft & spacecraft,SIC353);

(3) Medium low-tech: Coke & petroleum (SIC23); rubber & plastics (SIC25); other nonmetallic (SIC26); basic metals (SIC 27);fabricated metals (SIC28); ships & boats (SIC351);

(4) Low-tech: Food & beverages (SIC15); tobacco (SIC16); textiles (SIC17); clothing (SIC18); leather goods (SIC 19); woodproducts (SIC 20); paper products (SIC21); publishing, printing (SIC22); furniture and other manufacturing (SIC36); recycling(SIC37);

(5) High-tech KI: Telecoms (SIC642); computer & related (SIC72 exc. maintenance & repair, SIC725); R&D (SIC73); photo-graphic activities (SIC7481); motion pictures (SIC 921); radio & TV activities (SIC922); artistic & literary creation (SIC9231);

(6) KI services: Water transport (SIC61); air transport (SIC62); legal, accountancy & consultancy (SIC741 exc. managementactivities of holding companies, SIC7415); architecture & engineering (SIC742); technical testing (SIC 743); advertising (SIC744);

(7) Low KI: Hotels & restaurants (SIC55); land transport (SIC60); support for transport (SIC63); real estate (SIC70); rentingmachinery (SIC 71); maintenance & repair of office machines (SIC725); management activities of holding companies (SIC7415);labor recruitment (SIC745); investigation services (SIC746); industrial cleaning (SIC747); packaging (SIC7482); secretarial services(SIC7483); other business services (SIC7484); sewage & refuse (SIC90);

(8) Other low KI: Postal services (SIC641); membership organizations (SIC91); other entertainment services (SIC923 exc.artistic & literary creation, SIC9231); news agencies (SIC924); sporting activities (SIC926); other recreational activities (SIC927);other services (SIC93).

Other variables which would allow measurement of particular sources of agglomera-tion externalities (such as “labor-mix” and the location of upstream suppliers—cf. Baldwinet al., 2010—or the extent to which co-agglomerating firms trade with each other and/oremploy similar workers—cf. Ellison et al., 2010) are either not available or we prefer ourapproach (e.g., the use of employment fluctuations in individual establishments, relativeto fluctuations in their sector, is to us a rather limited proxy for “share” effects due toagglomeration).

Table 2 presents the mean (weighted) values for the variables, broken down by theeight sectors used in this study. That is, we decided to estimate production functions



for plants operating with similar technologies, with the sectors chosen based mostly onEurostat definitions (although with some minor amendments).9 High-tech manufacturingplants were largest (in terms of gross output, intermediate inputs, and employment), butthey were relatively younger (which in part explains why their capital stock was relativelysmall compared to other manufacturing sectors). In general manufacturing plants werelarger than those operating in the service sectors, although the age of the plants in serviceswas generally higher. Single-plant operations were more prevalent in manufacturing,while in services over 80 percent of plants belonged to enterprises operating plants inmore than one region.

Around 22–25 percent of plants in manufacturing (excluding low-tech manufacturing)were foreign-owned, with “brownfield” plants somewhat more likely to be in operation, andoverall EU-ownership predominating (except for high-tech manufacturing where U.S.-ownership was a little more likely). Plants owned by firms from other foreign-ownedcountries were in the minority (generally across all sectors). For low-tech manufacturingand high-tech knowledge-intensive (KI) services, around 18–20 percent of plants wereforeign-owned (EU-ownership was more likely in low-tech manufacturing; U.S.-owned inhigh-tech KI services); “brownfield” plants were generally more common than “Greenfield”plants.10 In the remaining service sectors, foreign-owned plants accounted for around12–14 percent of all plants (U.S.-ownership dominating KI services and other low KIservices; EU-owned more likely in low KI services), and again “brownfield” plants weregenerally more in evidence (except for other low KI services where “Greenfield” U.S.-ownedprevailed).

The average Herfindahl index of industrial concentration was generally very lowacross all sectors;11 the highest levels of competition were in KI services, low-tech manu-facturing, medium high-tech manufacturing, and high-tech manufacturing, while compe-tition was relatively low (on average) in high-tech KI services and other low KI services.Industry agglomeration was highest in the medium low-tech manufacturing sector (cov-ering such industries as metals and shipbuilding), followed by low-tech manufacturing; itwas lowest on average in low KI and other low KI services. Diversification was relativelyhigh in KI services (covering higher-level business services), but there was little differenceacross the other sectors covered. R&D was undertaken in around 24 percent of high-techmanufacturing plants, falling to 12 percent in low-tech manufacturing; far fewer servicesector plants undertook formal R&D (some 4–5 percent of those in high-tech KI and KIservices, and less than 1 percent for the other two service sectors included). Lastly, 16–23percent of plants were located in Assisted Areas, where they were eligible for help fromsuch U.K. Government schemes as Regional Selective Assistance, various R&D schemes,and EU assistance (mostly from the ERDF).

There are several approaches to estimating TFP using micro-level panel data. DelGatto et al. (2011) and Van Beveren (2012) provide useful surveys on these differentapproaches to measuring TFP. The analysis here is more limited as we consider onlymicro-econometric approaches. This allows us to concentrate on those methodologies thathave become the most commonly used in recent years, relying on micro-level (e.g., firm or

9 http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/Annexes/htec_esms_an3.pdf.10 We distinguish between plants newly set-up by foreign-owned enterprises (“greenfield” plants) and

those acquired from U.K.-owned companies (“brownfield” plants), given our expectation that there may bedifferences in productivity levels for these two subsets. See Harris and Robinson (2002) for evidence onwhether foreign-owned firms “cherry-pick” previously U.K.-owned plants.

11 Dividing these numbers into one gives the “numbers-equivalent” of equal-sized firms on averageoperating in each sector.



plant12) panel data that are much richer for analyzing heterogeneity across enterprisesand thus provide a better understanding of the causes of TFP differences.

Here we define TFP using a Cobb-Douglas log-linear production function approach(including fixed-effects, �i):13,14

yit = �i + �Eeit + �Mmit + �Kkit + �X Xit + �Tt + εit,(1)

where endogenous yit, eit, mit, and kit refer to the logarithms of real gross out-put, employment, intermediate inputs, and capital, respectively, in plant i in time t(i = 1, . . . , N; t = 1, . . . T); and Xit is a vector of observed (proxy) variables determin-ing TFP. In order to calculate TFP, Equation (1) is estimated directly (e.g., Harris, 2005)providing values of the elasticities of output with respect to inputs (�E, �M, and �K ), andthen TFP can be measured as the level of output that is not attributable to factor inputs(employment, intermediate inputs, and capital)—i.e., TFP is due to efficiency levels andtechnical progress. A commonly used alternative approach is to omit Xit when estimatingEquation (1), calculate TFP and then regress this, in a second-stage regression, on likelydeterminants (i.e., the variables in Xit). Clearly this latter approach would result in biasedestimates of � because of the omission of relevant variables in the first stage.

A large class of models have been used to estimate (1), using micro-level panel data.The two most popular in recent work are: (i) the Olley and Pakes (1996) and Levinsohnand Petrin (2003) approaches, which account for both endogeneity of inputs and outputsin the production function and selection bias due to firm entry and exit (which is likelyto be correlated with productivity), by using two-stage procedures where unobserved TFPis “proxied” by another state variable(s) such as investment or intermediate inputs; and(ii) our preferred approach of system GMM, which allows for fixed effects and tacklesendogeneity of the right-hand-side variables and selection bias by using their laggedvalues (in first differences and levels) as instruments. In essence, Olley and Pakes replaceEquation (1) with:15

yit = �0 + �Eeit + �Kkit + �Mmit + h (iit, kit) + εit,(1a)

where TFP is proxied by h(.)—which itself is approximated by a higher-order polynomialin iit and kit—and iit is investment. Levinsohn and Petrin replace h(.) with h(mit, kit).Both approaches do not allow for fixed effects and make some strong assumptions whencompared to the systems GMM approach (as discussed in Ackerberg et al., 2006).

Thus Equation (1)—in dynamic form with additional lagged values of output andfactor inputs—was estimated using the system GMM approach available in STATA 9.2(Blundell and Bond, 1998). This is sufficiently flexible to allow for both endogenous

12 Note, we use plant level data, rather than firm level, because the latter often comprises plantslocated across a number of geographic locations (especially in larger firms). Thus, using firm level datawould not allow us to fully disentangle the impact of geographic “spillover” effects.

13 The inclusion of fixed effects is necessary as empirical evidence using plant- and firm-level paneldata consistently shows that plants are heterogeneous (productivity distributions are significantly “spread”out with large “tails” of plants with low TFP) but more importantly that the distribution is persistent—plants typically spend long periods in the same part of the distribution. Evidence using the ARD has beenpresented in, for example, Haskel (2000) and more recently Martin (2008). Evidence from other countriesis presented in Baily et al. (1992), Bartelsman and Dhrymes (1998). Such persistence suggests that plantshave “fixed” characteristics (associated with access to different path dependent (in)tangible resources,managerial and other capabilities) that change little through time, and thus need to be modeled.

14 Note, if �E + �M + �K > 1, there are increasing returns-to-scale.15 A thorough description of the approach is provided in Van Beveren (2012) and Del Gatto et al.

(2011).



regressors16 and a first-order autoregressive error term. Note, as stated above, all datawere also weighted to ensure that the samples are representative of the population of GBplants under consideration.17

4. PLANT-LEVEL RESULTS

Our main results are provided in Tables A1–A4 in the appendix. Note, all eight mod-els are deemed sufficient in terms of tests for overidentification (i.e., the Hansen testof validity of the instrument set used—where instruments for the endogenous variablescomprised lagged values18 in first differences and levels, the former being used in thelevels equation and the latter the first differenced equation of the system GMM model),and for autocorrelation (cf . the AR(1) and AR(2) test statistics).19 To reduce the likelihoodthat our estimates are biased due to the problem of overinstrumentation (see Roodman,2009), the instrument set was “collapsed.”20 This had the expected effect of reducingdramatically the P-value of the Hansen test (although not to the extent that the null ofvalid instruments was rejected). In the interests of brevity, we do not discuss the resultsrelating to those determinants of TFP which are not the main focus of this paper: namely,R&D, the age of the plant, foreign-ownership, and economies-of-scale.21 However, with re-spect to the latter, we note here that manufacturing generally benefited from increasingreturns-to-scale, while services generally experienced decreasing returns. In most sectorsboth single-plant and multiregion plants had higher TFP, especially the latter plantssuggesting that external economies-of-scale are particularly relevant. This evidence onincreasing internal- and external plant economies-of-scale provides support for NewEconomic Geography and New Trade Theory models where spatial productivity effectsarise from plant/firm level increasing returns to scale and indivisibilities in production,which interact with transport costs to provide benefits from proximity to markets andsuppliers (Fujita et al., 1999).

16 Output, intermediate inputs, labor, capital, and R&D are treated as endogenous. Note, we did nottreat the agglomeration and diversification indices as endogenous as our expectation is that individualplants are unlikely to determine such “spillover” effects in the area in which they are located, especially overa relatively short time-period such as is used in this study. However, we did experiment by instrumentingthese two variables (using their lagged values) but this had little overall difference to the results reportedhere.

17 It is worth emphasizing that we have checked to ensure that our system GMM results are verydifferent to those obtained using OLS regression, or the Olley and Pakes (1996) and Levinsohn and Petrin(2003) approaches.

18 Based on running a large number of regressions with different lags of the instruments for differentendogenous variables, no general pattern on which lag-lengths should be used was found across bothvariables and the eight models estimated, other than lagged instruments starting from t − 4 or longerwere generally needed for all endogenous variables (often longer in the case of instruments for the capitalstock). In all instances the null hypothesis for the Hansen test is that the instruments used are exogenous(i.e., that they are correlated with endogenous regressors but are not correlated with the productionfunction error term—and hence productivity); this null was accepted in all but one model (it was onlyrejected at the 10 percent level in the case of hi-tech knowledge-intensive services).

19 STATA reports tests for the first-differenced residuals, thus there should be evidence of significantnegative first order serial correlation in differenced residuals and no evidence of second order serialcorrelation in the differenced residuals, which is the case here.

20 That is, we use Roodman’s suggested approach of typically using only the lagged values of the par-ticular variable being instrumented, rather than instruments based on the lagged values of all endogenousvariables (which often leads to overinstrumentation).

21 A discussion of these results as well as a fuller justification for including these variables (involvingboth an overview of the theoretical and empirical literature in these areas) is provided in a longer versionof this paper (see Harris and Moffat, 2011). Note, our results are consistent with a priori expectations.



Tables A1–A4 give the results for the variables measuring spatial spillover and“place” effects. We generally find that localization externalities are positive in the ser-vice sectors with plants located in “clusters” benefiting from higher TFP (doubling theproportion of industry output situated in each local authority would ceteris paribus in-crease TFP by around 6 percent). With respect to the Jacobian diversification measure,urbanization economies are mostly negative, and a doubling of the proportion of indus-tries present in each area would lower TFP by 9–19 percent (depending on the sector).In high-tech manufacturing we get the opposite effect; doubling the proportion of indus-try output situated in each local authority reduces TFP by on average 3 percent, whiledoubling Jacobian spillovers increases TFP by just over 19 percent.22 Tables A1–A4 alsoshow that on average plants located in Assisted Areas (and thus eligible for industrialassistance) had around 2–4 percent lower TFP.

As to the coefficients on the region dummies, with the South East as the benchmarkregion, the (ceteris paribus) impact on TFP of being located in a particular region isgenerally significant and numerically important. Regional impacts are mostly in accordwith expectations based on historical differences between the “core” and “peripheral”regions of Great Britain (Armstrong and Taylor, 2000). Overall, and based on takinga mean and median ranking across the eight sectors, plants located in the more ruralYorkshire-Humberside region experienced the largest negative impacts on TFP (e.g., TFPwas around 9–13 percent lower in high- and low-tech manufacturing, and most servicesectors); Wales was ranked next lowest (9–13 percent lower in high-tech manufacturingand most of services); followed by the North East (8–15 percent lower in services and low-tech manufacturing); East Midlands (7–12 percent lower in most of services and high- andlow-tech manufacturing); West Midlands (10–12 percent lower in most of services); theNorth West (6–10 percent lower in services and low-tech manufacturing); and Scotland(8–20 percent lower in services). Other regions of Southern England (the South West andEastern) did a little worse overall when compared to the South East (particularly in high-tech manufacturing where TFP was around 5–7 percent lower), while London performedon a par with the South East (slightly worse in medium low-tech manufacturing at justover 2 percent lower, and much better in low-tech manufacturing at 11 percent higher, withall other sectors not significantly different). Overall regional differences were particularlymarked in the service sectors, along with high- and low-tech manufacturing.

Our primary focus is on the “core” cities; thus there is a need to take into accountsimultaneously the impacts of the “place” effects associated with the region in whichthe city is located; the “place” effect associated with each city itself; and the differentimpact of industry agglomeration, diversification, and “assisted area” effects associatedon a particular city.23 Thus, we have calculated the average TFP effect based on all

22 As stated above, we have chosen to measure agglomeration and diversification with referenceto the 381 local authority districts in Great Britain; using instead the 243 travel-to-work areas of GreatBritain produces broadly similar results to those in Tables A1–A4. See Table A5 in the appendix fordetails.

23 Overman et al. (2009) take a different approach by first estimating a model with just city andregion dummies (and no other location variables), and then introducing other spatial variables (startingwith an urbanization index, then controlling for skill mix and transport links using aggregated localauthority data). They show how the results using city dummies change as other spatial variables areincluded, but they do not “back-out” the joint effects for each city of the combined impacts of the citydummies and these other spatial variables (as we do using Equation 2). This is important, as the resultsbased on just city dummies do not give the same results as taking together the impact due to all thespatial variables simultaneously, as each city may have both positive and negative impacts associatedwith agglomeration and diversification at the subcity level, as well as differences associated with thoseparts of the city located in assisted areas.



TABLE 3: Relative Mean TFP in Core Cities, 1997–2006, Great Britain: All SpatialFactors (Differences Are Expressed as Percentages; t-values in Parenthesis)

All Manufacturing Services

City– City–rest City– City–Rest City– City–RestSouth East of Region South East of Region South East of Region

London 1.6(0.78) 2.4(0.71) 1.7(0.85)Tyneside −1.2(−0.45) 7.6(2.63) 4.0(1.14) 5.9(1.63) −1.9(−0.76) 8.4(3.03)Manchester 1.7(0.82) 9.7(3.90) 1.0(0.27) 3.6(0.95) 1.8(0.96) 11.2(4.80)Liverpool −6.3(−2.29) 1.8(0.58) −8.6(−2.01) −5.9(−1.35) −6.0(−2.34) 3.4(1.17)Birmingham −3.8(−1.97) 2.0(1.77) 1.0(0.31) 1.0(0.30) −4.6(−2.67) 2.9(1.44)Coventry 0.9(0.43) 6.8(2.67) 8.6(2.22) 8.6(2.14) −0.7(−0.35) 6.8(3.12)Leicester −5.4(−2.24) 2.3(0.81) −14.3(−3.47) −12.0(−2.78) −3.6(−1.72) 6.0(2.36)Nottingham −1.6(−0.73) 6.1(2.28) 2.1(0.47) 4.4(0.94) −2.0(−1.07) 7.5(3.14)Bristol 8.9(4.66) 10.9(4.85) 1.5(0.42) 2.1(0.53) 9.7(5.78) 12.1(6.03)Glasgow −5.5(−2.10) 8.6(2.28) 1.0(0.29) 2.3(0.60) −6.4(−2.52) 11.2(2.97)Edinburgh −10.2(−4.53) 3.9(1.11) −2.6(−0.56) −1.3(−0.26) −10.8(−5.28) 6.8(1.97)Cardiff −10.0(−4.63) −0.7(−0.22) −0.8(−0.17) 0.9(0.18) −11.1(−5.95) 0.9(0.31)

Source: see text for details (especially Equation 2).

the (underlying) spatial parameter estimates (holding constant all other effects listed inTables A1–A4), separately for plants located in each city for comparison with those plantslocated in (i) the South East; and (ii) plants located in the noncity hinterlands of the regionin which a plant is located. That is we calculate separately s indices, one for each city, therest of the region (excluding any cities), and the South East region:

TFPsit = exp

[�1 ln aggloms

it + �2 ln diverssit + �3 ln aas

it +10∑

r=1

�nregrit +

12∑c=1

�ncitycit

],(2)

where agglomit refers to the industry agglomeration variable; diversit refers to diversifi-cation; aait refers to the assisted area dummy; regit covers each regional dummy variable;cityit refers to each city dummy variable; and �i, �n, and �n are the estimated param-eters in Tables A1–A4. Note we obtain the noncity index for each region by switchingthe regional dummy “on” and the city/cities dummy “off.” For each city, we then generate100[(TFPs

it ÷ TFPSEit ) − 1] where SE refers to the South East region to generate the “city

− South East” set of results in Table 3; a similar formula replacing the South East witheach relevant noncity part of the region is used to calculate the “city − rest of region” setof results in the table. The t-values reported are obtained using the “predictnl” commandavailable in STATA; this calculates the predictions of TFPs

it based on Equation 2 and theirassociated standard errors (using the delta method).24 We report the mean values of bothTFPs

it and the associated standard errors in Table 3.The first set of results reported are for all plants in the database, while the sec-

ond and third blocks present results for all manufacturing industries25 and those ser-vice sectors we have included in this study. Only Bristol had significantly higher TFPthan the South East, and this was mostly due to higher productivity (on average nearly10 percent) in services in that city. In five other cities (including London) there was nostatistically significant difference when compared to the South East region, while in the

24 The t-values reported all have asymptotically infinite degrees-of-freedom under the student’s t-test(based on the formula applicable to a test of independent means).

25 Note, the number of observations available for manufacturing plants in the cities covered wasmuch lower than those available for services; hence, the lower t-values associated with the manufacturingresults.



TABLE 4: Relative Mean TFP in Core Cities, 1997–2006, Great Britain: Agglomerationand Diversification Only (Differences Are Expressed as Percentages; t-values in

Parenthesis)

All Manufacturing Services

City– City–rest City– City–Rest City– City–RestSouth East of Region South East of Region South East of Region

London 1.7(1.04) −0.6(−0.18) 1.9(1.25)Tyneside 1.4(0.96) 1.4(0.92) 3.5(1.18) −0.1(−0.03) 1.1(0.89) 1.9(1.54)Manchester 2.6(1.71) 2.6(1.72) 1.2(0.46) −0.7(−0.27) 2.8(1.98) 3.2(2.44)Liverpool 2.1(1.32) 2.1(1.33) 0.1(0.04) −1.8(−0.69) 2.3(1.64) 2.7(2.05)Birmingham 3.3(2.01) 3.0(1.82) 5.5(1.90) 2.2(0.75) 2.9(2.03) 3.5(2.57)Coventry 1.5(1.02) 1.2(0.82) 5.4(1.82) 2.1(0.70) 0.6(0.55) 1.2(1.15)Leicester 1.2(0.82) 1.7(1.01) 2.8(0.91) 0.8(0.23) 0.8(0.71) 2.1(1.77)Nottingham 1.8(1.21) 2.3(1.34) 1.8(0.64) −0.2(−0.07) 1.8(1.39) 3.1(2.30)Bristol 2.0(1.38) 2.7(1.76) 1.2(0.46) 0.6(0.20) 2.0(1.64) 3.0(2.37)Glasgow 3.3(1.82) 3.8(1.97) 1.1(0.37) 1.2(0.37) 3.6(2.15) 4.3(2.54)Edinburgh 2.6(1.50) 3.2(1.68) −1.2(−0.35) −1.1(−0.29) 2.9(1.87) 3.6(2.29)Cardiff 1.5(1.06) 2.1(1.28) 0.3(0.10) −2.1(−0.66) 1.7(1.37) 3.3(2.61)

Source: see text for details (especially Equation 2 with �3 = �n = �n = 0).

remaining six cities TFP was lower (in Edinburgh and Cardiff the “gap” was around 10percent). The largest negative differences are caused by large “gaps” in the service sector(these also “explain” the poorer performance in Birmingham and Glasgow); but for citieslike Liverpool and Leicester poorer performance is associated relatively more with man-ufacturing rather than services (although services also have lower average TFP in thesecities). Lastly, in Coventry average TFP in manufacturing was nearly 9 percent higherthan in the South East, but the city did less well in services so that across all plants therewas no significant difference in performance.

With regard to whether cities had on average higher TFP than their (noncity) hin-terlands, Table 3 shows that overall this was mostly the case; although in Liverpool andCardiff there were no statistically significant differences across either sector, while inEdinburgh and especially Leicester higher TFP in services (of 6.8 percent and 6 percent,respectively) was not sufficient to overcome the poorer manufacturing performance inthese cities and give an overall positive and significant difference.

It is also possible to decompose the city-region differentials in Table 3 into a com-ponent due to industry agglomeration and diversification economies (the first two el-ements on the right-hand-side in Equation 2), separately from all other unmeasuredspatial characteristics of location (the dummy variables in Equation 2). This shows howmuch MAR- and Jacobian-economies account for differences in TFP across cities, andthus provides information on the degree to which such localization and diversificationeffects (as measured) assist with understanding differences in the economic prospectsof different spatial areas. The “city − South East” set of results in Table 4, coveringall of manufacturing and services, indicate that overall cities benefit from localizationand diversification economies, sufficient to offset (to some degree) the overall negativeTFP differential between the cities and the SE region (as shown in Table 3). For exam-ple, the overall negative differential for Birmingham in Table 3 of −3.8 percent is setagainst the 3.3 percent positive differential in Table 4, based on industry agglomerationand diversification economies alone, showing that plants experienced positive measuredspatial spillovers associated with being located in Birmingham but that other unmea-sured spatial effects (included in Equation 2) were much stronger. In fact, given that the



Birmingham dummy variable in Tables A1–A4 is often (significantly) positive, this pointsto the dominant negative characteristics of the West Midlands region as the main sourceof plants in Birmingham having lower TFP versus the South East of England.

With respect to the overall “city − rest of region” set of results in Table 4, theseshow that in general measured industry agglomeration and diversification economiesbenefited the cities relative to the rest of region in which they are located. However, theywere mostly not the dominant source of the positive “city − rest of region” differential(Table 3), except in Leicester, Birmingham, Liverpool, and Edinburgh.

Using the South East as the benchmark, Table 4 also shows that much of the offsettingbenefits due to measured spatial economies for the cities were concentrated in the servicesector, rather than in manufacturing; only in Birmingham and Coventry (both of whichare located in the West Midlands region) were industry agglomeration and diversificationeffects significantly positive in manufacturing (with a “city − South East” differentialof around 5.5 percent). In services, measured spatial effects were significantly positiveand helped to offset the overall negative differential with the South East in six cities(e.g., Liverpool, Birmingham, Glasgow, and Edinburgh), with a further two (Nottinghamand Cardiff) significant only at the 17 percent level (under a two-tailed test). However,industry agglomeration and diversification as measured were generally not the mainsource of the overall positive “city − rest of region” differentials in services in Table 3,except in Liverpool, Birmingham, and Cardiff; rather the unmeasured positive benefitsof being in a city (captured by the city dummies in Tables A1–A4) account for most of theobserved differentials in services.

Detailed results for the “city − South East” and “city − rest of region” differentialsby each of the eight industrial subsectors covered in this study are available—althoughspace limitations and the specific nature of the detailed results provided confine these toan unpublished appendix that is available from the authors. The more detailed resultsgenerally provide confirmation of those available in Table 3, in addition to information onwhich of the eight sectors had the largest positive externalities from city locations (vis-a-vis noncity locations). We find that plants in services (especially knowledge-intensivemarket services, such as services to the business sector) benefited most from spatialeffects, while the results for high-tech manufacturing were overall negative. Our resultsalso show that vis-a-vis the South East, city locations are particularly associated withlower TFP in high-tech manufacturing and much of services.

Based on the results in Tables 3, our overall conclusion is that, with the exceptionof Bristol, the core cities of Great Britain did not have significantly higher TFP whencompared to the South East region (our frontier benchmark), but they did tend to havehigher productivity when compared to their noncity hinterlands (especially with respect toTFP in services). However, we also found that overall cities benefit from (ceteris paribus)localization and diversification economies (especially in the service sector), sufficient tooffset (to some degree) the overall negative TFP differential between the cities and theSouth East region. As expected, our results show that not all cities do equally as well (e.g.,Leicester vs. Nottingham; Liverpool vs. Manchester; Edinburgh vs. Glasgow), and indeedthere is no overwhelming evidence from this study in support of British cities being theideal locations for encouraging growth, particularly in high-technology industries; espe-cially as diversification (or urbanization) economies (as opposed to localization economies)were largely negative.

5. CONCLUSION

This paper has examined the determinants of total factor productivity (TFP) usinga GB plant-level data set. The main findings relate to whether spatial spillovers and



“place” effects are important, and more particularly whether (ceteris paribus) plantslocating in the core cities of Great Britain have a productivity advantage over thosethat locate elsewhere. Our results are conditional on having controlled for the impact ofR&D, technical change, plant age, foreign ownership, multiplant economies of scale andcompetition. Estimates were obtained using system GMM as this allows for both fixedeffects and endogenous regressors. The sample was disaggregated into manufacturingand services and by technology to show whether different sectors perform differently.

Our main results are as follows: first, plants belonging to multiplant enterprisesoperating in more than one region generally had (ceteris paribus) higher TFP levels.This suggests that economies of scale arising from membership of a multiplant enterprisemay only become important over a large geographical area (and where supplying nearbymarkets is important). The evidence presented here therefore provides a slightly differentexplanation for agglomeration than that used by new economic geography models (e.g.,Krugman, 1991; Fujita et al., 1999) which rely on pecuniary26 rather than productionexternalities to explain agglomeration.

The industry agglomeration variable used in this study is positive and significantfor three of the four service sectors. However, of the four manufacturing sectors, industryagglomeration is only significantly positive for medium high-tech manufacturing. Thediversification measure is negatively associated with TFP for most sectors although thisassociation is only significant for three out of eight sectors. As suggested by Baldwinet al. (2010), who obtained a similar result, this may indicate that (inter alia) congestiondiseconomies are important. As expected, plants situated in an assisted area have lowerTFP for all sectors. The regional rankings of TFP are broadly in line with expectations,with plants in the South East and London generally experiencing a significant produc-tivity advantage associated with being located in these regions. Our results also suggestthat plants located in “core” cities generally perform better than plants in the same regionoutside of these cities; but with the exception of Bristol, no city has significantly higherTFP levels than the South East which suggests that the spatial externalities associatedwith city location are not as important as the benefits of being situated in the South East.The reasons for the dominance of this region clearly deserve more research.

According to a recent U.K. Department of Business, Enterprise and Regulatory Re-form (BERR, 2008)27 policy document: “larger, more diversified cities tend to be betterplaced to provide the flexibility required to take advantage of the opportunities and thechallenges of globalization and the knowledge-driven economy. However, specialized citiescan bring significant value to those industries that benefit from localization (includinginput sharing) and clustering with firms in the same sector.” This implies government ac-ceptance of the importance of both types of agglomeration externalities. This conclusion isstrengthened, specifically in relation to localization externalities, by the fact that the U.K.government has also attempted “to create the conditions that encourage the formationand growth of clusters” (BERR, 2008). For those industries with positive coefficients onthe industry agglomeration variable, this study would support such policies to encourageclusters, although the government should also be aware of the potential for congestiondiseconomies. However, overall the results presented in this paper do not provide supportfor policies that de facto accept as given that plants located in cities will do better. Rather,further research is required to enable the interpretation of the sometimes large differencesin the coefficient on the region and city dummies as this will allow the development ofpolicy to allow regions and cities to emulate the best performers. Research is also needed

26 Pecuniary externalities refer to the benefits which arise to workers due to being close to producersbecause the goods they consume are not subject to large transport costs.

27 BERR is now the Department of Business, Innovation and Skills (BIS).



that extends our results to include other definitions of “cities” to see to what extent theyhold when we include urban areas with smaller populations and/or lower density levels.For now our results point to the conclusion that the South East is still the “place” to be tobenefit from higher TFP.

APPENDIX

TABLE A1: Long-Run Weighted Systems GMM Production Function, High-Techand Medium High-Tech Sectorsa, 1997–2006

High-Tech Medium high-Tech

Dependent variable: ln gross outputt � z-statistic � z-statistic

ln intermediate inputst 0.438∗∗∗ 5.15 0.377∗∗∗ 6.15ln employmentt 0.427∗∗∗ 4.43 0.192∗∗∗ 3.93ln capitalt 0.191∗∗∗ 3.37 0.287∗∗∗ 10.73t 0.048∗∗∗ 8.59 0.019∗∗∗ 15.51ln aget −0.187∗∗∗ −3.56 −0.184∗∗∗ −11.69Single-plant enterpriset 0.077∗∗∗ 3.79 0.013∗ 1.69Enterprise operates in >1 regiont 0.089∗∗∗ 2.98 0.068∗∗∗ 4.63Greenfield U.S.-ownedt 0.165∗∗∗ 3.13 0.042∗ 1.84Brownfield U.S.-ownedt 0.122∗∗∗ 2.93 0.118∗∗∗ 4.08Greenfield EU-ownedt 0.106∗∗ 2.07 0.077∗∗∗ 4.13Brownfield EU-ownedt 0.084∗ 1.69 0.099∗∗∗ 5.71Greenfield Other FOt 0.131∗ 1.64 0.125∗∗∗ 4.39Brownfield Other FOt 0.080 1.24 0.118∗∗∗ 5.38ln Industry agglomerationt −0.034 −1.59 0.065∗∗∗ 7.57ln Diversificationt 0.193∗∗ 2.23 −0.094∗∗∗ −6.22ln Herfindahlt 0.015 1.45 −0.025∗∗∗ −6.64R&D undertakent 0.041∗∗ 2.20 0.108∗∗∗ 4.42Located in Assisted Areat −0.010 −0.49 −0.025∗∗∗ −7.34North Eastt −0.052∗∗∗ −1.31 −0.024∗∗∗ −3.86Yorks-Humbersidet −0.097∗∗∗ −2.86 −0.051∗∗∗ −10.02North Westt −0.054 −1.47 −0.049∗∗∗ −10.45West Midlandst −0.073∗∗∗ −2.67 −0.044∗∗∗ −8.03East Midlandst −0.114∗∗∗ −3.39 −0.030∗∗∗ −4.79South Westt −0.070∗∗∗ −2.80 −0.010 −1.49Easternt −0.051∗∗ −2.32 0.018∗∗∗ 4.03Londont 0.025 0.83 0.003 0.48Scotlandt 0.019 0.58 −0.031∗∗∗ −6.13Walest −0.096∗∗∗ −3.09 −0.056∗∗∗ −8.56Tynesidet −0.137∗ −1.78 0.091∗∗∗ 7.50Manchestert 0.014 0.18 0.041∗∗∗ 3.44Liverpoolt 0.080 1.19 0.031∗∗ 2.19Birminghamt −0.035 −0.62 −0.020∗∗ −1.99Coventryt 0.045 0.79 0.086∗∗∗ 5.92Leicestert −0.187 −2.45 −0.170∗∗∗ −8.03Nottinghamt −0.167 −1.05 0.105∗∗∗ 6.81Bristolt −0.015 −0.22 0.009 0.81Glasgowt 0.026 0.41 −0.018∗ −1.81Edinburght −0.112∗ −1.64 −0.047∗∗∗ −3.16

Continued



TABLE A1: Continued

High-Tech Medium high-Tech


Cardifft −0.095 −1.00 0.074∗∗∗ 5.11

Industry dummies Yes YesAR(1) z-statistic −6.71∗∗∗ −26.76∗∗∗

AR(2) z-statistic 1.19 −0.99Hansen test � 2 (df) 27.60(23) 22.41(15)No. of Obs. 12,906 40,834No. of groups 5,386 15,957

a See Table 2 for definition.∗∗∗significant at 1 percent, ∗∗significant at 5 percent, ∗significant at 10 percent.

TABLE A2: Long-Run Weighted Systems GMM Production Function, Medium Low-Techand Low-Tech Sectorsa, 1997–2006

High-Tech Medium High-Tech


ln intermediate inputst 0.451∗∗∗ 2.64 0.500∗∗∗ 5.72ln employmentt 0.423∗∗∗ 2.76 0.401∗∗∗ 4.16ln capitalt 0.141∗∗∗ 3.97 0.123∗∗∗ 2.77t 0.023∗∗∗ 4.59 0.023∗∗∗ 6.09ln aget −0.140∗∗∗ −4.74 −0.154∗∗∗ −5.82Single-plant enterpriset 0.080∗∗∗ 4.41 0.116∗∗∗ 8.18Enterprise operates in >1 regiont 0.120∗ 1.64 0.134∗∗∗ 2.71Greenfield U.S.-ownedt 0.024 0.56 0.117∗∗∗ 4.61Brownfield U.S.-ownedt 0.033 1.22 0.077∗ 1.75Greenfield EU-ownedt 0.066 0.91 0.165∗∗∗ 2.56Brownfield EU-ownedt 0.042∗∗ 2.03 0.203∗∗∗ 3.18Greenfield Other FOt 0.306∗∗ 2.10 0.016 0.27Brownfield Other FOt −0.024 −0.93 0.026 0.72ln Industry agglomerationt −0.036∗∗ −2.26 −0.005 −0.19ln Diversificationt −0.185∗∗ −2.11 0.035 0.81ln Herfindahlt 0.021 0.87 0.034∗∗ 2.23R&D undertakent 0.001 0.08 0.006 0.07Located in Assisted Areat −0.026∗∗∗ −4.28 −0.047∗∗∗ −4.02North Eastt −0.043∗∗ −2.08 −0.095∗∗∗ −4.11Yorks-Humbersidet 0.065∗∗∗ 2.69 −0.139∗∗∗ −2.78North Westt 0.048∗∗∗ 2.76 −0.067∗∗∗ −2.17East Midlandst −0.010 −0.66 −0.105∗∗∗ −3.92South Westt 0.022∗∗∗ 2.64 −0.005 −0.43Easternt −0.014 −0.93 −0.000 −0.03Londont −0.023∗ −1.81 0.103∗∗∗ 4.52Scotlandt 0.077∗∗∗ 4.57 0.006 0.35Walest 0.022 1.54 −0.055∗∗∗ −3.21Tynesidet 0.075∗∗∗ 3.48 0.144∗∗∗ 4.97Manchestert 0.039 1.44 0.075∗∗∗ 3.52Liverpoolt −0.051 −0.76 −0.097∗∗ −2.41Birminghamt 0.062∗∗∗ 2.63 0.099∗∗∗ 4.75Coventryt 0.076 1.45 0.335∗∗∗ 6.14Leicestert −0.052∗ −1.67 −0.061∗∗∗ −2.86

Continued



TABLE A2: Continued

High-Tech Medium High-Tech


Nottinghamt 0.108∗∗ 2.52 0.051 1.41Bristolt 0.115∗∗ 1.97 −0.040∗ −1.88Glasgowt 0.102∗∗∗ 3.64 0.074∗∗∗ 3.91Edinburght −0.066∗∗ −2.06 0.105∗∗∗ 4.78Cardifft 0.107∗∗∗ 5.14 −0.002 −0.06


AR(2) z-statistic 0.14 −1.09Hansen test � 2 (df) 9.34(6) 10.37(6)No.of Obs. 14,218 24,096No.of groups 4,854 7,750

aSee Table 2 for definition∗∗∗significant at 1 percent, ∗∗significant at 5 percent, ∗significant at 10 percent.

TABLE A3: Long-Run Weighted Systems GMM Production Function, High-TechKnowledge Intensive and Knowledge-Intensive Market Sectorsa, 1997–2006

High-Tech KI KI Market


ln intermediate inputst 0.420∗∗∗ 4.52 0.441∗∗∗ 7.72ln employmentt 0.466∗∗∗ 6.53 0.390∗∗∗ 5.79ln capitalt 0.060∗∗∗ 4.10 0.096∗∗∗ 2.59t 0.006∗∗∗ 2.57 0.005∗∗∗ 3.10ln aget −0.173∗∗∗ −8.39 −0.042 −1.19Single-plant enterpriset 0.021 0.41 0.105∗∗∗ 6.91Enterprise operates in >1 regiont −0.137 −1.60 0.211∗∗∗ 6.66Greenfield U.S.-ownedt 0.194∗∗∗ 13.47 0.094∗∗ 2.01Brownfield U.S.-ownedt 0.356∗∗∗ 9.95 0.044 1.45Greenfield EU-ownedt 0.175∗∗∗ 2.35 0.159∗∗ 2.51Brownfield EU-ownedt 0.183∗∗∗ 5.87 0.008 0.24Greenfield Other FOt 0.309∗∗∗ 3.60 0.174∗∗ 1.99Brownfield Other FOt −0.034∗∗∗ −2.75 0.200∗∗∗ 2.91ln Industry agglomerationt 0.058∗ 1.77 0.068∗∗∗ 6.96ln Diversificationt −0.167 −1.37 −0.168∗∗∗ −5.19ln Herfindahlt 0.092∗∗∗ 5.39 −0.007 −0.35R&D undertakent 0.225∗∗ 2.45 0.120∗ 1.65Located in Assisted Areat −0.030∗∗∗ −3.93 −0.035∗∗∗ −5.26North Eastt −0.169∗∗∗ −6.48 −0.144∗∗∗ −9.35Yorks-Humbersidet −0.127∗∗∗ −7.33 −0.111∗∗∗ −8.59North Westt −0.099∗∗∗ −5.86 −0.108∗∗∗ −11.61West Midlandst −0.112∗∗∗ −6.42 −0.106∗∗∗ −8.32East Midlandst −0.122∗∗∗ −6.33 −0.126∗∗∗ −8.62South Westt −0.012 −1.32 −0.027∗∗∗ −2.58Easternt −0.021∗∗ −2.43 −0.016∗∗ −2.02Londont −0.007 −0.57 0.008 0.51Scotlandt −0.218∗∗∗ −5.97 −0.083∗∗∗ −5.65Walest −0.143∗∗∗ −7.57 −0.120∗∗∗ −6.90Tynesidet 0.143∗∗∗ 5.41 0.158∗∗∗ 6.26

Continued



TABLE A3: Contnued

High-Tech KI KI Market


Manchestert 0.021 0.90 0.103∗∗∗ 6.98Liverpoolt 0.031∗∗ 2.19 0.119∗∗∗ 5.59Birminghamt 0.035∗∗ 2.27 0.058∗∗∗ 3.96Coventryt 0.086∗∗∗ 5.92 0.023 0.87Leicestert −0.170∗∗∗ −8.03 0.124∗∗∗ 4.85Nottinghamt 0.105∗∗∗ 6.81 0.049∗∗∗ 2.61Bristolt 0.030 1.15 −0.019 −1.37Glasgowt −0.037∗ −1.66 0.012 0.76Edinburght 0.045∗∗ 2.50 −0.087∗∗∗ −5.34Cardifft −0.090∗∗ −1.96 0.048∗∗∗ 2.97


AR(2) z-statistic −0.77 1.33Hansen test � 2 (df) 7.41∗(3) 8.27(6)No. of Obs. 32,971 27,995No. of groups 12,696 13,319

a See Table 2 for definition∗∗∗significant at 1 percent, ∗∗significant at 5 percent, ∗∗significant at 10 percent.

TABLE A4: Long-Run Weighted Systems GMM Production Function, LowKnowledge-Intensive Market Services and Other Low Knowledge-Intensive Services

Sectors,a 1997–2006

Low KI Other Low KI


ln intermediate inputst 0.821∗∗∗ 19.95 0.722∗∗∗ 68.61ln employmentt 0.159∗∗∗ 2.71 0.179∗∗∗ 3.30ln capitalt 0.043∗∗∗ 2.92 0.048∗∗∗ 3.95t −0.004∗∗ −2.46 −0.000 −0.26ln aget −0.062∗∗∗ −3.00 −0.070∗∗∗ −6.88Single-plant enterpriset −0.023 −0.68 0.425∗∗∗ 7.60Enterprise operates in >1 regiont −0.033 −0.73 0.378∗∗∗ 13.40Greenfield U.S.-ownedt −0.071∗∗∗ −4.37 0.042 1.16Brownfield U.S.-ownedt −0.092∗∗∗ −16.28 −0.152∗∗∗ −14.49Greenfield EU-ownedt 0.067∗∗ 2.04 0.113∗∗ 2.09Brownfield EU-ownedt −0.050 −1.01 0.117∗∗ 2.44Greenfield Other FOt −0.185∗∗ −2.00 −0.033 −0.52Brownfield Other FOt −0.124∗∗∗ −2.81 0.179∗ 1.80ln Industry agglomerationt 0.003 0.58 0.056∗∗∗ 2.65ln Diversificationt 0.002 0.10 −0.004 −0.14ln Herfindahlt −0.099∗∗∗ −29.20 0.012∗∗∗ 3.93R&D undertakent 0.153∗ 1.75 0.447∗∗∗ 5.58Located in Assisted Areat −0.013∗∗∗ −8.31 0.049∗∗∗ 10.47North Eastt −0.085∗∗∗ −9.80 −0.086∗∗∗ −5.95Yorks-Humbersidet −0.092∗∗∗ −49.47 −0.138∗∗∗ −6.89North Westt −0.089∗∗∗ −6.73 −0.066∗∗∗ −5.82West Midlandst −0.038∗∗∗ −5.63 −0.130∗∗∗ −13.85

Continued



TABLE A4: Continued

Low KI Other Low KI


East Midlandst −0.077∗∗∗ −6.97 −0.051∗∗∗ −7.21South Westt −0.000 −0.10 −0.042∗∗∗ −6.11Easternt −0.026∗∗∗ −5.02 −0.050∗∗∗ −13.92Londont −0.001 −0.35 −0.015 −0.94Scotlandt −0.144∗∗∗ −6.40 −0.194∗∗∗ −6.58Walest −0.109∗∗∗ −5.75 −0.047∗∗∗ −3.53Tynesidet 0.135∗∗∗ 4.43 −0.179∗∗∗ −10.72Manchestert 0.087∗∗∗ 7.59 0.023∗ 1.81Liverpoolt 0.005 0.81 −0.033 −1.05Birminghamt 0.001 0.09 −0.066∗∗ −2.37Coventryt 0.058∗∗∗ 10.16 0.066∗∗∗ 6.67Leicestert 0.067∗∗∗ 6.40 −0.021 −1.13Nottinghamt 0.048∗∗∗ 3.11 0.007 0.53Bristolt 0.137∗∗∗ 7.72 −0.021∗ −1.74Glasgowt 0.055∗∗∗ 12.08 0.089∗∗∗ 5.84Edinburght 0.026∗ 1.82 0.064∗∗∗ 9.06Cardifft −0.023∗∗ −2.29 −0.042∗∗∗ −4.76

Industry dummies yes yesAR(1) z-statistic −6.80∗∗∗ −5.73∗∗∗

AR(2) z-statistic −1.50 −1.18Hansen test � 2 (df) 3.24(3) 2.30(3)No. of Obs. 351,721 91,942No. of groups 131,150 32,975

a See Table 2 for definition.∗∗∗significant at 1 percent, ∗∗significant at 5 percent, ∗significant at 10 percent.

TABLE A5: Long-Run Weighted Systems GMM Production Function Estimates forAgglomeration and Diversification, 1997–2006

Local Authoritiesa Travel-to-Work Areasb

Spatial Unit � z-statistic � z-statistic

High-tech manufacturingln Industry agglomerationt −0.034 −1.59 0.013 1.24ln Diversificationt 0.193 2.62 0.043 1.34

Medium high-tech manufacturingln Industry agglomerationt 0.065 7.57 0.079 11.13ln Diversificationt −0.094 −6.22 −0.088 −7.57

Medium low-tech manufacturingln Industry agglomerationt −0.036 −2.26 0.025 1.56ln Diversificationt −0.185 −2.11 −0.176 −1.92

Low-tech manufacturingln Industry agglomerationt −0.005 −0.19 0.012 0.41ln Diversificationt 0.035 0.81 0.003 0.06

High-tech KI servicesln Industry agglomerationt 0.058 1.77 0.076 1.95ln Diversificationt −0.167 −1.37 −0.248 −2.33

Continued



TABLE A5: Continued

Local Authoritiesa Travel-to-Work Areasb

Spatial Unit � z-statistic � z-statistic

KI market servicesln Industry agglomerationt 0.068 6.96 0.059 8.83ln Diversificationt −0.168 −5.19 −0.200 −8.87

Low KI servicesln Industry agglomerationt 0.003 0.58 0.001 1.16ln Diversificationt 0.002 0.10 −0.002 −0.12

Other low KI servicesln Industry agglomerationt 0.056 2.65 0.046 2.45ln Diversificationt −0.004 −0.14 −0.006 −1.02

aEstimates taken from Tables A1–A4.bObtained from estimating Equation (1) with travel-to-work areas used to define agglomeration and diversi-

fication. Other parameter estimates not shown but (often very) similar to those shown in Tables A1–A4.

REFERENCESAberg, Yngve. 1973. “Regional Productivity Differences in Swedish Manufacturing,” Regional and Urban

Economics, 3, 131–155.Ackerberg, D. A., K. Caves, and G. Frazer. 2006. “Structural Identification of Production Functions.” Los Angeles,

UCLA. Available at: http://www.econ.ucla.edu/ackerber/ACF20withtables.pdf.Acs, Zoltan J., Luc Anselin, and Attila Varga. 2002. “Patents and Innovation Counts as Measures of Regional

Production of New Knowledge,” Research Policy, 31, 1069–1085.Andersson, Martin and Hans Loof. 2011. “Agglomeration and Productivity: Evidence from Firm-Level Data,” The

Annals of Regional Science, 46, 601–620.Anselin, Luc, Attila Varga, and Zoltan Acs. 1997. “Local Geographic Spillovers between University Research and

High Technology Innovations,” Journal of Urban Economics, 42, 422–448.Armstrong, Harvey and Jim Taylor. 2000. Regional Economics and Policy. Oxford: Blackwell.Arrow, Kenneth J. 1962. “The Economic Implications of Learning by Doing,” The Review of Economic Studies, 29,

155–173.Audretsch, David B. and Maryann P. Feldman. 1996. “R&D Spillovers and the Geography of Innovation and

Production,” American Economic Review, 86, 630–640.Baily, Martin Neil, Charles Hulten, and David Campbell. 1992. “Productivity Dynamics in Manufacturing Plants,”

Brookings Papers on Economic Activity, 1992, 187–249.Baldwin, John R., W. Mark Brown, and David L. Rigby. 2010. “Agglomeration Economies: Microdata Panel

Estimates from Canadian Manufacturing,” Journal of Regional Science, 50, 915–934.Bartelsman, Eric J. and Phoebus J. Dhrymes. 1998. “Productivity Dynamics: U.S. Manufacturing Plants,

1972–1986,” Journal of Productivity Analysis, 9, 5–34.Baumol, William. 1984. “On Productivity Growth in the Long Run,” Atlantic Economic Journal, 12, 4–10.BERR. 2008. “BERR’s Role in Raising Productivity: New Evidence,” BERR economics paper. London. Available

at: http://www.bis.gov.uk/files/file44504.pdf.Blundell, Richard and Stephen Bond. 1998. “Initial Conditions and Moment Restrictions in Dynamic Panel Data

Models,” Journal of Econometrics, 87, 115–143.Brulhart, Marius and Nicole A. Mathys. 2008. “Sectoral Agglomeration Economies in a Panel of European Re-

gions,” Regional Science and Urban Economics, 38, 348–362.Capello, Roberta. 2002. “Entrepreneurship and Spatial Externalities: Theory and Measurement,” The Annals of

Regional Science, 36, 387–402.Ciccone, Antonio. 2002. “Agglomeration Effects in Europe,” European Economic Review, 46, 213–227.———. and Robert E. Hall. 1996. “Productivity and the Density of Economic Activity,” The American Economic

Review, 86, 54–70.Combes, Pierre, Gilles Duranton, Laurent Gobillon, Diego Puga, and Sebastien Roux. 2009. “The Productivity

Advantages of Large Cities: Distinguishing Agglomeration from Firm Selection,” SERC Discussion Papers 27,SERC.

———, ———, ———, and Sebastien Roux. 2008. “Estimating Agglomeration Economies with History, Geology,and Worker Effects,” in E. L. Glaeser (ed.), Agglomeration Economics. Chicago: University of Chicago Press,pp. 15–65.



Corrado, Luisa and Bernard Fingleton. 2012. “Where is the Economics in Spatial Econometrics?,” Journal ofRegional Science, 52, 210–239.

Crepon, Bruno, Emmanuel Duguet, and Jacques Mairesse. 1998. “Research, Innovation and Productivity: AnEconometric Analysis at the Firm Level,” Economics of Innovation and New Technology, 7, 115–158.

Del Gatto, Massimo, Adriana Di Liberto, and Carmelo Petraglia. 2011. “Measuring Productivity,” Journal ofEconomic Surveys, 25, 952–1008.

Devereux, Michael P., Rachel Griffith, and Helen Simpson. 2007. “Firm Location Decisions, Regional Grants andAgglomeration Externalities,” Journal of Public Economics, 91, 413–435.

Duranton, Gilles and Diego Puga. 2004. “Micro-foundations of Urban Agglomeration Economies,” in J. V.Henderson and T. Jacques-Francois (eds.), Handbook of Regional and Urban Economics. Amsterdam; Oxford:Elsevier, pp. 2063–2117.

Ellison, Glenn, Edward L. Glaeser, and William R. Kerr. 2010. “What Causes Industry Agglomeration? Evidencefrom Coagglomeration Patterns,” American Economic Review, 100, 1195–1213.

Fingleton, Bernard. 2003. “Increasing Returns: Evidence from Local Wage Rates in Great Britain,” Oxford Eco-nomic Papers, 55, 716–739.

Fujita, Masahisa, Paul Krugman, and Tomoya Mori. 1999. “On the Evolution of Hierarchical Urban Systems,”European Economic Review, 43, 209–251.

Gertler, Meric S. 2003. “Tacit Knowledge and the Economic Geography of Context, or the Undefinable Tacitnessof Being (There),” Journal of Economic Geography, 3, 75–99.

Graham, Daniel J. 2009. “Identifying Urbanization and Localization Externalities in Manufacturing and ServiceIndustries,” Papers in Regional Science, 88, 63–84.

———. Patricia S. Melo, Piyapong Jiwattanakulpaisarn, and Robert B. Noland. 2010. “Testing for Causalitybetween Productivity and Agglomeration Economies,” Journal of Regional Science, 50, 935–951.

Griffith, Rachel. 1999. “Using the ARD Establishment Level Data to Look at Foreign Ownership and Productivityin the United Kingdom,” Economic Journal, 109, F416–42.

Harris, Richard. 2002. “Foreign Ownership and Productivity in the United Kingdom—Some Issues When Usingthe ARD Establishment Level Data,” Scottish Journal of Political Economy, 49, 318–335.

———. 2005. “Economics of the Workplace: Special Issue Editorial,” Scottish Journal of Political Economy, 52,323–343.

———. and Stephen Drinkwater. 2000. “UK Plant and Machinery Capital Stocks and Plant Closures,” OxfordBulletin of Economics and Statistics, 62, 243–265.

———, Qian Cher Li, and Mary Trainor. 2009. “Is a Higher Rate of R&D Tax Credit a Panacea for Low Levels ofR&D in Disadvantaged Regions?,” Research Policy, 38, 192–205.

———. and John Moffat. 2011. “Plant-level Determinants of Total Factor Productivity in Great Britain,1997–2006,” SERC Discussion Paper 64, SERC.

———, John Moffat, and Victoria Kravtsova. 2011. “In Search of ‘W’,” Spatial Economic Analysis, 6, 249–270.———. and Catherine Robinson. 2002. “The Effect of Foreign Acquisitions on Total Factor Productivity: Plant-

Level Evidence from U.K. Manufacturing, 1987–1992,” The Review of Economics and Statistics, 84, 562–568.Haskel, Jonathan. 2000. “What Raises Productivity? The Microeconomics of UK Productivity Growth,” Queen

Mary, University of London.Henderson, J. Vernon. 1986. “Efficiency of Resource Usage and City Size,” Journal of Urban Economics, 19, 47–70.———. 2003. “Marshall’s Scale Economies,” Journal of Urban Economics, 53, 1–28.Jacobs, Jane. 1970. The Economy of Cities. London: Jonathan Cape.———. 1986. Cities and the Wealth of Nations. Harmondsworth: Penguin.Kelejian, Harry. 2008. “A Spatial J-Test for Model Specification against a Single or a Set of Non-nested Alterna-

tives,” Letters in Spatial and Resource Sciences, 1, 3–11.Kominers, Scott Duke. 2008. “Measuring Agglomeration.” Available at: http://www.scottkom.com/articles/

measure_agglomeration.pdf.Krugman, Paul. 1991. “Increasing Returns and Economic Geography,” Journal of Political Economy, 99, 483–499.———. 1997. The Age of Diminished Expectations: U.S. Economic Policy in the 1990s. Cambridge, Mass.; London:

MIT Press.LeSage, James P. and C. Robert Pace. 2010. “The Biggest Myth in Spatial Econometrics.” Available at:

http://dx.doi.org/10.2139/ssrn.1725503.Levinsohn, J. and A. Petrin. 2003. “Estimating Production Functions Using Inputs to Control for Unobservables,”

The Review of Economic Studies, 70, 317–341.Marshall, Alfred. 1890. Principles of Economics. London: Macmillan.Martin, Philippe, Thierry Mayer, and Florian Mayneris. 2011. “Spatial Concentration and Plant-level Productivity

in France,” Journal of Urban Economics, 69, 182–195.Martin, Ralf. 2008. “Productivity Dispersion, Competition and Productivity Measurement,” CEP Discussion Paper

692, CEP.



Melitz, Marc J. and Gianmarco I. P. Ottaviano. 2008. “Market Size, Trade, and Productivity,” Review of EconomicStudies, 75, 295–316.

Moomaw, Ronald L. 1983. “Is Population Scale a Worthless Surrogate for Business Agglomeration Economies?,”Regional Science and Urban Economics, 13, 525–545.

Nakamura, Ryohei. 1985. “Agglomeration Economies in Urban Manufacturing Industries: A Case of JapaneseCities,” Journal of Urban Economics, 17, 108–124.

Olley, G. Steven and Ariel Pakes. 1996. “The Dynamics of Productivity in the Telecommunications EquipmentIndustry,” Econometrica, 64, 1263–1297.

Oulton, Nicholas. 1997. “The ABI Respondents Database: A New Resource for Industrial Economics Research,”Economic Trends, 528, 46–57.

Overman, Henry, Steve Gibbons, and Alessandra Tucci. 2009. “The Case for Agglomeration Economies,”Manchester Independent Economics Review. Manchester. Available at: http://www.manchester-review.org.uk/projects/view/?id = 718.

Puga, Diego. 2010. “The Magnitude and Causes of Agglomeration Economies,” Journal of Regional Science, 50,203–219.

Rice, Patricia, Anthony J. Venables, and Eleonora Patacchini. 2006. “Spatial Determinants of Productivity: Anal-ysis for the Regions of Great Britain,” Regional Science and Urban Economics, 36, 727–752.

Rigby, David L. and Jurgen Essletzbichler. 2002. “Agglomeration Economies and Productivity Differences in USCities,” Journal of Economic Geography, 2, 407–432.

Romer, Paul M. 1986. “Increasing Returns and Long-run Growth,” Journal of Political Economy, 94, 1002–1037.Roodman, David. 2009. “A Note on the Theme of Too Many Instruments,” Oxford Bulletin of Economics and

Statistics, 71, 135–158.Rosenthal, Stuart S. and William C. Strange. 2004. “Evidence on the Nature and Sources of Agglomeration

Economies,” in J. V. Henderson and T. Jacques-Francois (eds.), Handbook of Regional and Urban Economics.Amsterdam; Oxford: Elsevier, pp. 2119–2171.

Segal, David. 1976. “Are There Returns to Scale in City Size?,” The Review of Economics and Statistics, 58,339–350.

Sveikauskas, Leo. 1975. “The Productivity of Cities,” The Quarterly Journal of Economics, 89, 393–413.Van Beveren, Ilke. 2012. “Total Factor Productivity Estimation: A Practical Review,” Journal of Economic Surveys,

26, 98–128.Van der Panne, Gerben. 2004. “Agglomeration Externalities: Marshall versus Jacobs,” Journal of Evolutionary

Economics, 14, 593–604.Venables, Anthony J. 2011. “Productivity in Cities: Self-selection and Sorting,” Journal of Economic Geography,

11, 241–251.


Date post:	01-Oct-2016
Category:	Documents
Upload:	richard-harris
View:	218 times
Download:	4 times

IS PRODUCTIVITY HIGHER IN BRITISH CITIES?

Documents