The ability of current statistical classifications to separateservices and manufacturing

Ts

JA

ARRAA

JCLLE

KISI

1

feTyntttitbK1i

0h

Structural Change and Economic Dynamics 26 (2013) 47– 60

Contents lists available at SciVerse ScienceDirect

Structural Change and Economic Dynamics

jou rn al h om epage : www.elsev ier .com/ locate /sced

he ability of current statistical classifications to separateervices and manufacturing

esper Lindgaard Christensen ∗

alborg University, Department of Business and Management, Fibigerstraede 11, 9220 Aalborg OE, Denmark

a r t i c l e i n f o

rticle history:eceived July 2010eceived in revised form May 2013ccepted June 2013vailable online xxx

EL classification:8110

a b s t r a c t

This paper explores the performance of current statistical classification systems in classify-ing firms and, in particular, their ability to distinguish between firms that provide servicesand firms that provide manufacturing. We find that a large share of firms, almost 20%, arenot classified as expected based on a comparison of their statements of activities with theassigned industry codes. This result is robust to analyses on different levels of aggregationand is validated in an additional survey. It is well known from earlier literature that industryclassification systems are not perfect. This paper provides a quantification of the flaws inclassifications of firms. Moreover, it is explained why the classifications of firms are impre-

8001

eywords:ndustry classificationervices

cise. The increasing complexity of production, inertia in changes to statistical systems andthe increasing integration of manufacturing products and services are some of the primaryand interrelated explanations for this lack of precision. We emphasise, however, that suchclassification problems are not resolved using a ‘technical fix’. Any statistical classificationmethod involves a number of tradeoffs.

ndustrial dynamics

. Introduction

Statistical classification systems are useful for identi-ying specific features of firms, certain segments of theconomy, and the dynamics and structures of the economy.hese classifications have been developed over hundreds ofears, and they are used by multiple organisations for busi-ess, policies, planning, and research purposes. Therefore,he design and accuracy of industry classifications is impor-ant to our understanding of the structure and dynamics ofhe economy. In turn, the data used for prediction and pol-cy are influenced by these classifications. However, bothhe design of classification systems and their accuracy haveeen questioned (Jacobs and O’Neil, 2003; Clarke, 1989;
ahle and Walkling, 1996; Bhojraj et al., 2003; McKelvey,982; Hicks, 2011). There are relatively few previous stud-
es on this subject, but they have generally identified a

∗ Tel.: +45 99408261; fax: +45 98156013.E-mail address: [email protected]

954-349X/$ – see front matter © 2013 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.strueco.2013.06.002

© 2013 Elsevier B.V. All rights reserved.

number of problems with the current statistical classifi-cations.

The present paper follows these earlier contributions,by showing that industry codes do not precisely describethe activities of firms. Some of this lack of precision isbecause the nature of economic activities crosses the classi-fications of statistical systems. For example, firms classifiedas service firms do not generate their turnover only fromservice activities, and a considerable proportion of man-ufacturing firms conduct a substantial amount of serviceactivities. Other, frequently overlooked, explanations forthis lack of precision in the statistical classifications areexplored later in this paper. The paper provides the firstquantification of this phenomenon. In a sense, this paperextends a well-known statement by J.M. Keynes: ‘It may bebetter to be roughly right than exactly wrong.’ This paperconsiders exactly how ‘roughly right’ the current statistical
classifications are.
Previous studies have primarily used case studies toconclude that firms that are classified as manufacturingfirms also provide services. Services and manufacturing

dx.doi.org/10.1016/j.strueco.2013.06.002

http://www.sciencedirect.com/science/journal/0954349X

http://www.elsevier.com/locate/sced

http://crossmark.dyndns.org/dialog/?doi=10.1016/j.strueco.2013.06.002&domain=pdf

mailto:[email protected]

dx.doi.org/10.1016/j.strueco.2013.06.002

e and Ec
48 J.L. Christensen / Structural Chang
products are increasingly interwoven and are difficultto separate statistically. With the exception of a fewstudies that have used input–output data to demon-strate the services embodied in manufacturing products(e.g., Pilat and Wölfl, 2005; McCarthy and Anagnostou,2004), few quantitative studies examine the severity ofthis measurement problem. Earlier studies have primarilyinvestigated specific sub-industries rather than the wholeeconomy. This paper incorporates all industries in the pri-vate sector but concentrates primarily on the distinctionbetween services and manufacturing, because this discus-sion has been an important and recurrent issue in academicdebates.

An examination of the extent to which the presentstatistics place firms in the correct ‘box’ is at the heart of thisresearch as are explanations why this may not be the case.Furthermore, we examine whether it is possible to distin-guish between services and manufacturing in a meaningfulway. There may be anecdotal evidence that manufactur-ing products and services are linked, but the extent ofthis effect requires further investigation (Guerriere andMeliciani, 2005). We note that the inadequate precisionof statistical classifications is not only a question of tech-nicalities and resources, but it also involves fundamentalchoices regarding the criteria for classification and thelimitations of separating complex economic activities intopre-determined boxes.

On a more general level, this paper is also concernedwith the fundamental premise in the classical writings onclassifications (e.g., McKelvey, 1982): to classify organi-sations in homogeneous sub-populations, it is necessaryto develop an understanding of diversity so that unifor-mities across organisations can be understood. This paperwill not examine the science of classification and the clas-sification of classifications (Good, 1965), which has beenthoroughly discussed in biology, especially from the 1960sto the early 1980s.1 However, we will address the fun-damental question of the criteria for drawing boundariesbetween industries.

This paper begins by reviewing existing classificationcriteria and subsequently discusses the basic problems,as they are described in the literature. We then empiri-cally explore potential problems with misclassification andindicate the industries in which these problems may beparticularly severe. This discussion is followed by a reviewof three possible explanations for misclassification. First,we suggest that firm activities may not only be difficult toclassify, but they may also be interwoven, further compli-cating precise classifications. Second, the flaws in technicalassignation are assessed. Third, we test whether the find-ings are robust to introducing more than one industrycode for a firm. Although these three possible explana-tions are interlinked, they are presented separately forexpository reasons. The last section sums up our conclu-sions and discusses both the direct implications and the
broader issues regarding the criteria for the classification offirms.
1 See McCarthy (1995) for an overview.

onomic Dynamics 26 (2013) 47– 60

2. Criteria for classifications

Sector and industry breakdowns and the size distribu-tion of firms are two of the most important and widely usedsegmentations of the industrial structure. Other classifica-tions that are frequently used in industrial economics andinnovation studies include Pavitt’s taxonomy of sectoraltechnological regimes, the specialisation patterns of indus-tries or trade, and high vs. low-tech industries. Peneder(2003) provides an overview of some industry taxonomiesthat are focused upon technology and their techniques butdoes not mention the lack of precision in these classifica-tions stemming from incorrect industry codes. McCarthy(1995) explains the vocabulary behind the classificationsand indicates why classifications are useful.

The objective of grouping firms into different classes isto compile groups of firms that are similar in key dimen-sions. Statistical classifications strive to provide exhaustivecoverage of the observed universe. Classifications thereforeinvolve mutually exclusive categories and use methodolo-gies that allow for the consistent allocation of the entitiesobserved. Industrial classifications look for similarity inlabour and raw material inputs, production processes, andthe services and products produced. If these activities areintegrated into the same statistical unit, the combinationis regarded as one activity (Eurostat, 2008).

A fundamental point of departure for these classificationsystems is their criteria for classification. Some classifica-tions use the product range of the industries as a point ofdeparture, whereas others find technology to be a moreappropriate classification criterion. Despite attempts tointroduce consistency to industrial classification systemsand to classify firms by their type of economic activity, alack of precision persists in the principles of classification.Therefore, substantial variance exists in the classificationof firms by their end products, activities, end use, rawmaterials or market (U.S. Department of Labor, 2001). Eachcriterion could be argued to be correct, but it is difficult toargue that the system is consistent if all of these criteria areused simultaneously.

As will become evident from Section 3, the literaturehas pointed to the difficulties and lack of precision in theindustry classifications using existing classification crite-ria. The next question is, of course, what are alternativeclassification criteria?

Hicks (2011) review both a production oriented (NAICS)and a demand oriented classification system (Global Indus-try Classification Standard – GICS) and finds that neitherof them performs satisfactory. Although recognizing thatclassification systems cannot precisely reflect reality per-fectly it is argued that the current schemes could beimproved and that a new generation of classificationsshould avoid the mistakes entailed in the present system.

It is maintained in a paper by Dalziel (2007) that thechallenges facing classification systems are much largertoday due to more complex products and services, andincreasing interfirm relations. A demand-based system
would in her opinion better capture these phenomenon.
An alternative method of classifying firms is to considertheir relationships rather than focusing on their charac-teristics and the goods or services they produce. Thus, an

e and Ec

ifiapgu

tloicbe

3al

oiaitaaTe

3

cs1liisofitratcdAItfil

ctasfc(

J.L. Christensen / Structural Chang

ndustry would be, for example, a set of interdependentrms, in Cubbin’s (2001) view. Proponents of this approachrgue that classification based on processes rather thanroducts exhibits greater continuity because processes areenerally more generic and have lower turnover than prod-cts.

Carroll et al. (2000) and Peneder (2003) focuses uponaxonomies often used in empirical studies of the techno-ogical content of industries and point to the heterogeneityften hidden within the categories used. Although anntegrated approach would be desirable, separate classifi-ations for manufacturing and services could be consideredecause of their huge differences in technology and marketnvironments.

. The efficacy of industry classification systemsnd the boundaries between industries in previousiterature

Two closely interrelated issues dominate the discussionn the precision of industrial classifications. The first issues the extent to which it is possible and meaningful to sep-rate the manufacturing and service industries. Anotherssue is the technical precision related to assigning indus-ry codes to firms. These issues are addressed here withn emphasis on the boundaries between manufacturingnd services and the criteria for the classification systems.he issue of technical assignation is kept brief here andlaborated later in the paper.

.1. Precision of the classification of firms

Debates continue over whether the current statisticallassifications adequately capture an accurate and relevantegmentation of firms. This debate was noted as early as975 (Fertuck, 1975), but relatively few contributions in the

iterature have addressed this issue. Clarke (1989) exam-ned whether industry codes are useful for grouping firmsnto homogeneous segments. He concluded that firms withimilar characteristics are not easily identified when usingnly industry codes, especially 1- or 2-digit codes, as classi-cation devices. Dahlstedt et al. (1994) found in their studyhat industry classifications differed to an extent that couldender comparisons between industries misleading. Kahlend Walkling (1996) compared two databases and foundhat at the two-digit level, there was a 36% difference in thelassification of firms, which increased to 80% at the four-igit level. Bhojraj et al. (2003) compared NAICS (Northmerican Industry Classification System), and GICS (Global

ndustry Classifications Standard) classifications and foundhat GICS performed better in classifying firms based onnancial data and may be a better choice for financial ana-

ysts.The literature points to numerous examples of inac-

urate industry classifications. Livesey (2006, p. 8) notedhat software delivered on CD-ROMs or disks is classifieds manufacturing, but the same software is classified as a
ervice if it is delivered through the Internet. Timber meantor furniture may be classified as manufacturing, but it islassified as construction if it is used to build houses. Dalziel2007) listed several examples in which firms are classified
onomic Dynamics 26 (2013) 47– 60 49

without obvious logic. For example, the eight leading com-munication equipment manufacturers are classified intofour different industries and two different sectors. A fewcases of inconsistent allocation would not pose problemsfor the analysis of industries, but if this phenomenon ispervasive, misleading perceptions of the industry structureand dynamics may result.

Generally, tests of the efficacy of classification schemeshave used an ‘internal benchmarking’ approach thatcompares across different classification schemes if theseschemes group firms consistently. In some cases, thesestudies apply one or more validation variables, such asfinancial ratios (e.g., Guenther and Rosman, 1994; Bhojrajet al., 2003; Hicks, 2011). Although this is a valid exercisethat may illuminate some of the necessary choices betweenclassification schemes, it does not solve the fundamentalconceptual issue of expedient criteria for classification.

Hicks (2011) noted that classification systems differin their ability to precisely classify firms, and she main-tained that the categories of the current statistical systemare inadequate to describe the modern economy. Newindustries emerge at a pace that significantly exceedsthe changes in statistical systems (Graham et al., 2007).Moreover, there is an unbalanced structure in the systembecause some industries with little weight in the economyhave fine-grained categories that differentiate betweensub-industries, whereas other industries with much largereconomic weight are pooled into broad categories (Hicks,2011).

The examples of inaccurate classifications mentionedabove might be partly explained as a technical problemrelated to the assignment of industry codes. However, otherphenomena may distort the precise allocation of types ofactivities in the economy. Such phenomena include blurredboundaries between industries due to multiple activities,the convergence of industries, and inter-related activi-ties. In contrast, some features of the current organisationof production contribute to clearer boundaries betweenindustries.

3.2. Multiple types of activities in firms

Firms are often understood to belong to a certain indus-try and to perform activities typical of that industry. Inreality, most firms perform a range of activities that canbe assigned to industries other than the one indicated bytheir primary industry code. Services may be attributed tothe service sector of the economy when they are performedby specialised service firms; however, the service functionand service products may be generated in all firms in aneconomy regardless of their statistical classification. Theseactivities may span traditional statistical classifications. It isdifficult to quantitatively study these activities at an aggre-gated level because firms are usually grouped according tothe product in their product portfolio that produces thehighest revenue.

As a result, the phrase ‘growth of the service economy’
(Fuchs, 1965; Greenfield, 1966; Gershuny and Miles, 1983)may have dual meanings. This phrase encompasses notonly the growth of the service industry as a group ofspecialised service providers but also the fact that services,

e and Ec
as a function, may have increased in the economy as awhole. Many functions are generic, which means thatactivities that are strictly industry related are reduced toa smaller share of the total activities of the firm. The trendtowards more complex and cross-disciplinary productionprocesses means that both the knowledge bases and thepure activities of the firm span a range of different typesof processes, thereby complicating the design of a classifi-cation system that reflects the industry structure (Dalziel,2007). Consequently, it may be necessary to relax the sharpdistinction between industries (e.g., manufacturing andservices) because of the widespread presence of serviceactivities in non-service sectors (Nählinder, 2005) andbecause empirical findings suggest a similarity betweensome types of service industries and manufacturingindustries (Preissl, 2000).

3.3. Convergence of industries

Related to the discussion above, some studies havenoted a convergence between services and manufactur-ing (Coombs and Miles, 2000; Gallouj and Weinstein, 1997;Pilat et al., 2006). The manufacturing sector adopts charac-teristics of service firms and has a growing share of revenuefrom services. Moreover, the service sector has an increas-ing impact on other sectors. The similarity of productionmodes in manufacturing and services suggest that manu-facturing and service are becoming similar. Some firms thatare classified as manufacturing generate large amounts oftheir turnover by selling services, as illustrated by Howells(2004), who reports that firms such as IBM and Siemensderive more than 50% of their turnover from service activ-ities. AEGIS (2002) found that services are essential inputsfor all modern manufacturing, often accounting for 60–75%of input costs. Pilat and Wölfl (2005) confirmed this find-ing, observing that workers in firms in the manufacturingindustry are increasingly occupied with service activities.In some countries, up to half of these workers are engagedin services.

Consequently, the statistical classifications of firms andindustries into either manufacturing or services makesit difficult to statistically aggregate activities, and indeedthis problem is enhanced when service and manufactur-ing activities are intertwined. Related, Howells relates theincreasing focus on service activities in manufacturing to achange in the perception of consumption, with a growingtendency to perceive consumption as a continuing pro-cess involving long-term customer contact through servicedelivery rather than as a one-off contact through a productsale. Howells argues that the shift in selling and consump-tion may increase firms’ concerns about reliability and easeof servicing because these firms may eventually have tobear the costs of these activities.

3.4. Interwoven products and services

In the same way that manufacturing companies increas-
ingly link services to their production, service firms havebecome more engaged in delivering physical products aspart of a service package or support for a service, andprovides firms with a competitive edge (Nordås and Kim,

2013). There is a significant amount of interdependencebetween the two sectors because many products com-prise both a tangible product and services (such as training,maintenance), that are sold in packages delivery (Howells,2004; Sundbo, 2001). Examples of the complementari-ties between tangible and service products include theincreased tendency to sell all-coverage insurance togetherwith electronics or the service programme and the (differ-entiated) guarantee offered with a new car. In the latterexample, a range of services may be offered, such asfinancing and maintenance services. Manufacturing firmshave increasingly integrated products and services as partof their strategy, which contribute to removing barriersbetween industries (AEGIS, 2002). In some cases it mayeven be difficult to distinguish between products and ser-vices that are packaged together. A mobile telephone soldwith a subscription could be seen as both a product with aservice and a service with a product. The price structure ofthe product/service in this example is not very helpful indetermining how it should be classified because the priceis often advertised as a skewed distribution; for example,the phone may cost 1 Euro, with the subscription garneringall of the revenue for the seller.

3.5. More focus – clearer boundaries?

Another trend in the organisation of production ismoving in the opposite direction by providing clearerboundaries between industries. Following a period in the1970s and 1980s in which leading firms engaged in ver-tical integration and ‘big is beautiful’ was the mantra,there has been an increasing tendency to outsource andfocus on core capabilities (Carlsson, 1989; McCarthy andAnagnostou, 2004). Instead of firms attempting to acquirecontrol over more parts of the value chain through owner-ship by vertical integration, strategic sourcing (Venkatesan,1992) has become popular. In addition to a cost-reductionelement, surveys among outsourcing firms generally men-tion a “focus on core activities” as a motive for outsourcing,in line with the strategic sourcing trend. Whereas out-sourcing was once primarily confined to low-skill activities,it has recently become prevalent in knowledge-intensiveactivities. This change points to a clearer division of labourbetween firms in the value chain as well as the possibilityof a clearer statistical separation between manufacturingand service activities.

Some of the growth of the service sector may thusbe ascribed to the statistical separation and re-groupingof activities that have not necessarily increased in quan-tity (OECD, 1999). However, according to Schettkat andYocarini (2006), this effect is not very significant. Althoughoutsourcing increased service employment, it explains lit-tle of the shift from manufacturing to service employment.It is possible. However, that this clearer separation betweenactivities is only legal in nature. The activities are separatedto a greater extent in different entities, but, at the sametime, these same entities interact and trade intensively
and in complex ways, and they may even have cross-ownership. Moreover, some of this clearer separation ofactivities occurs between different establishments withinthe same firm and in conglomerates of firms.

e and Ec

‘rekdMspot

iobtdmciotseteG2

sctfmtCa2t

4

4

eiqfifctshocNs

rc


Clearer boundaries may also stem from a trend towardsdown-sizing.’ In some firms, there has been a counter-eaction to mergers and larger entities. To ensure anntrepreneurial culture in the firm, some firms attempt toeep the size of units down, in some cases, even splittingivisions as soon as they exceed a certain threshold size.odularisation is another trend that may spur a clearer

eparation of different parts of the value chain. Finally, asroduction becomes increasingly specialised, the divisionf labour increases, which may produce a clearer separa-ion between activity types.

Thus, a growing body of evidence suggests increas-ngly blurred boundaries between industries, whereasther trends point in the direction of a clearer separationetween manufacturing and services. Although the struc-ure and divisions between industries may be difficult toetermine when coupled with problems with the assign-ent of individual firms to industries, this difficulty is often

ompounded by insufficient studies on the dynamics ofndustries. Service researchers have argued that the extentf service activities, such as new forms of organisation, newypes of customer relations and delivery, and new packageolutions, has not been adequately measured in existingmpirical studies. Furthermore, these researchers suggesthat studying the dynamics of services is distinctly differ-nt from studying manufacturing innovation (Sundbo andallouj, 1998; Coombs and Miles, 2000; Djellal and Gallouj,001).

Based on the above discussion, the borders betweenervice and manufacturing activities appear to be diffi-ult to determine using existing statistics. Furthermore,he extent to which it is meaningful to separate manu-acturing and services is questionable. These two sectors

ay converge due to an increased focus on selling solu-ions rather than products (Vandermerwe and Rada, 1988;oombs and Miles, 2000; Gallouj and Weinstein, 1997),nd the two main sectors have similar dynamics (Howells,006). Therefore, more precise knowledge on the extent ofhis problem is desirable.

. Empirical study of classifications of firms

.1. Classification criteria

Based on survey data from Denmark, this sectionxplores the extent of the blurred boundaries betweenndustries and the ability of a statistical system to ade-uately describe the type of activity in firms. To classifyrms this investigation uses information obtained directly

rom interviews with firms on their share of activitieslassified as respectively services, and the production ofangible products. Hence, our classification criteria in theurvey are the firms’ self-perception of what they do; andow much of their activities should be grouped in two typesf activities: production and services. This information isompared to firms’ official industry codes according to theACE (European Classification of Economic Activities) clas-

ification system.In the NACE system, the value added is the primary crite-

ion when industry codes are assigned. Firms with activitiesovering several parts of the value chain or firms with


horizontal integration should be classified according to thepart that contributes the most to the total value addedrather than according to the end-product. This method isstraightforward when an activity represents more than 50%of the value added. In cases where a unit performs morethan two activities in more than two NACE categories andnone of these exceeds half of the value added, the ‘top-down’ method is used to identify the primary activity (see,e.g., Eurostat, 2008).

4.2. The data

The main information on firms’ activities was drawnfrom telephone interviews with representative panel ofmanagers of 1007 private firms in Denmark. This data col-lection was part of a survey on the business cycle butincluded a few questions on the nature of firms’ activ-ities. These questions were specifically designed for thepurposes of the present paper. Respondents were asked tocharacterise and quantify the activities of the firm into ser-vices, production of tangibles, and other activities. In turn,these data were compared with the firms’ official industrycodes.

Firms throughout the entire private sector with at least5 employees were selected. Within this segment, the samp-ling was random. The data were subsequently weighted bysize (employment) and industry. Comparing the realisedsample with the population by size and industry groupsrevealed that the realised sample closely reflected the pop-ulation, with deviations up to 4%. The composition of thepanel was regularly adjusted to ensure representativeness,and it was possible to dynamically adjust the interviewsaccording to possible skewness. The take-up rate was sat-isfactory, and the interviewed firms represented more than20% of employment. In summary, the data were balancedand representative for the purposes of this study.

Respondents should be able to provide valid answersto interview questions. Although there may be differencesbetween firms in the perception of products and services,two points indicate that this was not a significant issue inthis study. First, as shown in Table 1, only 2% of the respon-dents answered ‘Do not know/other’. Second, debriefingwith the interviewers revealed that respondents had noproblems in understanding or categorisation.

4.3. Industry classification and firms’ own views of theiractivities

Tables 1 and 2 list the share of firms that assigned apercentage of their activities to ‘services’ and ‘production’,respectively (the residual ‘other’ was listed by very fewfirms). These statements were compared with the assignedindustry codes following the NACE classification system.The data were grouped into the firms within an industryor size group with no services, 1–49% services, 50–99% ser-vices, and 100% services. A similar grouping was applied forproduction. This method allowed us to compare the data
to the “officially” assigned industry codes. The results aredisplayed in Table 1 (services) and Table 2 (production).
The two tables show that 32% of the firms had bothservice and production activities. This result indicates that

52 J.L. Christensen / Structural Change and Economic Dynamics 26 (2013) 47– 60

Table 1Share of activities listed as services, by industries and size groups.

Sector affiliation of the responding firm No service 1–49% 50–99% 100% Don’t know/other Number of cases

Agriculture, fisheries, raw materials 46 20 15 18 43Industry 41 35 14 10 1 226Building and construction 7 18 30 41 4 153Trade and restaurants 7 10 17 65 1 326Transport and telecommunications 1 5 94 59Finance, business services 3 6 16 72 2 200

5–10 employees 11 10 19 58 2 40110–19 employees 13 17 19 49 2 27920–49 employees 18 23 14 46 17750–100 employees 31 22 11 33 2 91>101 employees 38 26 9 24 3 34

Total 15 15

a classification system based upon distinguishing betweenservices and manufacturing will have difficulty consis-tently allocating a third of the firms in the economy. Inthe area of Industry, half of the firms had both types ofactivities. Table 1 shows that half of the firms in the survey(51%) are classified as pure service firms, and two-thirds(68%) listed the majority of their activities as services. Fur-thermore, 18% of the firms in Agriculture, Fisheries, andRaw materials and 10% of the firms classified as Industrycharacterised all of their activities as services. Many firmsin areas that are usually regarded as production-intensivesectors listed service as 50% or more of their total activities.Thus, a large share of firms – 33%, 24% and 71% in Agricul-ture, Fisheries, and Raw Materials, Industry, and Buildingand Construction, respectively, claimed that at least halfof their activities were services rather than production.Accordingly, these firms should be classified as services ifthe term “activities”, as reported by the firms, is used as thesole criterion for classification.

The propensity to list activities as pure servicesdecreases with firm size, which may be explained by thefact that many large firms are multi-product firms thatinclude either different and complementary products ormore than one link in the value chain. The larger share ofservice activities in large firms may also indicate that thesefirms are more complex organisations that must engage inauxiliary activities, such as human resource management.Large firms may also have internal functions that wouldotherwise be acquired from external providers, such asIT, building and machinery maintenance. Finally, the sizeeffect may be explained by the fact that service firms aregenerally smaller and are therefore more likely to fall intothis group.

We now consider firms’ propensity to list production asthe main activity. Table 2 mirrors Table 1, to some extent,it also supplements the insights of Table 1.

The first result worth noting is that there is a remarkabledifference between the share of firms with no produc-tion (53%) and those with no services (15%). The mostobvious explanation for this difference is that manu-
facturing firms are dependent on a number of serviceactivities that are integrated through in-house production.In addition to the examples mentioned above, infor-mation processing is another such service activity. This
17 51 2 1007

finding may also be interpreted in line with the find-ings in Pilat and Wölfl (2005), which suggest that theservice sector is more independent of other industriesthan the manufacturing sector. This may be becauseservice firms often source inputs to their activities fromthe sector itself, whereas manufacturing industries aredependent upon inputs from other industries, includingservice industries, and act as providers of intermediateinputs.

As expected, the table shows that firms in Industry andin Agriculture, Fisheries and Raw Materials list produc-tion more frequently than do firms in other industries.However, even within these industries, 18% and 11% offirms, respectively, list ‘no production’ corresponding tothe shares listed in Table 1 (18% and 10% under ‘100%service’). Furthermore, in these two sectors, 25% and 19%of firms, respectively, characterise more than half of theiractivities as outside manufacturing. Again, Building andConstruction appears to be dominated by activities that arenot considered manufacturing by the firms themselves. In65% of the firms in Building and Construction, a minimumof 50% of the activities are primarily outside production.Probably maintenance and repairs are considered servicesby the firms, whereas the construction of houses, for exam-ple, is seen as production because it involves the creation ofnew, tangible goods. As also expected, production by firmsize produces results opposite those of the accounts of ser-vices: smaller firms are less ‘manufacturing intensive’. Thisfinding is consistent with the fact that service firms areusually relatively small.

This discussion has indicated the shares of firms in dif-ferent industries by categories of activities and has allowedus to specify the share of firms that are classified inindustries that intuitively do not correspond to their self-reported, dominant activities. In addition to categories ofactivities, we are interested in the precise share of eco-nomic activity ascribed to services and manufacturing. Themean values for the share of activities characterised by thefirms as services (Table 3) show that two-thirds of the eco-nomic activity in the private sector involves services. This
result corresponds to the findings of Pilat and Wölfl (2005),who contend that approximately 70% of all employment isworkers who are employed in the service sector, and 66%of employment is required to meet the final demand for

J.L. Christensen / Structural Change and Economic Dynamics 26 (2013) 47– 60 53

Table 2Share of activities listed as production, by industries and size groups.

Sector affiliation of the responding firm No production 1–49% 50–99% 100% Don’t know/other Number of cases

Agriculture, fisheries, raw materials 18 7 29 46 43Industry 11 8 40 41 1 226Building and construction 46 19 24 7 4 153Trade and restaurants 67 11 14 6 1 326Transport and telecommunications 96 3 1 59Finance, business services 73 8 14 3 2 200

5–10 employees 59 11 17 10 2 40110–19 employees 53 10 23 12 2 27920–49 employees 47 9 25 18 17750–100 employees 32 8 26 31 2 91101–199 employees 24 6

Total 53 11

Table 3Means of the share of activities listed as services, by industries and sizegroups.

Sector affiliation of the responding firm Mean Number of cases

Agriculture, fisheries, raw materials 30 43Industry 23 226Building and construction 65 153Trade and restaurants 79 326Transport and telecommunications 99 59Finance, business services 84 200

5–10 employees 72 40110–19 employees 65 27920–49 employees 59 17750–100 employees 44 91

swcBt

hTpcetvNrb

alt

Sm(pUewl

>101 employees 33 34

Total 66 1007

ervices.2 An industry breakdown shows results in lineith those discussed above. For example, firms in Industry

laim that one-fourth of their activities are services,3 anduilding and Construction firms characterise two-thirds ofheir activities as services.

These results indicate that there is substantial sectoraleterogeneity, especially in the two ‘production’ sectors.his heterogeneity becomes particularly visible when welot the firms’ responses against their designated industryodes (Fig. 1). If we interpret Industry and Agriculture, Fish-ries and Raw Materials as manufacturing industries, thenhe dotted vertical line separates manufacturing and ser-ices. The y-axis to the left of the line includes firms with
ACE codes 11–39, with the remainder of the firms to the
ight. A total of 181 firms in the top left quadrant and theottom right quadrant are classified in industries that we

2 Similar figures for the manufacturing industries in their study are 18%nd 22%. Their numbers are for France, but the authors claim that simi-ar numbers are found in all countries for which there are input–outputables.

3 A somewhat comparable study in four OECD countries (Finland,weden, Japan, New Zealand) found that the share of service activity inanufacturing is increasing (Pilat and Wölfl, 2005). Another OECD study

Pilat et al., 2006) found that in the manufacturing sector in 13 Euro-ean countries, only 60% of workers were engaged in production in 2002.sing employment data, that study also found an increase in the share ofmployment in service-related occupations since 1995, although a declineas noted in a few countries. This increase may be even higher over a

ong-term perspective.

29 38 3 34

21 14 2 1007

typically do not relate to what they claim are their mainactivity. Thus, the manufacturing firms have over half oftheir activities in services and the service firms have lessthan 50% of their activities in service. Hence, although theterm ‘misclassified’ may be too strong, the total share ofmisclassified firms is 18% (181/1007).

Fig. 1 illustrates that there are differences across indus-tries in how many firms are classified outside their mainactivity (in the following these firms are for simplicitytermed ‘misclassified’, even if this indicates that the NACEclassification is unable to correctly classify firms). TheTransport and Telecommunications and Finance and Busi-ness Services industries have the greatest number of firmsthat are assigned the correct industry codes. One-fourthor more of the firms in the Agriculture, Fisheries and RawMaterials, Industry and Building and Construction sectorswere ‘misclassified’. This intra-industry heterogeneity ispotentially important because it may indicate that ‘mis-classification’ problems are confined to specific segmentsor types of firms.4 Further fine-grained analyses of the datamay reveal whether specific firms within these broad cat-egories are ‘misclassified’ particularly often.

The high level of aggregation of this analysis may poten-tially obscure the fact that some of the 181 apparently‘misclassified’ firms may be in industries that would cor-respond to the activities, as stated by the firms, at a moredetailed level of aggregation. For example, the three two-digit industries Building and Construction, Agriculture, andIndustry all include sub-industries that can be expectedto be dominated by service activities, such as machin-ery rentals, maintenance and repair, and branch-specificconsultancy.5 To control for this possible source of error,
the 181 firms that were apparently misclassified were ana-lysed further using their 6-digit industry codes. An analysisof industry differences based upon the NACE 6-digit level
4 According to Statistics Denmark, small firms are categorised into thewrong industry group more often than large firms. Therefore, the share ofturnover or employment in the economy classified in the wrong industryis smaller than the share of firms. Comparing the average size of firms inthe ‘181/misclassified’ group with the rest of the firms reveals that themisclassified firms are indeed smaller.

5 A communication with Statistics Denmark indicated that qualitychecks have revealed a marked effect from the aggregation level of anal-ysis on the number of firms that are placed in the wrong group.


ice acti
Fig. 1. Plot of share of serv
of aggregation reveals that a broad variety of firms showsto be ‘misclassified’. Omitting Building and Construction,which was mentioned above as an industry that could beregarded as belonging to either manufacturing or services,then industries such as Bakeries, Publishing/Printing, andMachine processing pop up as some of the most frequently‘misclassified’ firms. The firms’ responses regarding theshare of service activities were compared to the industrycode and were manually assessed to determine whethertheir industry codes corresponded to their responses aboutthe type of their activities. After removing 10 firms thatlacked a 6-digit industry code and 6 firms with invalidanswers, we identified 16 firms that could be consideredmisclassified by a 2-digit industry code, but perhaps not at amore detailed level. These firms were agricultural contrac-tors, engine repair, landscape gardening (1 firm), pre-presswork (1 firm), software (which could be classified in eithergroup) (3 firms), and non-financial holding companies (notspecified; could be classified in either group) (4 firms).Thus, we identified at least 149 (181 − 32) misclassifiedfirms at the more detailed level.

Vice versa it may be that some of the firms that werefound to be correctly classified in the 2-digit analysis were,in fact, misclassified. Therefore, the rest of the sample wasanalysed using the 6-digit codes. Omitting firms without6-digit codes or invalid percentages of service activities,an additional 50 firms were found to be misclassified,rendering a total of 199 misclassified firms. Because thedenominator was reduced to 893 due to the subtractionof missing or invalid values, we found that 22% of thefirms were misclassified. As mentioned above, the Build-ing and Construction industry may be a grey area in whichfirms could be perceived as either production or services.Removing Building and Construction from the analysisresulted in 151 misclassified firms out of 785 (19%). Thus,this robustness analysis suggests that the results do not
change significantly with the use of more detailed indus-try codes. Slightly less than one in five firms claim that themajority of their activities are services even if they are clas-sified as manufacturing or that their activities are primarily
vities against NACE codes.

production even if they are officially listed as service firms.An additional validation was performed using a similar sur-vey of another segment of firms. This analysis is reportedin Annex B.

We have established that a substantial share of firmswere misclassified. The analysis was based upon a compar-ison of the primary NACE codes from the business registerand the firms’ claims regarding their activities. In Section5, we examine this result further with a view to try outplausible explanations.

5. Discussion: possible explanations ofmisclassifications

The literature indicates that there are currently manyinterwoven activities in production and suggests that thissituation presents a challenge to any classification sys-tem, particularly when the primary classification criterioninvolves production. Another possible explanation is thatthe assignation of industry codes is not precise. A third pos-sible explanation is that firms may have more than oneindustry code; in this case, analysing all of the codes ratherthan only the primary code could potentially change theresults regarding the classifications.

Even if these explanations are presented separately theyare fundamentally inter-linked and reflect the general dif-ficulty in separating what firms do in well-defined boxes.These difficulties emanate at different levels and perspec-tives, such as at the level of the single product and the levelof the firm. Likewise, both the firm and statistical officeshave difficulties assigning the activities of the firm to thecorrect ‘boxes’.

5.1. Interwoven activities

The integration of products and services complicates a
statistical separation of the two. It is problematic at twolevels of aggregation; on a product level the bundling andintegration of products and services contributes to blur-ring boundaries between statistical groups; and on a firm

e and Ec

li

BmuetuhvcwoTfiditb(t2

sbpcDrwTssiivfsitAF(dTbtahadp

Si2Avs

do not expend significant time or effort to calculate the


evel the fact that many firms do both also means that it isncreasingly difficult to assign firms to specific industries.

The servitization debate (Vandermerwe and Rada, 1988;aines et al., 2009; Santamaría et al., 2012; Lodefalk, 2013)ay be evidence on the tight coupling of services and man-

facturing. This debate got momentum in the late 1980s,specially with the Vandermerwe and Rada 1988-paperhat defined the concept in the title: adding value (to prod-cts) be adding services. Thus, the literature focuses onow product offerings are often sold together with ser-ices, which partly reflects an increasing complexity inonsumer demands, but indeed also is created by firmsho engage in maintaining competitive advantage through

ffering complex product-services (Nordås and Kim, 2013).he interwoven product-services hence pose challenges torms’ strategic orientation, as they are no longer solely pro-ucers but also service providers. As part of a strategy an

ntertwined product-service may be a way to enhance cos-umer loyalty and differentiate products. Moreover, it haseen said to be a way to sustain growth in mature industriesSantamaría et al., 2012), and it has caused many manufac-uring firms to move massively into services (Baines et al.,009).

Among several researchers, Sundbo (2001) notes thatervices may be sold as part of a package that involvesoth pure services and tangibles. To investigate if thishenomenon is also present in the Danish context, weonducted a survey of knowledge-intensive services inenmark, as described in Annex B. The survey asked

espondents in these firms whether products and servicesere developed and delivered in the form of a package.

he results show that 37% of the knowledge-intensiveervice firms that developed new services also developedervices that were delivered as part of a product packagencluding tangible products. Most often, the service firmtself supplied the product package that combined ser-ices and physical products, but in more than one out ofour cases, another firm supplied the product package.6 Aimilar survey found that 24% of all Danish manufactur-ng firms developed one or more new services in relationo their product development (Christensen et al., 2004).

survey of 640 firms in the Agriculture, Fisheries andorestry sector in Denmark found that 32% of these firms240 firms) developed one or more new services that wereelivered together with products (Christensen et al., 2011).his finding indicates that in a large share of the firms, theoundaries between production and services are obscuredo such a degree that innovative activities combine tangiblend intangible products. The empirical evidence referred toere suggests that it is a pervasive phenomenon that holds
cross size groups and sub-groups of the industry and inifferent industries. It should be noted that the questionosed in the surveys only asked about new services. It is
6 A similar question was posed in the Danish Community Innovationurvey. The results of this survey show that 31% of innovative knowledge-ntensive service firms introduced both new services and products in004–2006, and 26% claimed to have only introduced new products.mong manufacturing firms, 28% introduced both new products and ser-ices. An additional 6% of the manufacturing firms introduced only newervices.


likely that if the question were not limited to innovation,then the share of firms with combinations and packageswould be much higher. This survey cannot indicate withcertainty the extent of total revenues represented by thesepackages; we can only determine how many firms providedthis type of packaged service or production. It is difficult todetermine the extent of this source of misclassification, butour findings suggest that the interweaving of activities isa prevalent phenomenon and call into question the ratio-nale for maintaining a distinct separation of services andmanufacturing.

As discussed above, the tendency to outsource and focuson core activities has created a clearer distinction betweenmanufacturing and services, whereas the trend towardsselling solutions and complementary products rather thanstand-alone, specific products tends to dissolve the bound-aries between industries. We are unable to distinguishbetween these two effects in the data or to assess the pos-sible development of these trends over time.7 However,we have identified solid evidence that the statistical clas-sification of firms and their activities do not necessarilycorrespond.

5.2. Assignation – a failure to capture industrycharacteristics?

Difficulty in assigning industry codes is also related tothe ever-increasing speed of change in industry structuresand the upsurge of new products that may resist classifica-tion into existing statistical categories (Hicks, 2011). Thisis a recurrent problem faced by statistical bureaus. Fur-thermore, the industry code assigned to a firm may notbe changed when the product range of the firm changes.Firms are not primarily concerned with verifying whetherthe industry codes in statistical accounts precisely reflecttheir changes in activities. Even if firms do change theirprimary or secondary industry codes, the codes are notalways changed at a more detailed level. A study by Kahleand Walkling (1996) reported that changes in 4-digit codesoccurred in only 24% of firms during a 20-year period.8

Bhojraj et al. (2003) found that in each of the years between1994 and 2001, only 2.8% of Standard & Poor’s 1500 firmschanged their industry codes. Industry codes may inade-quately reflect contemporary activity types, which makesit difficult to assign correct codes at a four-digit level, asdemonstrated by Hicks (2011).

Industry codes are primarily self-assigned by firms.Although some firms may seek advice in assigning the cor-rect industry code, one study has shown that many firms

correct code or to determine whether the available codesadequately describe the activities of the firm (Jacobs and

7 As mentioned above, a study in four OECD countries found that theshare of service activity in manufacturing is increasing (Pilat and Wölfl,2005). A number of earlier studies (see Mathe and Shapiro, 1993, foran overview) and, more recently, Pilat et al. (2006) have found that theamount of services included in manufacturing goods has risen over time.

8 Statistics Denmark has found that firms’ perceptions of their relevantclassification do not always correspond with the criteria used by statisticalbureaus.


Table 4Secondary industry codes by industries’ share of firms listing minimum of 2 NACE codes.

Percentages horizontally Secondary industry

Primary industry Agricul F&B Iron Elec Other B&C Trade Transp. Finance Total

Agriculture, fisheries, raw materials 26.07 0.29 1.50 0.16 0.86 15.87 10.53 1.50 43.21 5128Food and beverages 2.79 35.41 1.93 0.43 51.93 0.43 7.08 466Iron and metal 6.55 0.13 35.86 3.69 5.91 7.25 29.24 0.51 10.87 1573Electronics 1.74 10.00 19.57 5.43 3.91 42.61 0.22 16.52 460Other industry 3.70 0.77 3.10 .03 37.28 2.93 31.64 0.69 18.85 2323Building and construction 12.54 0.03 3.67 0.95 2.42 33.66 17.12 2.80 26.80 3675Trade and restaurants 4.33 0.89 2.05 0.80 3.36 3.09 57.74 2.16 25.58 10,259

0.480.41

0.91

firms = 0.75%) included more than two NACE codes. Giventhe discussion above on multiple activities, it seems at firstglance surprising that such a high share of firms identify

9 A 5-country OECD project in a collaboration between statistical officesinvestigated micro-level data on manufacturing and services and foundthat the data for Denmark were very similar at the enterprise and estab-

Transport and telecommunications 9.20 0.21 0.90

Finance, business services 6.56 0.15 0.64

Total 8.88 0.80 3.04

O’Neil, 2003). Consequently, industry codes do not changeoften, even if there are substantial changes in the orga-nisation of economic activities. Given the rapid change inindustrial and economic structures and in the configurationof firms, industry codes should change frequently. As illus-trated above, however, this is not the case in practise. Again,it should also be re-iterated that there are inherent difficul-ties in separating manufacturing and service activities.

In periods of rapid technological change fitting aproduct range to a category may prove difficult. Large con-glomerates and diversified companies, which may havesubstantial revenues in diverse product groups, oftenreport only a few industry codes. According to Leiponenand Drejer (2007), very few firms diversify across 2-digitindustries; even at more detailed levels of aggregation, onlya minority of large firms are diversified. The ownershipstructure may also distort the interpretation of the data.As mentioned above, the financial sector may be registeredas the owner of a significant amount of industrial activ-ity, as noted in the above discussion on outsourcing andthe changing boundaries between manufacturing and ser-vices. For example, large manufacturing companies nowhave financial services, and they may provide a full rangeof services by bundling products and services. This processcomplicates the assignment of industry codes that reflectthe true nature of these firms’ activities.

Industry codes may differ according to the level ofaggregation (Clarke, 1989; Kahle and Walkling, 1996).Clarke (1989, p. 21) presents an example of a 3-productfirm to indicate the importance of the level of aggregationat which the match between industry codes and activitiesis analysed. He describes a firm with revenues stemmingfrom three products, one from SIC 3211 (40%), one from SIC2842 (30%) and one from SIC 2845 (30%). In this example,the correct four-digit industry code is 3211, the three-digitcode is 321, and the two-digit code is 32, even thoughthe majority of the firm’s revenues stem from SIC 284/28;firms are assigned codes in the primary activity, whichis the one that generates the largest value added. How-ever, Clarke’s example is only relevant if the ‘bottom-upapproach’ is used. The usual method of assignation is cur-rently the ‘top-down approach’, in which the contributions
of related activities are added at a higher level of aggrega-tion in a hierarchical procedure. Using this approach wouldrender the classification code 28 in the example above.However, this method is not without flaws. Fully diversified
0.83 6.85 27.46 25.31 28.77 1446 3.70 4.23 22.13 1.95 60.23 14,427

5.07 8.25 30.98 2.71 39.36 39,757

firms may not be assigned the most appropriate indus-try code. The method may result in higher-level industrycodes that differ from the main activity of the firm. Evenat a detailed level of aggregation, classification accordingto the top-down method may result in the assignment ofa 4-digit code that differs from the largest 4-digit-levelactivity because of the need for consistency between assig-nations at different levels. Additionally, it is important toconsider the level of units that are analysed. Generally,at the enterprise level, many details may be overlookedbecause a range of different divisions may encompass activ-ities across different activity types.9

5.3. Are economic activities better reflected by severalNACE codes?

The majority of industry analyses are limited to only theprimary industry code, and it is often the only informationavailable to users of data. The use of the primary industrycode only may potentially obscure an accurate understand-ing of firms’ economic activities. On the other hand, theremay be cases where firms reported the industry codes forall of their major types of activities rather than only the pri-mary code, allowing us to take this source of informationinto account. Thus, one possible explanation for misclassifi-cations may be that the firms correctly report their diverseactivities by reporting several industry codes, but we rarelyinclude more than one industry code in an analysis.

To look into this potential error we obtained infor-mation on the number of Danish firms that list multipleindustry codes to determine whether firms report indus-try codes for all of their main activities. The overall resultwas that 577,612 (94%) of 617,369 firms assigned only oneNACE code (at any level). Most of the other cases con-tained information on 2 NACE codes, and less than 1% (4500

lishment levels (Pilat and Wölfl, 2005). A 5-country OECD project in acollaboration between statistical offices investigated micro-level data onmanufacturing and services and found that the data for Denmark werevery similar at the enterprise and establishment levels (Pilat and Wölfl,2005).

e and Ec

oBgIirioic

tstFtisNttipcfii

atFosoiprTtbaeaMsaimC

hctclfi

6

sw


nly one industry code. But as noted in the literature (e.g.,hojraj et al., 2003), the low share illustrates that firms areenerally not concerned with their industry classification.n particular, firms do not consider a second industry codemportant. Moreover, there may be conceptual problemselated to classification, as noted in Section 2. For example,t may be easier for firms to consider the different segmentsf customers and needs addressed by their activities than its to understand when they should have additional industryodes in a production-oriented classification system.

Of the analysed firms, 3% were classified into two indus-ries within the same sector, 1% listed Finance as theirecondary sector, 1% listed Trade as their secondary sec-or, and 1% listed another secondary sector, cf. Table 4.irms in Industry were most likely to identify an addi-ional industry code; on average, these firms identified 1.25ndustry codes, compared to an average of 1.05. Table 4hows how the 39,757 firms that assigned more than 1ACE code listed their secondary NACE codes, allowing us

o further analyse how industries are ‘connected’ by indus-ry codes from presumably related industries. Firms in thendustries listed in the first column are ordered by theirrimary industry code, and the industry of their secondaryode is listed horizontally. The diagonal shows the share ofrms with secondary industrial codes within their primary

ndustry.A small share of firms have a secondary NACE code,

nd half of these firms (48.4%) list secondary codes withinheir main industry. Three other results stand out. First,inance/Business Services was frequently used as the sec-ndary code. This finding may be explained by the legaltructure of some firms; a substantial number of firms arerganised as holding companies or may have service activ-ties and financial and/or business services related to theirrimary activity. Second, the category of Trade and Restau-ants was listed as the secondary industry by 31% of firms.his may be because some trading activities are allocatedo a specific department that engages in wholesale tradingeyond the narrow needs of the company. Third, Buildingnd Construction seems to be related to Agriculture, Fish-ries, and Raw Materials. A relatively large share of Buildingnd Construction firms list Agriculture, Fisheries, and Rawaterials as their secondary industry (12.5%), and a large

hare (15.9%) of Agriculture, Fisheries, and Raw Materi-ls firms list Building and Construction as their secondaryndustry. The most plausible explanation is that the raw

aterials part of the industry is related to the Building andonstruction industry in vertical organisations.

Although there may be some justification for theypothesis that some firms assign more than one industryode, this cannot be the primary reason for misclassifica-ions because the vast majority of firms identify only oneode. Of the firms with two or more codes, the majorityist a secondary code within their own industry or withinnance/business services or trade/restaurants.

. Conclusions

The purpose of the present paper is to deepen our under-tanding of one of the fundamentals of industry analysis:hether and to what degree we can rely on industry codes


to characterise industrial structures. There is a debate aboutthe appropriateness of industry codes as descriptions offirms’ activities. As a contribution to this debate, the cur-rent paper demonstrates the extent to which firms that arecategorised as belonging to one main sector in the econ-omy conduct activities that, according to a strict distinctionbetween manufacturing and services, actually “belong” toanother main sector.

The analysis shows that two-thirds of the economicactivity in the private sector involves services. These activi-ties, however, do not only occur in firms classified as servicefirms by industry codes. Service firms do not generate theirturnover only from service activities; conversely, a consid-erable proportion of manufacturing firms conduct serviceactivities. We have shown that the statistical classificationof firms (using the NACE classification codes, in this case)does not reflect all of these firms’ self-reported activities.Although this information is not new, this analysis con-tributes to a quantification of this phenomenon as well aspossible explanations.

At a 2-digit level of aggregation, 18% of ‘misclassifica-tions’ were identified across industries. In three industries,Building and Construction, Agriculture, Fisheries and RawMaterials, and Industry, a large share of activities are clas-sified by the firms as services; 33%, 25% and 24% of thefirms in these industries, respectively, claim that the bulkof their activities are services rather than production. Thereis heterogeneity both among and within industries. Somefirms fully belong to their assigned industry, whereas oth-ers seem to perform large parts of their activities in otherareas.

Findings were robust when considering that some firmsmay have more than one industry code. The results didnot change when the analyses were performed at a moredetailed level of aggregation or when the Building andConstruction industry was omitted. The results were alsoroughly similar in a complementary survey that asked morespecific questions about the revenues from end-productsrather than broad questions about activities.

Any classification system involves benefits and draw-backs. The basic criterion for a classification technique hasa major impact on how firms are grouped. Thus, we shouldnot expect any classification system to be perfect. In fact,there are limits to how precisely we are able to make ourstatistics, even if we pour in more resources. As Einsteinnoted, ‘Not everything that counts can be counted, and noteverything that can be counted counts.’

Different explanations have been provided for why thestatistical classification of firms based on the current sys-tem does not correspond to the activities reported byfirms. One challenge in the current classification systemsrelates to the nature of firms’ activities. Because firmsperform multiple activities that are largely and increas-ingly interwoven, it is difficult to obtain an adequateand aggregated picture of the share of different types ofactivities in the economy. Thus, service activities in theeconomy are not limited to the activities of service firms.
Additional explanations note difficulties in classificationstemming from trends such as outsourcing, captive compa-nies and the inertia in changes to categories in the statisticalsystem.

e and Ec
Livesey (2006) contends that a revision of the statisticalsystem is needed and notes that an obvious opportu-nity to perform this revision was the so-called Allsoppreview (2004) of the use of the statistical system for policymaking. This review provided an opportunity for radicalchanges and supplementary modifications to the exist-ing statistical system. However, the author claims thatthere is inherent resistance to such changes. The result-ing recommendations in the Allsopp review were modest,proposing the addition of codes to capture emerging indus-tries and to provide greater regional disaggregation. Theresistance to change in classification systems is explainedby a basic dilemma. On the one hand, it is necessary toadjust classifications to changes in the economic, tech-nological and organisational methods of production. Newproducts emerge, and the importance of existing productschanges; likewise, the organisation of industry undergoessubstantial changes. In a dynamic economy, the statisti-cal system must be a work in progress. On the other hand,continuity is necessary to maintain long-range data and acorresponding outlook.

Moreover, changing statistical classifications is not asimple process. Basically, one could pursue the avenue ofincreasing detail and the categories in the statistical sys-tem, or some of the categories could be lumped together.Both strategies could be defended, however, currentlythere seem to be an inexpedient skewed distribution inthe coverage of economic activities. Generally, the serviceindustry is not described at a detailed level, but alsowithin manufacturing there are areas that are detailedin the statistics although not very important in real life(Hicks, 2011; Graham et al., 2007). Apparently this is eas-ily resolved. However, request for historical continuity inour data is one hindrance to this process. Another is the factthat, at least in Europe, the classification system is under EUlegislation, which complicates agreement and implemen-tation of changes. Any extensive recoding of classificationswould also involve substantial costs. Although such modi-fications may enhance the precision of the current system,they would not eliminate the basic problems in assigningindustry codes, nor would they address the fundamentalquestion raised in this paper and in the literature regardingthe criteria for classification.

Two fundamental questions raised by this study areat what point in the value chain the statistical systemshould measure activities and whether the system shouldmeasure only marketable products and services or activ-ities in a broad sense. Another fundamental issue is thatstatistical classification systems are designed to groupfirms in roughly homogeneous classes, but with the exten-sive international division of labour and specialisation,many firms’ products and services are increasingly het-erogeneous. Seemingly simple products, including rawmaterials, are increasingly specific. In the extreme case,this makes each firm unique and complicates the sta-tistical grouping of firms. Dalziel (2007) shows that apurely activity-based statistical classification system may
be inadequate. She proposes a system that considers theneeds to which firms respond rather than their activi-ties alone. A demand-oriented classification system suchas GICS (Global Industry Classification Standard) would,

however, entail a number of other difficulties and chal-lenges similar to those indicated here.

A number of actors, such as researchers and statisticalbureaus, extensively utilise industry statistics. Economet-ric and statistical analyses of industry structures may betechnically sophisticated and include high levels of statis-tical significance, but if the basis of these analyses (i.e., theway data are organised) is not precise, there is a risk ofover-interpreting the results. Despite the potential bias inresults from the way data are produced this is rarely in theradar of economists. In particular, classification issues areclose to absent in economic papers; the available data andtheir classifications are taken for granted.

The problems pointed to here and in other studies are,of course, well known at statistical offices. An ad-hoc sur-vey with simple techniques apparently reveals a number ofdeviations between the self-perception of firms regardingtheir activities and the statistical coding of these firms.Aside from explanations above it may be hypothesised thatthe perception of ‘activities’ in the survey is understoodamong respondents as not only related to production andvalue added but a broader range of activities including forexample marketing and distribution. There is, however, nocertain knowledge on whether there are differences in theperceptions with respondents in official statistics and ad-hoc surveys as in this case. Another explanation relatesto the above-mentioned delay (and absence) in changesof industry codes. To ensure that the statistical classifica-tion system is ‘on track’ regular quality checks should beperformed, and the overarching reluctance to change clas-sifications because of the need for historical time seriesshould not be exaggerated.

The main finding of the present study is not that indus-try classifications are completely misleading and should bediscarded; rather, these classifications inevitably involvetradeoffs regarding their categorisation of firms. We mustbe aware of the potential flaws and limitations in the pre-cision of the data if analyses or policies involve activitiesthat span sectors and firms. Finally, there is a clear needto supplement industry classifications with more in-depthstudies of the true range of activities of the firms that oper-ate within these industries. According to McCarthy (1995),a classification system should ideally be accurate, stable,timeless and general. Although this is an ideal, probablyunrealistic, state of affairs, there seems to be room for mov-ing further towards that situation.

Future research on this issue could extend the cover-age of countries beyond Denmark to obtain more preciseknowledge on the performance of statistical systems inother countries.

Acknowledgements

Comments on earlier drafts of this paper are grate-fully acknowledged from Ina Drejer, Dept. of BusinessStudies, Aalborg University; Jon Sundbo, Roskilde Uni-versity; Brian Wixted, Simon Fraser University, Canada;
Margaret Dalziel, Telfer School of Management, Universityof Ottawa; Andre Lorentz, Max Planck Inst. of Economics;Michaela Trippl, Vienna University of Economics and Busi-ness Administration; Ron Boschma, Utrecht University;

e and Ec

NStt

A

InAiwn

ttWuaBi

tcees

Ipticiiistdreicwr

amcoAwi‘fn1tnD


icola Wall, Manchester Business School; Ole Berner,tatistics Denmark; and Søren Schiønning Andersen, Statis-ics Denmark. Responsibility for any errors rests solely withhe author.

nnex A. Statistical classification systems

Industry codes, such as the NAICS (North Americanndustry Classification System), ISIC (United Nations Inter-ational Standard Industrial Classification of Economicctivities), NACE, and GICS, are used in most industry stud-

es for research and the design of policy programmes asell as within private businesses (e.g., marketing compa-ies).

The NAICS was established in the late 1990s to replacehe SIC as a reaction to the inexpedient inconsistencies inhe previous classification system (Clarke, 1989; Kahle and

alkling, 1996; Bhojraj et al., 2003). The SIC was widelysed prior to its replacement by the NAICS/NACE, evenfter problems related to the SIC had been recognised. Thus,hojraj et al. (2003) found that more than 90% of the studies

n leading academic journals used SIC codes.The statistical classification system has evolved over

ime.10 The NACE system is used by more than 150ountries. Because of its wide use and its inclusion in thempirical section of this paper, the NACE system will bexplained in more detail (for a more detailed explanation,ee, e.g., Eurostat, 2008).

In 1961, NACE (initially called NICE) was implemented.t included only manufacturing, construction and energy-roducing companies, but it was subsequently extendedo cover more types of firms. By 1970, all economic activ-ties were covered. To ensure comparability with otherlassification systems, a convergence towards the ISIC wasnstituted and implemented in 1990. After a minor revisionn 2002, a major international revision of NACE (rev.2) wasmplemented in 2007 based upon work to revise the ISICince 2000 (rev.4). NACE is based upon ISIC in that it is iden-ical at the higher level of aggregation, but NACE is moreetailed at lower levels of aggregation. The purpose of theevision was to adjust the statistical system to the currentconomies, specifically to include and denominate new
ndustries and to enhance harmonisation across statisticallassification systems, such as NACE, ISIC, and NAICS,ithout losing the ability to review data historically. The
evised version of NACE attaches greater importance to the

10 The census of manufacturers in the U.S. was implemented in 1810fter several limited versions in the preceding two decades. By theid-1850s, statistical systems had begun to emerge in several western

ountries. In Denmark, the case country in this paper, the productionf statistics was confined to population counts in 1769, 1787 and 1801.n organisation for producing statistics was established in 1797, but itas dismantled in 1819. In the 1830s and 1840s, there was an increas-

ng demand in Western Europe for statistics on public finances (the termstatistics’ was then commonly used for facts about the state), and an officeor compiling statistics was established in 1833. After several years witho statistics, the statistical office ‘Statistics Denmark’ was established in850. The industrial development in the second part of the century accen-uated the need for industrial statistics. However, these statistics wereearly non-existent in Denmark until the end of the century (Statisticsenmark, 2000).


production process in defining the individual classes. Ata higher level of aggregation (groups and divisions), thecharacter of the goods and services produced is moreimportant. Thus, the criteria for categorisation includethe character and use of the goods and services and theinputs and technologies used. The relative importance ofthese criteria varies by the industry category and type ofproduction (Eurostat, 2008).

Annex B. Additional robustness analysis

A further robustness check uses a disaggregated case,that of knowledge-intensive business services. In an addi-tional survey of the total population of Danish firms withat least 10 employees, respondents were asked about thenature of their activities. In this survey, 732 firms par-ticipated, rendering a 53% response rate. We calculatedthe percentage of the turnover in the firms identified asknowledge-intensive service firms (based on NACE codes),which was generated from the sale of services. One out offive firms (19%) characterised as service firms in the offi-cial industry classifications generated less than half of theirturnover from the sales of services. On these grounds, thesefirms may be misclassified. The results show that eventhough this complementary survey asked specific ques-tions about the turnover from end-products or servicesrather than broad questions about activities, the overallnumbers of firms that fall outside of the expected classi-fication code are comparable across both surveys. At thismore disaggregated industry level, there are marked dif-ferences between sub-sectors. Especially within IT services,30% of the firms ascribe less than 50% of their turnover tothe sales of services, and only half of these firms generate76–100% of their turnover from services. This may be dueto the combination of IT services with different types ofhardware, illustrating the issue of interwoven activities.

References

AEGIS, 2002. Selling Solutions: Emerging Patterns of Product-ServiceLinkage in the Australian Economy. Australian Expert Group in Indus-try Studies, University of Western Sydney, Sydney, Australia (JaneMarceau, Nicole Cook, Bronwen Dalton, Brian Wixted).

Allsopp, C., 2004. Review of Statistics for Economic Policy Making. HMT,London.

Baines, T.S., Lightfoot, H.W., Benedettini, O., Kay, J.M., 2009. The servitiza-tion of manufacturing. A review of literature and reflection on futurechallenges. J. Manuf. Tech. Manag. 20 (5), 547–567.

Bhojraj, S., Lee, C.M.C., Oler, D.K., 2003. What’s my line? A comparison ofindustry classification schemes for capital market research. J. Account-ing Res. 41 (December (5)), 745–774.

Carlsson, B., 1989. The evolution of manufacturing technology and itsimpact on industrial structure: an international study. Small Bus. Econ.1, 21–38.

Carroll, P., Pol, E., Robertson, P.L., 2000. Classification of industries by levelof technology: an appraisal and some implications. Prometheus 18 (4),417–436.

Christensen, J.L., Drejer, I., Vinding, A., 2004. Produktudvikling I danskfremstillingsindustri. ACE Notat nr. 8, Aalborg Universitet, Aalborg,Denmark.

Christensen, J.L., Dahl, M.S., Eliasen, S., Nielsen, R.N., Østergaard, C.R., 2011.
Patterns and collaborators of innovation in the primary sector: a studyof the danish agriculture, forestry and fishery industries. Ind. Innov.18 (2), 203–225.
Clarke, R.N., 1989. SICs as delineators of economic markets. J. Bus. 62 (1),17–31.

http://refhub.elsevier.com/S0954-349X(13)00041-6/sbref0005

















































































































































































e and Ec
Coombs, R., Miles, I., 2000. Innovation, measurement and services: thenew problematique. In: Metcalfe, J.S., Miles, I. (Eds.), Innovation Sys-tems in the Service Economy. Measurement and Case Study Analysis.Kluwer, Boston, pp. 85–103.

Cubbin, J.S., 2001. Market structure and performance. In: Lesourne, J., Son-nenschein, H. (Eds.), Industrial Economics I. Foundations of Pure andApplied Economics. Routledge, London.

Dahlstedt, R., Salmi, T., Luoma, M., Laakkonen, A., 1994. On the usefulnessof standard industrial classifications in comparative financial state-ment analysis. Eur. J. Oper. Res. 79, 230–238.

Dalziel, M., 2007. A systems-based approach to industry classification. Res.Policy 36 (10), 1559–1574.

Djellal, F., Gallouj, F., 2001. Patterns of innovation organisation in servicefirms: portal survey results and theoretical models. Sci. Public Policy28, 57–67.

Eurostat, 2008. NACE Rev.2. Statistical classification of economic activitiesin the European Community. Methodologies and Working Papers. EC.

Fertuck, L., 1975. A test of industry indices based on SIC codes. J. Financ.Quant. Anal. 10 (5), 837–848.

Fuchs, V.R., 1965. The growing importance of the service industries. J. Bus.38, 344–373.

Gallouj, F., Weinstein, O., 1997. Innovation in services. Res. Policy 26,537–556.

Gershuny, J., Miles, I.D., 1983. The New Service Economy. The Transfor-mation of Employment in Industrial Societies. Francis Pinter, London.

Good, I.J., 1965. Categorisation of Classification. Mathematics and Com-puter Science in Medicine and Biology. HMSO, London, pp. 115–128.

Graham, J.M., Davis, S., Lippoldt, D., Mann, C.L., Triplett, J., 2007. The Mea-sure of a Nation: Quantifying Innovative Strength through ImprovedService Sector Metrics. NBR Special Report 11. National Bureau ofAsian Research, Seattle, Washington.

Greenfield, H.I., 1966. Manpower and the Growth of Producer Services.Columbia University Press, New York and London.

Guerriere, P., Meliciani, V., 2005. Technology and international competi-tiveness: the interdependence between manufacturing and producerservices. Struct. Change Econ. Dynam. 16, 489–502.

Guenther, D., Rosman, A.J., 1994. Differences between COMPUSTAT andCRSP SIC codes and related effects on research. J. Account. Econ. 18(1), 115–128.

Hicks, D., 2011. Structural change and industrial classification. Struct.Change Econ. Dynam. 22, 93–105.

Howells, J., 2006. Where to from here for services innovation? Paper forKISA Conference, Sydney, March.

Howells, J., 2004. Innovation, consumption and services: encapsulationand the combinatorial role of services. Serv. Ind. J. 24, 19–36.

Jacobs, G., O’Neil, C., 2003. On the reliability (or otherwise) of SIC codes.Eur. Bus. Rev. 15 (3), 164–169.

Kahle, K.M., Walkling, R.A., 1996. The impact of industry classifications onfinancial research. J. Financ. Quant. Anal. 31, 309–335.

Leiponen, A., Drejer, I., 2007. What exactly are technological regimes?Intra-industry heterogeneity in the organization of innovation activ-ities. Res. Policy 36, 1221–1238.


Livesey, F., 2006. Defining High Value Manufacturing. IfM, University ofCambridge, Cambridge, U.K.

Lodefalk, M., 2013. Servicification of manufacturing – evidence fromSweden. Int. J. Econ. Bus. Res. 6 (forthcoming).

Mathe, H., Shapiro, R.D., 1993. Integrating Service Strategy into the Man-ufacturing Company. Chapman & Hall, London.

McCarthy, I., Anagnostou, A., 2004. The impact of outsourcing on the trans-action costs and boundaries of manufacturing. Int. J. Prod. Econ. 88,61–71.

McCarthy, I., 1995. Manufacturing classification: lessons from organiza-tional systematics and biological taxonomy. Integr. Manuf. Syst. 6 (6),37–48.

McKelvey, B., 1982. Organizational Systematics: Taxonomy, Evolution,Classification. University of California Press, Berkeley.

Nählinder, J., 2005. Innovation and Employment in Services. TheCase of Knowledge Intensive Business Services in Sweden.Linköping Studies in Arts and Sciences, Linköping University(Ph.D. Thesis).

Nordås, H.K., Kim, Y., 2013. The role of services for competitiveness inmanufacturing. In: OECD Trade Policy Papers No. 148. OECD Publish-ing.

OECD, 1999. Business and Industry Policy Forum on the Service Economy.OECD, Paris.

Peneder, M., 2003. Industry classifications: aim, scope and techniques. J.Ind. Compet. Trade 3 (1/2), 109–129.

Pilat, D., Wölfl, A., 2005. Measuring the interaction between manufactur-ing and services. In: OECD Science, Technology and Industry WorkingPapers, 2005/5. OECD Publishing.

Pilat, D., Cimper, A., Olsen, K.B., Webb, C., 2006. The changing nature ofmanufacturing in OECD economies. In: OECD Science, Technology andIndustry Working Papers 2006/9. OECD Publishing.

Preissl, B., 2000. Service innovation: what makes it different? Empiricalevidence from Germany. In: Metcalfe, J.S., Miles, I. (Eds.), InnovationSystems in the Service Economy. Measurement and Case Study Anal-ysis. Kluwer, Boston, pp. 125–148.

Santamaría, L., Nieto, M.J., Miles, I., 2012. Service innovation inmanufacturing firms: evidence from Spain. Technovation 32,144–155.

Schettkat, R., Yocarini, L., 2006. The shift to services employment:a review of the literature. Struct. Change Econ. Dynam. 17 (2),127–147.

Statistics Denmark, 2000. Dansk Statistik 1850–2000 (Danish Statistics1850–2000) (Copenhagen).

Sundbo, J., 2001. The Strategic Management of Innovation: A Sociologicaland Economic Theory. Edward Elgar, Cheltenham, UK.

Sundbo, J., Gallouj, F., 1998. Innovation in Services. SI4S Synthesis PapersS2. STEP, Oslo.

U.S. Department of Labor, 2001. Report on the American Workforce.Vandermerwe, S., Rada, J., 1988. Servitization of business: adding value by

adding services. Eur. Manag. J. 6 (4), 314–324.Venkatesan, R., 1992. Strategic sourcing: to make or not to make. Harv.

Bus. Rev. 70, 98–107.



























































































































































































































































































































































































































































































































































































































































































































The ability of current statistical classifications to separateservices and manufacturing

Business

Transcript of The ability of current statistical classifications to separateservices and manufacturing