Journal of Economic Surveys

ORGANIZATIONAL CHANGE, SKILL FORMATION, HUMAN CAPITAL MEASUREMENT: EVIDENCE FROM ITALIAN MANUFACTURING FIRMS

Authors


Abstract

Abstract This paper emphasizes the role of labour demand as a determinant of human capital formation. After a section in which the alternative conceptions on the functioning of labour markets are presented and different ways of measuring human capital are compared, an applied analysis is carried out in which we provide a labour-demand-oriented measure of human capital, as defined by the amount of specific skills firms generate through work-based training (WBT) activities. By merging three rich firm-level datasets, we estimate the impact of a set of variables supposed to affect both the propensity to invest in WBT and the intensity of training within the Italian manufacturing industry over the period 2001–2005. Special attention is devoted to the variables characterizing within-firm organization of knowledge, organizational change and the formation of competence pipelines: among them, innovation, internationalization commitment, out-sourcing and new hirings. The estimates show that the effect of innovation on WBT is higher when the introduction of new technologies is supported by organizational innovations. When looking at the nature of WBT, we investigate the different determinants of the firms' propensity to provide both in-house and outside training. We measure training intensity in terms, respectively, of the number of provided training activities, private and total training costs and share of trainees.

1. Introduction

The aim of this paper is to develop a theoretical framework for the measurement of human capital. Differently from the standard human capital approach, which is based on labour-supply-oriented measures, i.e. mainly schooling, education and experience, our approach relies in a labour demand perspective.1 From an empirical point of view, this approach results in an analysis of the drivers of firm-provided work-based training (WBT), as a measure of human capital.

We address these problems by first developing a conceptual framework in which human capital is acquired not only at school, but primarily on-the-job, and, second, by estimating training propensity and intensity on a sample of manufacturing firms operating in Italy over the period 2001–2005.

The structure of the paper is the following. In Section 2, we derive two different ways of measuring human capital by describing two main theories of how labour markets work in generating and allocating working skills. Particular attention is devoted to the job-competition model as opposed to standard wage-competition models of the labour market and on how the organization of knowledge within the firm may play a relevant role in affecting both the firms' decision to train and the amount of training to provide. In this respect, we identify four main drivers of WBT: technological and organizational innovation, internationalization activities, the out-sourcing of production activities and, finally, the previous recruitment of new personnel.

Relying on this theoretical framework, in Section 3 we describe the data we use for the following empirical analysis, while in Section 4 we present a set of empirical models for estimating the impact of those variables that are supposed to affect both the propensity to invest in WBT and the intensity of training within the Italian manufacturing industry. In Section 5, we discuss the results of the estimates. In particular, our exercise seems to show that innovation impacts more on WTB than international commitment. In particular, technological and organizational innovations seem to be complementary strategies that positively affect firms' propensity to train. Finally, Section 6 concludes.

2. The Conceptual Framework

The issue of how to measure human capital from a labour demand perspective can be tackled by referring to at least two different theories of the labour markets.2

The first – the wage-competition model – is the dominant one, while the second – the job-competition model – is a less frequently adopted framework of analysis, in particular with respect to its implications for the enhancement of the notion and measurement of human capital. In the following two sections we briefly describe how the two models work in allocating skills and in determining the patterns of labour incomes.

2.1 Two Different Conceptions of Labour Market Functioning

In standard economic models the match between labour demand and supply is based on the wage-competition mechanisms. At each market level, potential employees compete among each other for a job on the basis of a wage bidding device. When labour supply exceeds labour demand, potential employees underbid the prevailing level of wage, until equilibrium between demand and supply is reached. For each single vacancy, the adjustment process works in the same way and employees' selection occurs on the basis of wage underbidding.

Potential employees are assumed to be homogeneous, so that individual characteristics, such as education, previous working experience, sex, race and the like, are irrelevant. Potential employees differ from one another only as far as their individual reservation wages are concerned. Even when the vacancy is specifically addressed to a well-defined professional profile, hiring occurs on the basis of a mechanism of wage competition, in which the supply side is defined and limited with respect to the skills attached to the specific professional profile. The market for skills overlaps the market for labour services.

The adjustment process works differently according to Thurow's (1975)‘job-competition model’. In this case a pivotal role is played by the labour demand side. This means that the employer behaviour and firm's characteristics become more relevant.3 Therefore, in this model the stress is on the match between employees' characteristics and the firm needs. In order to maximize the return of this match, employers rank the applicants on the basis of the desired individual and specific traits. These are conceived as proxies of their degree of trainability and capability to adapt to the requirements attached to each specific job, and therefore to their training costs. Definitely, this ranking depends on the idiosyncratic features of the job and cannot be uniquely defined for all firms. For instance, in certain jobs previous working experiences can be relevant in defining the individual position in the ranking, whereas, in other job positions, ranking mainly depends on the education or other specific characteristic. However, in any case the employer puts together a queue of applicants which runs from the most favoured applicant to the least preferred one on the basis of perceived training costs.

As far as wage setting is concerned, in the job-competition model the wage level is exogenously determined with respect to the process of both hiring and individual positioning in the labour queue. Basically, one can point out two different types of drivers in the process of wage setting: external and internal drivers. External drivers refer to factors which affect a labour market as a whole: in particular, centralized bargaining can determine the minimum wage attached to each specific job position. As far as the internal factors are concerned, the structure of each internal labour market, and the career paths and the hierarchy of relative wages, define the wage level within the firms.4

In synthesis, we can identify that differences between the standard model and the job-competition model are threefold and concern, respectively: (i) the wage setting process; (ii) the process of skill formation and development; and (iii) the role of labour demand.

2.1.1 The Wage Setting Process

In the standard model wages are endogenously determined by the interaction between supply and demand, whereas in the job-competition model the process of wage determination is exogenous with respect to the dynamics of hiring.

However, it is important to emphasize that in the wage setting process the firm's organization of knowledge, its incentive schemes and, overall, the structure of each internal labour market affect deeply individual wage determination.

2.1.2 The Process of Skill Formation and Development

In standard models, the labour market provides the employer with the skills needed. Of course, this does not mean that firms do not train their employees. The traditional Beckerian framework (Becker, 1964), based on the distinction between general and specific training, provides the standard interpretation of the process of skill formation and development within firms. Nevertheless, in the wage-competition approach recruitment and skill formation are activities related to two different and distinct processes, as opposed to the job-competition approach, in which recruitment occurs on the basis of indicators of trainability.

In the latter, skill formation and development are processes inherently linked to firm activities; i.e. a minimum amount of training is always needed, even for the most menial job positions. The match between labour demand and supply is always mediated by the expected dynamics of skill formation and development, which adapts the employees' characteristics to the firm's technology and organization of knowledge and work. Skill formation and development allow the establishment of complementary relationships between the techno-organizational framework of the firm and the skills of the single employee.

2.1.3 The Role of Labour Demand

The structure of labour demand, determined by the organization of knowledge within the firm and its techno-organizational framework, does not play any significant role in the standard model. In the job-competition model, on the contrary, labour demand plays a central role both in the process of wage setting and in the process of skill formation and development.

When one refers to the organization of knowledge and the techno-organizational framework, three elements have to be emphasized. First of all, we have to take due account of the functioning of internal labour markets and, especially, the organization of knowledge and work, the architecture and the hierarchy of job positions, the capability to introduce and exploit organizational innovations and the structural characteristics of the firm (e.g. size, industrial specialization, location). Second, we should consider the technological setting and the capability of firms to innovate both their products and productive processes. Third, one should consider the entire value chain in which the firm operates, thus looking also at international trade activities, like export, foreign market penetration, off-shoring and market-oriented strategies.

All these three factors constrain the individual productivity, which is not only a characteristic of the individual employee, but also, and mainly, the outcome of a multi-layer process influenced by the organization of knowledge, the techno-organizational structure, the process of matching between the individual characteristics of the employees and the internal labour market. Finally, it is important to observe that the level of employment takes a Keynesian flavour in the job-competition model, since, given the level of wages, labour demand sets the amount of employees hired.

2.2 Human Capital Measurement in a Labour Demand Framework

In economic theory the concept and measurement of human capital has been mostly conceived and developed in the framework of the wage-competition model, with special emphasis on labour-supply-based proxies.5 The reference to the Smith compensating differentials principle as a keystone of human capital theory is particularly revealing in this respect.

However, the pragmatic concept has remained rather general and open to a demand side contribution to its formation as the following sentence shows:

Just as physical capital is created by making changes in materials so as to form tools that facilitate production, human capital is created by changing persons so as to give them skills and capabilities that make them able to act in new way. (Coleman, 1990, p. 304)

The same is true when it is defined as

The knowledge, skills, competencies and attributes embodied in individuals that facilitate the creation of personal, social and economic well-being. (OECD, 2001, p. 18)

In devoting time and funds to the development of human resources individuals and households, firms and governments carry out forms of activities which are undertaken not only for the sake of present benefits (consumption, availability of required skills), but also for future pecuniary returns (investment) and non-pecuniary advantages (risk aversion and precautionary behaviour).

Many activities, like, for instance, schooling, training, experience, mobility and migration, health and sport, influence the stock of human capital the individual, the firm and the society are endowed with and affect the quality of available labour services.

Accordingly, several groups of variables can be devised in trying to describe the nature and scope of human capital. However, the measurement of human capital can be best conceived in the framework of Petty's ‘political arithmetic’. This implies the capability to implement a theoretically based and empirically robust systematic set of accounts in which ‘number, weight and measure’ can be assigned to human resources. Such a complete set has not yet been developed and different measurement frameworks still coexist.6

2.2.1 The Original Question

It is well known how relevant has been the notion of capital by Fisher (1906) in making possible the ‘human capital revolution’7 and gradually overtaking Marshall's objections to it. In Fisher's framework, every stock of assets existing at a given point of time which allows a flow of services (payments) over time can be defined as capital. The value of each type of capital is given by the present value of the specified stream of payments over time. Time is conceived as an undifferentiated input.8

A totally different view was held by Alfred Marshall, according to whom human capital as a pure private good was not a realistic and logically consistent concept. In order to support his influential view he made reference to the end of the slavery system, to market failures, but especially to the nature of knowledge: non-storable, non-numerable, non-additive, highly idiosyncratic and conceivable as a set of connections based on conjectures.

Even if the notion of human capital was already recognized in the English mercantilist thought and above all in the Smithian vision,9 and in spite of the attention given to the quality of human resources by well-known economists,10 the accomplishment of the human capital theory had to wait till the second half of the 20th century (Blaug, 1972; Antonelli, 2003).

The notion of human capital developed by this theory falls in between Smith's reproducibility framework, in which no absolute limit exists to productivity increases when human knowledge improves, and an absolute scarcity framework, in which talents are limited by nature. An explanation of the process of human capital formation is provided. And this tends to differentiate it with respect to a purely ‘virtual’ notion of human capital that can be found in endogenous growth models as well as in real business cycle models.

2.2.2 Human Capital Measurement in the Two Different Conceptions

Considered from this particular perspective, one of the main outcomes of the human capital theory is the measurement of human capital as a magnitude with the following attributes: expressed in monetary terms; extrapolated from actual income profiles; derived ex ante; stressing the potential side of human capital.

Human capital is conceived as a monetary stock which is accumulated over the life cycle by individuals. It is assessed starting from the inspection of the life-time profile of the actual earnings of working employees (extrapolated measure). The amount of stockpiled human capital is determined by the rational behaviour of the individuals aiming at maximizing their earnings streams (ex ante measure). For each individual the life-time profile of the actual earnings depends

  • 1on the life-time profile of capacity or potential earnings (potential nature);
  • 2on the amount of human capital employed over time for producing further human capital.

When we come to examine the measurement of human capital in the job-competition model, three remarks can be made. At first sight, no monetary measure of human capital is directly designed. Background characteristics can be seen as a vector of variables expressed in real terms and identifying proxies of needed competences by the firms. The vector of job structure and characteristics describes the direct requirements of labour demand, again expressed in real terms, and is the key determinant of every human capital formation.11

In this respect, we can observe that in the job-competition model the measurement of human capital leads to a magnitude with the following attributes: derived from actual job characteristics; expressed as a vector of variables in real terms; derived ex post; stressing the actually used side of human capital.

In Section 2.2.3, we will explore to what extent a monetary measure of human capital can be derived also in the job-competition model. In any case, it should be clear that the job-competition approach involves the measurement of human capital on two distinct levels. The first level involves the individual characteristics of employees and their endowment of education, previous working experience, functional flexibility, capability to work in a team. This is a direct measurement of individual human capital, not mediated by the performance of the labour markets since, as observed in the previous section, the process of wage setting is constrained by both the system of centralized bargaining and the binding structure of relative wages fixed in the internal labour market.

The second level takes account of the stock of human capital measured at a firm level. As previously stated, the creation and development of human capital in firms rests on the techno-organization framework, the internal organization of knowledge and the internal dynamics of skill formation and development. This means that the human capital stock depends on four different sets of drivers.

  • (a) The internal labour markets, the organization of labour and the coordination of tasks among job positions. These factors define the productivity of the single employee, along with the aforementioned individual characteristics. For instance, the employer's strategy to adopt one or more of the so-called high productivity working practices such as task/job rotation, teamwork and so on define the range and the boundaries of each specific job position.
  • (b) The technology and the propensity of the firm to innovate. The stock and kind of machineries, capital equipment and, in general, the technology adopted by the firm constrains the skills which are actually used and their return in the work process. Any innovation introduced by the firm enriches (upskilling), but it can also decrease (deskilling), the skills required for the performance of production.
  • (c) The dynamics of skill development and the process of adjustment of individual characteristics to the techno-organizational framework. The strategies implemented by the firm for the employees' training affect directly the endowment of human capital available.
  • (d) The economic network to which the firm belongs, intended as the position of the firm in the overall value chain. The positioning of the firm in the value chain defines the level of dynamic transaction costs, i.e. ‘the costs of not having the capabilities you need when you need them’ (Langlois, 1992). These costs can affect the firm's decision to out-source a specific stage of production, when the costs implied by the process of human capital formation are too high for the firm to meet. In a sense, one can say that the position of the firm in the value chain limits the amount of investment in human capital that the firm can sustain and, hence, it fixes a ceiling to the stock of human capital. On the other side, the integration or the sub-contracting of production activities to external suppliers may determine the complexity of the production process and the number or quality of tasks to be performed at the workplace, thus affecting the amount of skills required and the amount of training to be provided. In both the theoretical and empirical literature, the role played by these variables has been analysed in depth. In particular, knowledge-intensive business services can make the process of out-sourcing convenient as they are key actors in the process of conversion of institutionalized knowledge into tacit knowledge, through the interactions between firms and different types of consultants.12 Consequently, knowledge-intensive business services can affect positively the amount of specialization advantages deriving from out-sourcing and various forms of delocalization (Mazzanti et al., 2008), favouring the process of learning required for the implementation of these strategies through different practices of WBT.

2.2.3 A Closer Examination

An increasing awareness is gaining ground according to which it is difficult to talk about a high qualification level of human resources without taking into account continuity and modular experience in skill formation. Moreover, production sectors, filières and territorial contexts are deemed relevant in order to assess the evolution over time of learning processes.

These considerations lead us to take into consideration the organization of knowledge in the economic system.13 How can we conceive the interactions and trade-offs originated by the organization of knowledge internal and external to the firm? In our view, settings and changes linked to the internal organization of knowledge are better explained in the framework of the job-competition model and the notion of ‘competence pipeline’.

Along these lines we may envisage two ways in which the mark left by the division of labour14 in terms of labour inputs and outputs can be detected and (ex post) measured.

  • (a) A disaggregated pipeline can be conceived, through which we may be able to assess the differential value in the cumulative flow of labour services due to skill improvement and personally delivered by each single employee working in a given firm for a certain period of time.15 A relevant determinant of this value is the expenditure in WBT jointly faced by the firm and the single employee in the definite period of time.
  • (b) In a parallel way, an aggregated pipeline can be conceived, through which we may be able to assess the differential value in the cumulative flows of labour services due to skill improvement and collectively delivered by the whole set of different employees working in a certain interval of time within a firm, sector, filière, territorial or network context.16 A relevant determinant of this value is the expenditure in WBT jointly faced by firms and employees within the boundaries of the selected unit of analysis.

If we examine, for each of these two dimensions, the overall outcome in terms of skill differential incorporated in the different vintages of the competence pipeline, it is possible to derive two indirect monetary measures of human capital. Namely, we may envisage a two-tier measurement of human capital, both for the disaggregated and aggregated pipeline, taking into consideration both its input and output side.

With reference to dimension (a), the input-oriented measure of human capital is the compound expenditure in WBT jointly faced by the firm and the single employee in a certain period of time, while the output-oriented one is the compound increment in productivity17 generated over time by each employee, plus the opportunity cost of the time spent by him in non-rewarded training activities.

In version (b), the input-oriented measure of human capital is the compound expenditure in WBT jointly faced by firms and employees in a certain interval of time within the boundaries of the selected unit of analysis, while the output-oriented one is the compound increment in productivity generated over time by the whole set of employees working in a certain interval of time within the boundaries of the selected unit of analysis, plus the opportunity cost of the time spent by them in non-rewarded training activities.

Particularly when input-oriented measures in both dimensions are concerned, caution must be stressed from the beginning. Several aspects can increase the difficulty of deriving genuine measures of the overall expenditure in WBT. Among them the most problematic to deal with are the following: the co-financing nature of many training programmes in Europe and the dominant role of public contribution in many instances; the high differentiation of skill categories, which implies the availability of a large spectrum of formal and informal training activities;18 systemic impacts and biases determined by the interaction of the main features of the economic and training environment in which firms operate with the predominant policy approaches.19

In any case, these are measures with the attributes previously stressed: derived from actual job characteristics; expressed as a vector of variables in real terms; derived ex post; stressing the actually used side of human capital.

The notion of competence pipeline is able to prop up the measurement of human capital from the labour demand side in that it allows us to think in terms of ‘core borings’ of the human capital passed through the firm and, to a certain extent, still available according to its past and present needs.

2.2.4 Work-based Training Definition

WBT plays a crucial role as a determinant of human capital formation on the demand side. In our definition it includes all the training activities accomplished by the employees and performed after the end of the schooling period under the (total or partial) responsibility of firm management.20

Even though this contribution is based on a different approach with respect to the seminal Becker analysis (1964), especially as far as the process of skill formation and the role of structural variables are concerned, the notion of WBT has some similarity with the Beckerian notion of specific and on-the-job training.

In the applied analysis, we will distinguish between training provided within the firm (in-house training), training provided from outside (outside training) and training provided within the firm through the placement of workers side-by-side with other colleagues (coaching).

WBT activities are not only a direct way to form and accumulate specific skills, but also a measure of the mismatch between the amount of skills acquired by an individual before entering the labour markets and the amount of skills required by the job, or the tasks, in which she/he is employed.

However, in a job-competition framework, skill mismatch may depend on (i) a ‘pathological disconnection’ between the educational system and the economic system in the demand and supply of knowledge and skills (like in the wage-competition framework); (ii) a ‘physiological’ division of labour between the two systems in the process of skill formation.

Human capital investments, in this respect, are treated by the firms much like material capital investments: normally they have to face at least part of their cost in order to reduce the mismatch. The measurement of WBT could run in parallel to the measurement of years of schooling in the human capital model, a measure of human capital that is conceived in a labour supply framework. In our case it represents a first step for sketching a more complex model.

3. Dataset Creation

Relying on the theoretical framework presented in Section 2, we now turn our attention to the available descriptive and econometric evidence on the factors that affect the training decisions of Italian manufacturing firms. The higher degree of homogeneity of manufacturing firms, with respect to those operating in the services sector, has been a key factor in the choice to focus the empirical analysis on this kind of firms. The implication is that, given the different nature of market products and production processes, one cannot expect to apply a priori the conclusions drawn from the analysis of the manufacturing firms to the services ones.

Our data have been obtained through the merge of three distinct data sources: (i) the IX Survey on manufacturing firms (Indagine sulle Imprese Manifatturiere), carried out by Unicredit-Capitalia21 for the period 2001–2003; (ii) the Excelsior reports (Sistema Informativo Excelsior), fulfilled by the Research Centre of Italian Chambers of Commerce (Centro Studi Unioncamere) for the period 2003–2005; (iii) the Observatory on the balance sheets of joint-stock companies (Osservatorio sui bilanci delle società di capitale), created by InfoCamere on behalf of Centro Studi Unioncamere. To our knowledge, this is the first attempt to join the three data sources, so that our empirical analysis can be considered as completely new in the framework of the empirical studies on the Italian manufacturing industry.

The IX Survey on manufacturing firms gathers information on a representative sample of 4289 firms over the period 2001–2003. While firms with 500 or more employees correspond to the universe, firms employing between 11 and 499 employees represent a sample stratified by size, geographical area and sector of economic activity.22

Table 1 shows the structure of the Unicredit-Capitalia sample by firm size and geographical area.

Table 1.  The Structure of Unicredit-Capitalia Sample by Size and Geographical Area.
Firm size%Firm size%Geographical area%Geographical area%
11–2022.1511–50 51.69 Northwest 35.91  North 66.03
21–5029.5451–249 36.93 Northeast 30.12  Centre 17.65
51–24936.93≥250 11.38 Centre 17.65  South 16.32
250–499 5.27   South 16.32  
≥500 6.11      
Total  100.00 100.00 100.00

This survey is of particular importance since it offers a rich set of information on manufacturing firms' characteristics and activities. In our paper we focus on the sections concerning firm size; industrial specialization and geographical location; labour force composition by occupation; technology, as given by firms' investments in new equipment, product and process innovation, techno-organizational innovation; firms' internationalization activities, with particular reference to export, production off-shoring and the purchase of business services from abroad; market-oriented activities, like the out-sourcing of goods and services.

The Excelsior reports are issued by Unioncamere, in cooperation with the Italian Ministry of Labour and Social Welfare and the European Social Fund, and provide information on the year-by-year labour demand of a sample of about 100,000 privately owned firms with more than one employee and distributed all over the Italian territory (Centro Studi Unioncamere, 2007).

The sample refers to 27 sectors, primarily belonging to the manufacturing industry, but ranging also over the agricultural and the public sectors.23 Data on labour demand are collected following two criteria: the International Standard Classification of Occupations-88, on the one side, and the Excelsior classification, on the other. The latter accounts for the individual worker's level of competence, as defined on the base of the complexity of the tasks operated at the workplace, and the degree of skill specialization, where this degree is determined by the interaction between the knowledge content of tasks and their operating context, i.e. the industry.

The Excelsior dataset is composed of different sections which provide data collected by the local and regional Chambers of Commerce and other administrative archives or through direct interviews to firms carried out in years 2003, 2004 and 2005. The sections selected for the empirical analysis concern (i) the annual stock of the labour force at the end of each year (in 2003, 2004, 2005) and the annual entry and exit flows of labour by occupation;24 (ii) the volumes of WBT (in 2003, 2004, 2005), with particular reference to the number of trainees, the costs and the place of training activities supplied.

Finally, the third data source provides information on the balance-sheet variables relative to joint-stock companies active in the period 2001–2003. The Observatory on the balance sheets is managed by Unioncamere on the base of the information contained in the National Register of firms, in which all Italian stock companies are recorded. This dataset represents a unique source of data since it covers the whole population of Italian stock companies: hence, it allows handling more than 600,000 balance sheets every year and analysing their main financial indicators.

After merging the three data sources, the final sample contains 1545 manufacturing firms with 11 or more employees and active all over the period 2001–2005. Using the Eurostat definition, we obtain the following size composition of the sample: 20% small firms (11–49 employees); 54% medium firms (50–249 employees); 26% large firms (from 250 employees onward). Seventy-six per cent of the firms are geographically located in the north of Italy, while 15% are located in the centre and 8% in the south.

Traditional (46%) and specialized suppliers (31%) seem to be the sectors counting the highest number of firms, while the science-based sector (6%) is less populated.25Table 2 summarizes the firms' distribution by size and geographical area of localization.26

Table 2.  Final Sample Structure by Size and Geographical Area (Raw%).
 NWNECentreSouthTotal
Small110965545306
%35.9531.3717.9714.71100.00
Medium315288125134862
%36.5433.4114.5015.55100.00
Large1661245730377
%44.0332.8915.127.96100.00
Total5915082372091545
%38.2532.8815.3413.53100.00

Table 3 compares the 2001 distribution of firms in our Capitalia-Excelsior-Observatory (CEO) sample with the one emerging from the official 2001 Census on manufacturing industry, provided by the National Statistical Office (ISTAT, 2001), with respect to firm size and geographical area.

Table 3.  Sample Coverage: CEO and Census Data, 2001.
 ISTAT (%)CEO (%)
Size
Small8720
Medium1154
Large226
Area
North5571
Centre2115
South2414
Total100100

As can be clearly seen, the main bias in our sample concerns firm size. In particular, while the 2001 Census data show that only 13% of firms belong to the medium-large size category, in our sample this percentage rises up to 80%. A similar picture emerges when looking at the macro-area in which firms are located: while around 76% of the entire firms' population is located in the Centre and North of Italy, in the CEO sample this fraction rises to 86%.

We identify three possible causes of such a bias. As Tables 1 and 3 show, a first cause can be traced back to the characteristics of the Unicredit-Capitalia sample, which tends to under-represent the 2001 population of small firms (around 52%) and to over-represent that of medium-large firms (a bit more than 48%) with respect to the picture emerging from Census data.

A second cause lies in the fact that both in the Unicredit-Capitalia dataset and in the Excelsior one, small firms are selected by sampling, while the population of large firms interviewed represents the universe. The implication is that, when merging the two datasets, the probability to exclude small firms is higher than the probability to exclude large firms.

Finally, the third cause is due to the higher mortality rate of small firms with respect to large firms. When dealing with longitudinal data, the probability for a small firm to survive is generally lower than that for a large firm, as recent empirical studies on firms' demographic evolution and determinants of firms' survival rate clearly show (Agarval and Audretsch, 2001; ICE-ISTAT, 2007). Moreover, smaller firms are also more subject to mergers or take-overs, i.e. activities that tend to change their average dimension towards larger sizes.

However, our merged sample is representative of those firms that, over the period under investigation, constitute the ‘backbone’ of the Italian manufacturing industry, particularly when referring to activities like technological and organizational innovation, R&D and international trade. In other words, the sample is representative of those firms that show the highest capacity to introduce and manage innovation processes as well as to penetrate foreign markets.

In this respect, for instance, the data from the Community Innovation Survey 3 (CIS3) (European Commission, 2004) show that medium and large firms, although less numerous in absolute terms, constitute the highest fraction of firms, respectively, introducing product and/or process innovations; introducing successful innovations; employing more workers in innovation activities; gaining the highest profits from innovations. In this context, Table 4 compares the distribution of firms in the CIS3 (in 1998–2000) and in the CEO sample (2001–2003) by the type of innovation-related activity. Although some slight discrepancies still persist, mainly due to the different time span considered, the picture described in the CEO sample seems to be in line with the one described in the CIS3. The emerging result is that medium and large firms in 2001–2003 not only invest more in R&D and innovation activities, but, in these sectors, tend to employ more labour than smaller firms.

Table 4.  Firms and Employment Distribution in Italian Manufacturing (%).
Typology of firmCIS3 (1998–2000)CEO (2001–2003)
SmallMediumLargeSmallMediumLarge
  1. Source: Table IT.2B, p. 163 (European Commission, CIS3) and authors' elaboration on CEO dataset.

Innovative376078567484
Non-innovative634022442616
Total100100100100100100
R&D316179345860
Non-R&D693921664240
Total100100100100100100
New equipment677179859096
No new equipment33292115104
Total100100100100100100
 
 Employment share
 
Innovative416283587560
Non-innovative593817422540
Total100100100100100100

A similar picture can be derived when looking at the international trade volumes of Italian manufacturing firms. Data from Capitalia (2005) and ICE-ISTAT (2007), for instance, show that the weight of medium and large firms in foreign expansion activities is not only greater than the one of small firms, but is also increasing over time. This occurs not only when looking at export data, but also when looking at the trends in off-shoring, foreign direct investments and foreign-controlled firms.27

4. Descriptive Analysis

In this section we provide a descriptive analysis referring to our sample and concerning the main features of the 2005 labour demand predictions by type of education (Table 5) and experience required (Table 6) for the year 2006. In Table 7 the characteristics of the training activities provided in 2004 and 2005 are portrayed.

Table 5.  Labour Demand by Firm Size and Level of Education Required in Firm Predictions (Year 2006).
SizeEducation requiredNo.%Cumulated %
LargePrimary education19112.512.5
Regional vocational training362.314.8
Vocational training1519.824.7
Secondary school56536.961.5
College degree59038.5100.0
Total1533100.0 
MediumPrimary education29822.222.2
Regional vocational training654.927.1
Vocational training19314.441.5
Secondary school55841.683.1
College degree22616.9100.0
Total1340100.0 
SmallPrimary education5334.934.9
Regional vocational training95.940.8
Vocational training2516.457.2
Secondary school5536.293.4
College degree106.6100.0
Total152100.0 
Table 6.  Labour Demand by Firm Size and Type of Experience Required in Firm Predictions (Year 2006).
SizeExperienceNo.%Cumulated %
LargeFully specific51933.933.9
Sector specific48631.765.6
General22914.980.5
No experience29919.5100.0
Total1533100.0 
MediumFully specific32424.224.2
Sector specific50938.062.2
General17312.975.1
No experience33424.9100.0
Total1340100.0 
SmallFully specific2415.815.8
Sector specific5838.253.9
General1912.566.4
No experience5133.6100.0
Total152100.0 
Table 7.  Firms Providing Training Programmes to their Employees by Size.
SizeTraining20042005
No.%No.%
Large Yes29778.7829879.05
 No8021.227920.95
 Total377100.0377100.0
Medium Yes36442.2337543.50
 No49857.7748756.50
 Total862100.0862100.0
Small Yes8126.477626.50
 No22573.5323073.50
 Total306100.0306100.0
Total sample Yes74248.0374948.48
 No80351.9779651.52
 Total1545100.01545100.0

With respect to the first feature, as Table 5 shows, a positive correlation is evident between education required in firm predictions and firm size, especially if one focuses on the demand for graduates and for employees with only primary education. While 38.5% of the new labour demand in large firms concerns individuals with a college degree, medium-sized firms are primarily interested in workers with a secondary school degree (41.6%), leaving only 16.9% of their recruitment to people with a college degree. This picture is even more emphasized for small firms, the interest of which is more heavily concentrated on secondary school leavers (36.2%) or vocational training programme leavers (22.3%).

With reference to the previously acquired experience, Table 6 shows that, in general, the need for experience increases with firm size: the demand for totally inexperienced individuals is much higher for small firms (33.6%) than for large firms (19.5%). Considering the type of experience required, large firms prefer to hire employees with fully specific (33.9%) and sector-specific (31.7%) experience, whereas small- and medium-sized firms are more willing to employ employees with sector-specific experience (38% and 38.2% respectively).

Next to the propensity to train, we also provide some preliminary evidence on the intensity of training in Italian manufacturing firms. In particular, we focus on the average share of trainees and on average training costs per trainee. In Figure 1, we present some graphical statistics for the average share of trainees by firm size, propensity to innovate and propensity to export. As expected, large firms train a higher fraction of their workforce both in 2004 and in 2005 in comparison with small- and medium-sized firms. In addition, innovative and export-oriented firms tend to train a higher fraction of their workforce than non-innovative and domestic competitors. Finally, one should note that a declining trend seems to prevail with the share of trainees slightly decreasing over time, from an average of 16.3% in 2004 to an average of 15.6% in 2005.

Figure 1.

Mean Share of Trainees by Firm Size in 2004 and 2005.

Figure 2, instead, shows the trends in 2004 and 2005 of average nominal training costs. In particular, the upper box refers to total (nominal) training costs, while the lower one specifically refers to entirely firm-provided (nominal) expenditures. In a different way from before, medium-sized firms tend to spend more on average than small and large competitors. Innovative and export-oriented firms spend less than non-innovative and domestic counterparts only in 2004, the opposite being true in 2005. Finally, we note that total average training costs seem to increase over time, particularly if we refer to private expenditures.28

Figure 2.

Average Training Cost per Trained Employee in 2004 and 2005.

Finally, our dataset allows us also to identify three different forms of WBT: (a) in-house training; (b) training from outside; (c) coaching (affiancamento).

In particular, the first refers to training programmes provided within the firm through formal classes and courses, the second to formal training programmes that employees attend outside the firm boundaries, while the third concerns training activities like mentoring, learning side-by-side of other colleagues, or person to person experiential learning.

As Table 8 shows, in-house is the most preferred form of training by medium and large firms, while small firms prefer to train workers outside. Interestingly enough, large firms are also highly involved in providing training through coaching, while small and medium firms are more willing to provide formal types of training, although coaching is increasing from 2004 to 2005 for small firms only.

Table 8.  Forms of Training by Firm Size (%).
Size20042005
In-houseOutsideCoachingIn-houseOutsideCoaching
  1. Note: Percentages refer to the sub-sample of firms providing training in 2004 and 2005.

Large92.5982.4980.4792.2875.8477.18
Medium75.2770.6035.1675.2066.4032.27
Small51.8569.14 9.8859.2163.1611.84
Total79.6575.2050.5480.3769.8348.06

5. Estimating an Empirical Model

The main aim of the subsequent analysis is to estimate the patterns of skill formation in a labour demand framework. In particular, we assume that, next to standard ways to measure human capital from a labour supply perspective – primarily based on education and training acquired by individuals – an alternative measure of human capital is given by the amount of WBT firms provide to their workforce. WBT activities are not only a direct way to form and accumulate specific skills, but also a measure of the mismatch between the amount of skills acquired by an individual before entering the labour markets and the amount of skills required by the labour markets.

Measuring human capital from the labour demand side is not an easy task. Apart from a chronic lack of firm-level data on training activities, we face also the difficulty of measuring the level and quality of the competences required by firms, as well as the skill content of tasks. However, a first step in this direction may consist in exploring some measures of training intensity and in investigating which factors, among firms' strategies, activities and structural characteristics, significantly affect the training decisions of firms.

Using the CEO dataset, we investigate how firms' characteristics and activities carried out in 2001–2003 do affect training investments in 2004.

Keeping in mind the conceptual framework presented in Section 2, we now turn attention to the explanatory variables that, according to us and to the (limited) related literature (Baldwin and Johnson, 1995; Baldwin et al., 1995; Antonietti, 2007; Bassanini et al., 2007; Hollenstein and Stucki, 2008), may play a role in affecting firms' decision to train workers. In particular, we focus on activities like technological and organizational change, internationalization and market-oriented activities like out-sourcing.

In the base version of our empirical model, the dependent variable is given by the choice to provide training in 2004 (in general terms, or by type of training or by occupation). The explanatory variables concern firm size, industrial specialization, previous recruitment of personnel, technological and organizational innovation, out-sourcing and foreign expansion. All these variables are measured in the previous three years (2001–2003), so to avoid a priori simultaneity issues in the estimates.

With respect to the factors that affect our human capital variable, we particularly emphasize two aspects that characterize the ‘competence pipeline’ described in Section 2: the role played by the recruitment of new personnel and the role played by organizational change, the latter conceived both in terms of innovative activities and in terms of commitment to internationalization.

5.1 Estimates Based on the Logit Model

In estimating our empirical model we follow a two-stage approach. In the first stage we are interested in estimating the firms' propensity to invest in training in the year 2004, both in general terms and as referred to the different occupations and to the different places in which it occurs. Since we deal with binary dependent variables, we rely on a logit specification in order to estimate our model.

Our first empirical model is based on a standard logit specification of the type (Wooldridge, 1999)

image(1)

where Λ is the cumulative standard logistic distribution function of the random variable WBT. Through this equation we estimate the impact of a vector x of regressors on the probability for a firm i to invest in training (WBTi = 1) in the year 2004.

Since our dataset allows us also to distinguish the destination of training activities by occupation, we also specify our dependent variable in terms of WBT for the top and middle management (WBTM) and WBT for plant operators (WBTO).29

In addition, we also define three dummies measuring the three different places of training: WBT in-house for firms providing in-house training, WBT outside for firms providing training from outside, WBT coaching for firms training workers by placing them side-by-side with other colleagues. Table 9 summarizes the three sets of dependent variables that we consider for our logit estimates.

Table 9.  Training 2004 by Occupation and Place (% in parentheses).
WBTManagersPlant operators
  1. Note: aPercentages in parentheses refer to the total amount of firms providing each type of training respectively. In the first cell, 94.75% of firms providing in-house training involve managers, while 84.43% involve plant operators. This means that in-house training is almost common for both managers and machine operators.

In-house560 (94.75)a499 (84.43)
Outside525 (94.09)448 (80.29)
Coaching357 (78.98)335 (74.12)
Total WBT685 (44.34)591 (38.25)

As independent variables, we include (i) a set of controls and (ii) measures of organizational change that are supposed to underlie the training decisions of firms, as suggested by the theoretical framework and the economic literature. Again, all these independent variables refer to the period 2001–2003 in order to avoid problems of simultaneity with the training decision variable.

As controls we include30 (i) four geographical area dummies (Northwest, Northeast, Centre and South) in order to capture time invariant area specific spillovers; (ii) firm size as given by 2001–2003 average employment (natural logs); (iii) 14 industry dummies, according to the industry to which the firm belongs, which, instead, capture time invariant industry specific spillovers due to sectoral specialization; (iv) a variable measuring the capital intensity of production, as given by deflated net technical assets per employee, in natural logs (Log_K/L); (v) a variable reflecting the skill composition of the labour force, i.e. the natural log share of white collars (Log_WC/L); (vi) two dummy variables capturing, respectively, the acquisition of public funds for training in year 2003 (public) and the use of private funds for financing training in 2003 (private).31

As explanatory variables, we focus on the firms' propensity to hire new workers in year 2003 and on three activities that are supposed to capture organizational change: innovation, internationalization commitment and out-sourcing.

We expect that innovation positively affects the demand for new competences, particularly when it involves the design of new organizational assets and the need to develop new skills on the job. At the same time, one can conceive training as a tool on which the firm relies in order to increase the level of productivity after the introduction of new products, processes or organizational practices.

Similarly, internationalization activities, like export and the cross-border relocation of production activities, may require the firm to develop new competences, like managerial capabilities, language skills or foreign market knowledge. Moreover, training may be required when firms purchase advanced, knowledge-intensive, business services, like administrative services, accounting and bookkeeping, design, R&D or financial services and software.

Finally, the out-sourcing of low-skill-intensive phases of the production process, and the maintenance of high-skill-intensive phases at home, may require the continuous upgrading of existing skills through training activities.

Operationally, the recruitment of new personnel is measured by a dummy variable (hire 2003) which equals 1 if the firm hires new workers in the period 2001–2003 and 0 if not. Innovation is captured by three dummies concerning firm's investments in new machinery and equipment (investment), the introduction of new product and/or processes (inno_tech) and the introduction of organizational innovations due to the introduction of new products and/or new processes (inno_org).

Firms' internationalization is measured through three dummies: one equal to 1 if the firm engaged in exporting activities over the period 2001–2003 (export) and 0 otherwise, another equal to 1 if the firm moved production abroad (off-shoring) and the third equal to 1 if the firm purchased service activities from abroad during 2001–2003 (services).

Finally, out-sourcing is measured by a dummy variable (out-sourcing) equal to 1 if the firm contracted out activities that were previously integrated and 0 otherwise.

5.2 Ordered Logit Estimates on Training Intensity

After focusing on the propensity to train, in this second stage we estimate training intensity by first looking at the cumulative combination of different modalities of training according to the place in which it occurs: in-house, outside and side-by-side to colleagues through coaching.

We measure training composition by defining an ordered index (WTI) assuming a value equal to 0 for firms which do not supply any WBT programme, 1 for firms supplying only in-house training programmes, 2 for firms providing both in-house and outside training programmes, and 3 for firms providing all three forms of training programmes.

The basic idea is that the amount of training supplied can be thought of as a proxy of the mismatch between the skills acquired by workers through education and the skill required within the firm. In addition, WTI is a direct way for the firms to create or update technology-specific competences that are generally made obsolete by technological or organizational change.

As far as the index is conceived, WTI captures the degree of ‘training variety’ provided by a single firm. The higher WTI, the higher the diversification of the firm training portfolio, and thus the higher the number of sources and channels from which new knowledge is created and accumulated. This index can be also thought of as measuring training complexity, as firms providing different training programmes may require different training inputs, different types of trainers or administrative staff and are perhaps less able to benefit from economies of scale in training provision.

Table 10 shows the various modalities of training and the relative frequency distributions.

Table 10.  Firms' Distribution Among the Four Modalities of WBT.
Types of WBTNo.%Cumulative %
095461.7561.75
118411.9173.66
21509.7183.37
325716.63100.0
Total1545100.00 

Since our training index is an ordered categorical variable, ranging from 0 to 3, we use an ordered logit model as more appropriate for explaining variations in the training composition than a linear regression model. Whereas in the latter, a firm with a training index of 2 would be twice as training intensive as one with an index of 1, in the ordered logit model no such assumption of cardinality is made: a training index of 2 simply indicates more training patterns than a training index of 1.

The basic idea underlying the model is the existence of a latent continuous variable, WTI*, that denotes the degree of training diversification of a firm. The relationship between this latent variable and the set of explanatory variables is the following:

image(2)

where xi is the vector of regressors influencing the level of WTI and ɛi is the random error component drawn from a standardized normal distribution. Although WTI* is not observed, the integer index WTI is observed and related to WTI* by the following relationship:

image((3.1))
image((3.2))
image((3.3))
image((3.4))

where δi are the unobserved free threshold parameters (cut points) which define the boundaries between the different levels of WTI. Given the relationship between WTI and WTI* and the distribution of the error term, we can write the probability of observing a firm as having a zero value of WTI as

image(4)

where Λ is the standard logistic distribution function. Similarly, we can specify the other probabilities as

image((5.1))
image((5.2))
image((5.3))

Estimates are obtained by maximum likelihood. However, simply estimating β is of limited value, since we are not interested in E(WTI*|x) =x′β as WTI* is just an abstract construct. Instead, we are interested in assessing the response probabilities, i.e. the impacts of the explanatory variables on WTI. This information can be extracted from calculating the marginal effects, i.e. the effects that a change in the explanatory variables has on the cell probabilities. These effects can be written as

image(6)

where λ is the standard logistic density. Since the marginal effects vary with the levels of the explanatory variables, they are calculated at the mean values of x. While for continuous variables such marginal effects can be interpreted as elasticities, for dummy variables they represent changes in the predicted probabilities for unit changes from a status of 0 to a status of 1.

As before, we include a set of control variables and a set of additional variables capturing aspects of technological and organizational change. Moreover, in the ordered logit estimates on training composition, as well as in the estimates on training intensity, we define five new variables which measure, respectively, the intensity of recruitment at time t−1, the intensity of investment in new equipment, the intensity of innovation activity, the intensity of internationalization commitment and the intensity of out-sourcing.

The intensity of recruitment is given by the natural log share of new entrants in year 2003 (Log_hire2003). The intensity of investment in new machines and equipment is measured, instead, as the (natural log) of average 2001–2003 investments per employee, deflated by a 2001-based business investments price index (Log_inv).

As regards techno-organizational change, we first build an index of innovation intensity (inno_comb) that equals 0 for non-innovative firms (32.04%), 1 for firms that introduced a new product in the period 2001–2003 (19.81%), 2 for firms that introduced a new product and a new process (29.26%) and 3 for firms that, after product and process innovations, also introduced new organizational practices (18.90%).

Second, we define a foreign expansion index (FEI), as developed in Basile et al. (2003), in order to measure the degree of internationalization of the firm. This index integrates various dimensions of commitment to internationalization: exports, production off-shoring and the purchase of services from abroad.32 Thus, FEI ranges between 0 and 3: 0 for the firms not involved in any international trade activity (28.03%), 1 for firms that only export goods and services (52.36%); 2 for firms exporting and purchasing knowledge-intensive services from abroad (15.99%); 3 for firms exporting, purchasing services and off-shoring production abroad (3.62%).

Finally, we measure out-sourcing intensity by defining an index (out_int) that takes the value of 0 for firms developing all the phases of their production process internally (77.99%), 1 for firms out-sourcing one phase (13.92%), 2 for firms out-sourcing two phases (3.62%) and 3 for firms out-sourcing more than two phases (4.47%).

5.3 Estimates on Training Intensity

Finally, we estimate training intensity by relying on both real training costs per employee (in natural logs) and the (natural log of the) number of trainees over the total number of employees in 2004 (Log_trainees). With respect to the former, in particular, we explicitly distinguish between real total training costs (Log_TTC) and real private training costs (Log_TPC).33

Since we observe our dependent variables only for firms providing training in 2004, it is possible that firms with positive expenditure levels or positive shares of trainees are not randomly selected from the population, so that a sample selection may arise, which biases standard ordinary least squares (OLS) estimates.

In order to avoid such a problem, we employ a Heckman two-step selection model (Heckman, 1976, 1979). For this purpose, we first define TC* as our outcome of interest, i.e. training expenditures in 2004.34 This outcome is observed only if TC* > 0. We then introduce a latent variable, WBT*, so that it is observed only if WBT* > 0, i.e. only when firms do invest in training activities. Assuming a linear specification of the model, we obtain the following system of two equations:

image((7.1))
image((7.2))

with ɛ1 and ɛ2 jointly normally distributed and homoskedastic. The two-step method is based on the conditional expectation:

image(8)

in which inline image is the so called inverse Mills ratio and σ12 is the covariance between ɛ1 and ɛ2, and where inline image is obtained by probit regression of WBT on x1, and the second step OLS regression of TC on x2 leads to a semi-parametric estimate of (β2, σ12).

Therefore, we first estimate a probit model on the propensity to train in year 2004. Then we include such a probability into the second-stage estimation of training costs (and training share) in order to obtain consistent estimates of the parameters of interest.

As explanatory variables, we include the firm employment size, the skill intensity, the capital intensity, the share of previously recruited personnel, the previous use of public funds for financing training, the investments in new machinery per employee, the innovation intensity, the degree of engagement in foreign market penetration and the intensity of out-sourcing.

6. Main Results

Tables 11–13 show the results of the logit estimates on the propensity to invest in WBT in 2004, where WBT differs by occupation (managers and plant operators) and place of training respectively.

Table 11.  Propensity to Train 2004: Logit Estimates.
 Model 1Model 2Model 3
  1. Notes: Dependent variable WBT. The estimate of the coefficients refers to marginal effects at the sample mean. Heteroskedastic-consistent standard errors in parentheses. Intercept coefficients are not reported. Hosmer–Lemeshow goodness-of-fit test (H-L Gof test) does not reject the null hypothesis of correct specification of the models (p-values in parentheses).

  2. *Significant at 10%; **significant at 5%; ***significant at 1%.

Log_employees0.150  (0.017)***0.150  (0.017)***0.187  (0.017)***
Log_WC/L0.089  (0.029)**0.087  (0.029)**0.105  (0.029)**
Log_K/L0.013  (0.019)0.012  (0.019)0.013  (0.018)
hire 20030.031  (0.039)0.026  (0.039)0.033  (0.038)
private0.235  (0.033)***0.235  (0.033)*** 
public  0.048  (0.044)
investment0.114  (0.059)*0.110  (0.060)*0.126  (0.058)**
inno_tech0.024  (0.036)−0.008  (0.039)0.003  (0.038)
inno_org 0.073  (0.036)**0.069  (0.035)**
export−0.074  (0.046)−0.072  (0.046)−0.062  (0.045)
off-shoring−0.030  (0.051)−0.034  (0.052)−0.033  (0.049)
services0.082  (0.039)**0.083  (0.039)**0.083  (0.038)**
out-sourcing0.063  (0.040)0.060  (0.040)0.071  (0.039)*
Area dummiesYesYesYes
Industry dummiesYesYesYes
No. obs.141014101410
Pseudo R20.20650.20870.1842
H-L Gof test χ21383.56  (0.4831)1383.33  (0.4773)1391.42  (0.4166)
LogPseudolike−774.78982−772.64293−796.56206
Table 12.  Propensity to Train by Occupation 2004: Logit Estimates.
 WBTM (1)WBTM (2)WBTO (1)WBTO (2)
  1. Notes: The estimate of the coefficients refers to marginal effects at the sample mean. Heteroskedastic-consistent standard errors in parentheses. Intercept coefficients are not reported. Hosmer–Lemeshow goodness-of-fit test (H-L Gof test) does not reject the null hypothesis of correct specification of the models (p-values in parentheses).

  2. *Significant at 10%; **significant at 5%; ***significant at 1%.

Log_employees0.176  (0.018)***0.216  (0.018)***0.137  (0.016)***0.170  (0.016)***
Log_WC/L0.117  (0.030)***0.134  (0.030)***−0.002  (0.025)0.015  (0.026)
Log_K/L−0.008  (0.019)−0.006  (0.019)0.049  (0.018)**0.048  (0.017)**
hire 20030.027  (0.040)0.035  (0.038)0.018  (0.036)0.024  (0.034)
private0.250  (0.033)*** 0.168  (0.032)*** 
public 0.053  (0.044) −0.004  (0.038)
investment0.102  (0.060)*0.118  (0.057)**0.082  (0.056)0.093  (0.055)*
inno_tech−0.010  (0.040)−0.002  (0.039)0.016  (0.036)0.021  (0.035)
inno_org0.063  (0.036)*0.058  (0.036)0.080  (0.033)**0.078  (0.033)**
export−0.043  (0.049)−0.032  (0.048)−0.051  (0.044)−0.043  (0.044)
off-shoring−0.017  (0.053)−0.017  (0.050)0.001  (0.045)0.003  (0.044)
services0.057  (0.039)0.059  (0.039)0.029  (0.036)0.032  (0.035)
out-sourcing0.058  (0.041)0.070  (0.039)*0.030  (0.036)0.039  (0.036)
Area dummiesYesYesYesYes
Industry dummiesYesYesYesYes
No. obs.1410141014101410
Pseudo R20.24500.21720.18340.1687
H-L Gof test χ21379.84  (0.5037)1391.95  (0.4128)1416.15  (0.2496)1422.52  (0.2133)
LogPseudolike−731.03972−757.91239−768.21212−782.0354
Table 13.  Propensity to Provide In-house, Outside and Coaching Training.
 WBT in-houseWBT in-houseWBT outsideWBT outsideWBT coaching
  1. Notes: The estimate of the coefficients refers to marginal effects at the sample mean. Heteroskedastic-consistent standard errors in parentheses. Intercept coefficients are not reported. Hosmer–Lemeshow goodness-of-fit test (H-L Gof test) does not reject the null hypothesis of correct specification of the models (p-values in parentheses).

  2. *Significant at 10%; **significant at 5%; ***significant at 1%.

Log_employees0.197  (0.017)***0.226  (0.017)***0.127  (0.016)***0.164  (0.016)***0.195  (0.015)***
Log_WC/L0.064  (0.027)***0.077  (0.027)***0.081  (0.027)**0.098  (0.027)**0.019  (0.022)
Log_K/L0.008  (0.018)0.009  (0.018)0.007  (0.018)0.007  (0.018)0.019  (0.015)
hire 2003−0.012  (0.038)−0.006  (0.037)0.013  (0.035)0.020  (0.035)0.088  (0.025)***
private0.179  (0.033)*** 0.180  (0.031)*** 0.107  (0.026)***
public 0.044  (0.041) −0.015  (0.037) 
investment0.060  (0.058)0.074  (0.058)0.151  (0.050)**0.158  (0.048)**−0.075  (0.062)
inno_tech−0.009  (0.038)−0.001  (0.037)−0.032  (0.036)−0.023  (0.035)0.020  (0.029)
inno_org0.057  (0.034)*0.053  (0.034)0.065  (0.033)**0.063  (0.032)*0.018  (0.030)
export−0.070  (0.048)−0.063  (0.048)−0.041  (0.045)−0.030  (0.045)−0.020  (0.039)
off-shoring−0.044  (0.044)−0.043  (0.044)0.018  (0.047)0.020  (0.046)−0.065  (0.031)**
services0.086  (0.038)**0.088  (0.038)**0.037  (0.035)0.039  (0.034)0.014  (0.030)
out-sourcing0.074  (0.040)*0.083  (0.039)**0.031  (0.035)0.042  (0.035)−0.019  (0.029)
AreaYesYesYesYesYes
IndustryYesYesYesYesYes
No. obs.14101410141014101410
Pseudo R20.25240.23700.19640.17870.2793
H-L Gof test1371.561378.071389.981404.281450.49
(0.5664)(0.5172)(0.4273)(0.3252)(0.100)
LogPseudolike−701.54837−716.00523−743.69391−760.14362−615.76708

Relating to the former, we present three different specifications of the model. In the first (Model 1), after controlling for geographical location and industrial specialization, we focus on technological innovation, as the capacity of the firm to introduce, in the three years before training, new products and/or new processes. In the second specification (Model 2), instead, we focus on techno-organizational innovation, as the capacity of the firm to re-organize its production process after the introduction of a technological innovation. Finally, in the third specification (Model 3) we consider the two variables together.

As the estimates show, the choice to invest in WBT positively depends on firm size (increasing the labour force by 1% leads to an increase in the probability to train above 15%), skill intensity (around 9%), use of private funds for financing training (around 24%), investments in new machinery (around 11%), techno-organizational change (around 7%), and, more weakly, the purchase of services from abroad (more than 8%) and the out-sourcing of production phases to external suppliers (around 7%).

Interestingly, the use of public funds does not seem to stimulate the training propensity. The same is true for engagement in exporting activities and off-shoring. The previous recruitment of new personnel does have a positive impact, but not significantly different from zero.

With reference to organizational change, we note that only the adoption of new organizational practices, after the previous introduction of technological innovation, has an impact on WBT. Since our variable captures the joint occurrence of technological and organizational change, we speculate that technology only stimulates training when it is followed by an organizational change. Therefore, technological and organizational change may be considered as complementary inputs in the training decision of manufacturing firms.

When looking at international trade activities, only the acquisition of services from abroad seems to have a positive effect on WBT. The acquisition of knowledge-intensive business services, like, for instance, financial services, insurance, transport, R&D and design, may require firms to develop the necessary skills in order to operate them.

Finally, when dealing with market-oriented activities, the out-sourcing of ancillary production and service activities may allow firms to specialize on the high-skill-intensive phases of their value chain, thus stimulating them to invest more intensively in skill formation through WBT.

When we split the dependent variable by occupation, we find some interesting results, as shown in Table 12. First of all, the impact of firm size seems to be higher when firms train managers with respect to plant operators. One possible explanation is that managers working in large firms, other than processing a wider set of activities than managers working in smaller firms, have also to deal more intensively with activities like supervision, control, process design and work recruitment, that require a continuous upgrading of their skills.

As expected, skill intensity is important only when referring to the choice to train managers, whereas capital intensity is important when dealing with the decision to train plant operators. The latter piece of evidence, thus, seems to confirm the complementarity between physical and human capital at the workplace.

As before, the use of private funds increases both the propensity to train managers and the propensity to train plant operators, the former being particularly more affected (25%). On the contrary, the use of public funds in 2003 does not have any significant effect.

Techno-organizational change, instead, seems to stimulate more the propensity to train plant operators (8%) rather than managers (6%), while the purchase of knowledge-intensive services and out-sourcing seem to be relevant only when a more general propensity to train (managers, executives and operators) is relevant.

A look at the determinants of WBT by place of training adds some more information to the previous picture. As clearly emerges from Table 13, in-house training seems to be more chosen when high skills are available (more than 6%), when the firm previously utilized private funds for financing training activities (around 18%), when techno-organizational innovation occurs (around 6%), when the firm purchases knowledge intensive business services from abroad (around 9%) and when the firm contracts out phases of the production process that were previously developed internally (around 8%).

A slightly different picture seems to characterize the choice to provide training from outside. In this case, skills availability seems to gain importance (more than 8%) as well as techno-organizational change (more than 6%) and previous investments in capital-embodied technology (more than 15%). The availability of new machines as well as the introduction of new organizational practices, thus, play the most significant role in driving the firm decision to invest in external training courses.

Finally, training through coaching seems to be induced mainly by the previous recruitment of workers and by previous private spending, even if in the latter case the impact is lower than for in-house and outside training. In this respect, coaching can be thought of as a particularly useful form of training when the firm, after hiring new personnel, needs to make such new workers rapidly adapt to the technology and the organization of production, i.e. to the working environment. Put another way, while more formal types of training seem to be linked to complex activities, like techno-organizational innovations or international trade, coaching can be considered as a first step through which newly hired workers approach the firm and its organization.

Now we turn to the second stage, in which the ordered logit estimates aim at identifying to what extent the variables previously identified do affect the composition – or the heterogeneity – of WBT, assumed to be a proxy of firm-specific human capital.

Table 14 reports the estimated coefficients of the significant variables in the ordered logit regressions. Interestingly, the firm size, the share of high-skilled personnel and the intensity of innovation are the most significant drivers of WTI, while the recruitment of new personnel and foreign expansion activities are not significant. Estimates based on Model 1 also show that the index of innovation intensity is significant (p < 0.05). From this result we argue that the more heterogeneous the set of innovation activities in which the firm is engaged, the more heterogeneous seems to be the training portfolio implemented.

Table 14.  Training Index 2004: Ordered Logit Estimates.
 Model 1
  1. Notes: Heteroskedasticity-robust standard errors. The likelihood ratio (LR) test confirms that the hypothesis of proportional odds is not violated.

  2. *Significant at 10%; **significant at 5%; ***significant at 1%.

Log_employees1.021  (0.095)***
Log_WC/L0.403  (0.152)**
Log_K/L−0.008  (0.105)
private0.799  (0.162)***
Log_hire 20030.035  (0.086)
Log_inv−0.013  (0.064)
inno_comb0.185  (0.073)**
FEI−0.096  (0.106)
out_int0.167  (0.087)*
Area dummiesYes
Industry dummiesYes
No. obs.882
Pseudo R20.1814
Log likelihood−786.23683
Cut pointδ15.422  (1.165)
Cut pointδ26.207  (1.170)
Cut pointδ37.058  (1.176)
LR test for proportional odds χ2(50)54.04 (p-value 0.3227)

However, as stated before, the information contained in Table 14 is relevant just for suggesting the variables that are important in explaining individual heterogeneity in WTI among firms. Table 15, instead, reports the marginal effects of such variables on the three categories in which WTI > 0.

Table 15.  Marginal Effects on the Index of Training Composition 2004.
VariablesPr(WTI = 1)Pr(WTI = 2)Pr(WTI = 3)
Size0.066 (0.010)***0.077 (0.011)***0.097 (0.011)***
Log_WC/L0.026 (0.010)**0.030 (0.012)**0.038 (0.015)**
private0.049 (0.011)***0.060 (0.013)***0.079 (0.017)***
inno_comb0.012 (0.005)**0.014 (0.006)**0.018 (0.007)**
out_int0.011 (0.006)*0.013 (0.007)*0.016 (0.008)*

What clearly emerges is that the marginal impact of each variable on WTI increases when passing from one category to the following one. For our purpose this means that previous private spending, skill, innovation and – more weakly – out-sourcing intensity are drivers of training intensity. In particular, adding organizational change to technological innovation requires the firm to activate all the three forms of training, even if this impact is not particularly strong (1.8%). Interestingly, training intensity and complexity are also increasingly driven by the utilization of private funds, while public funds do not have any impact.

Finally, Table 16 presents the results of the Heckman selection models for unit training costs, both private (Log_TPC) and total (Log_TTC), and for the share of trainees (Log_trainees).

Table 16.  Training Intensity: Heckman Two-step Estimation.
 Log_TPCLog_TTCLog_trainees
  1. *Significant at 10%; **significant at 5%; ***significant at 1%.

Log_employees0.302  (0.179)*0.260  (0.170)0.372  (0.168)**
Log_WC/L0.603  (0.160)***0.580  (0.156)***0.516  (0.152)**
Log_K/L−0.013  (0.086)0.013  (0.082)−0.081  (0.081)
Log_hire 20030.061  (0.072)0.045  (0.69)0.064  (0.068)
Log_inv0.080  (0.064)0.057  (0.062)0.130  (0.060)**
public0.151  (0.185)0.246  (0.178)0.171  (0.177)
inno_comb0.159  (0.065)**0.133  (0.063)**0.119  (0.063)*
FEI−0.056  (0.082)−0.079  (0.078)−0.095  (0.078)
out_int0.010  (0.073)0.022  (0.069)0.085  (0.069)
Area dummiesYesYesYes
Industry dummiesYesYesYes
No. obs.911914914
Censored obs.452452452
Uncensored obs.459462462
Mills lambda1.387  (0.780)*1.176  (0.770)1.399  (0.745)*

Both the training costs and the share of trainees are positively affected by firm size, skill intensity and innovation intensity. With respect to the latter, in particular, we find that adding organizational changes, after the introduction of new product or process technologies, increases the intensity of training by an amount between 12% (for trainees) and 16% (for private costs). Investment in capital-embodied technology, instead, seem to positively affect only the relative number of trainees (13%).

Interestingly, previous spending of public funds does not seem to have any influence on future training activity. The same is true for internationalization, out-sourcing and the recruitment of new personnel in 2003.

Summing up, we find evidence that the more the production process is characterized by changes in its technological and organizational structure the more firms form and upgrade their workforce skills. However, while a relatively wide set of factors seem to affect the decision to train or not, few variables do also have an impact on the intensity of training. Once decided to train, the amount of training35 supplied depends on how intense is the process of techno-organizational change and on the stock of high-skilled personnel available.

7. Conclusions

In the present paper we provide a first theoretical and empirical attempt to measure human capital in a labour demand perspective.

From a theoretical point of view, we assume that a labour market characterized by job competition can be a useful framework in order to stress the importance of training as a type of firm-level activity devoted to the creation of new skills. In particular, we stress the concept of ‘competence pipeline’ as a useful theoretical tool for better describing how changes linked to the internal organization of knowledge lead to changes in the amount of specific human capital acquired by individuals on the job.

This concept is able to prop up the measurement of human capital from the labour demand side in that it allows us to think in terms of ‘core borings’ of the human capital passed through the firm and, to a certain extent, still available according to its past and present needs. At least four alternatives can be defined in order to pursue an applied analysis. In one of them, an aggregate pipeline and an input-oriented measure can be conceived, through which we are allowed to assess the cumulative value of the flows of heterogeneous labour services delivered by all the different employees working in each period of time and over a certain span of years within a sector, filière or territorial context. A relevant determinant of this value is the amount of WBT firms supply to their workforce in a given sector, filière or territorial context.

Relying on this framework of analysis, we merge three rich firm-level datasets and estimate the impact of a set of variables which are supposed to affect both the propensity to invest in WBT and the intensity of training within the Italian manufacturing industry over the period 2001–2005. In this respect, we devote special attention to technological and organizational innovation, international trade, the out-sourcing of production activities, the use of private versus public funds for financing previous WBT and previous recruitment of new personnel.

Our estimates show that, when looking at the probability to invest in WBT programmes in general, innovation impacts more than international trade, in particular when new technologies are followed by organizational innovations. When we disaggregate our dependent variable by occupation and by the place of training provided, we find that techno-organizational innovation, as well as capital-embodied technological change, seems to affect more the propensity to train plant operators than managers. These results do not resemble any other previous studies.

When looking at the form of WBT, we find that, apart from size and skill intensity, while out-sourcing and the purchase of business services positively affect the propensity to provide in-house training, techno-organizational change seems to drive the choice to train outside the firm. A coaching-like training activity, instead, seems to be the first form of skill formation strategy devised by firms, as it is mainly affected by previous recruitment. Finally we estimate training intensity in terms, respectively, of the number of training activities provided, private and total training costs and the share of trainees. Our results point to a positive and significant effect of skill intensity and innovation intensity, while no significant effect is found for the degree of commitment to internationalization.

This paper can be considered as a first exploratory exercise towards a more exhaustive and robust study of human capital measurement conceived in a labour demand perspective. Our present aim is to test, theoretically and empirically, if and to what extent WBT can be considered a proxy of human capital. In addition to that, we analyse which factors contribute to explain the heterogeneity of WBT among manufacturing firms. As schooling, background characteristics, ability and age can measure human capital from the point of view of the labour supply, firm's investments in new technology, in new organizational practices, in highly skilled personnel and in international trade activities can be taken as important determinants of human capital from the point of view of labour demand.

Acknowledgements

We are extremely grateful to all the participants at the workshops organized by Giorgio Vittadini and Piergiorgio Lovaglio in the framework of the Lombardy Regional Research Institute (IRER) research project ‘Human capital in Lombardy’. We also acknowledge with special thanks Unioncamere and, in particular, Claudio Gagliardi and Francesco Vernaci for their invaluable help in providing and creating the dataset used in this paper. A preliminary version of this paper has been presented at the ISA-RC33 7th International Conference on Social Science Methodology, Campus of Monte Sant'Angelo, Naples (Italy), 1–5 September 2008. All the usual disclaimers apply.

Notes

  • 1

    The answer to this question may have strong implications in terms of the evaluation of educational outcomes and quality. For a very preliminary account see Antonelli (2008).

  • 2

    A third view is given by job-matching theories of the labour market. Since we particularly stress the role of labour demand, for simplicity, we focus on the two polar theories, i.e. wage versus job competition. For a review of the different mechanisms underlying all these models see, for instance, Heijke and Muysken (2000).

  • 3

    The techno-organizational framework, the related organization of labour, and, more generally, all features concerning the internal labour markets and the institutional settings become key factors, as suggested, for instance, by Antonelli (2003) and Antonelli and Guidetti (2008a, b).

  • 4

    Additionally, as analysed in Villa's contribution (1986), the strategies implemented by employers as to hiring, training, division of labour, recruitment and structuring of career paths are affected by economic (especially, the product market) and technological constraints.

  • 5

    For an up to date survey see Folloni and Vittadini in this issue.

  • 6

    See, for instance, Le et al. (2003, 2006), OECD (1998), Oxley et al. (2008)Stroombergen et al. (2002) and Woessmann (2003).

  • 7

    Gary Becker published in 1964 the first edition of his well known book Human Capital. A Theoretical and Empirical Analysis with Special Reference to Education. However, Jacob Mincer published in advance at least two seminal works in human capital theory: ‘A study of personal income distribution’, his doctoral dissertation, in 1957, and an article on ‘Investment in human capital and personal income distribution’ in August 1958. Then he published the article ‘The distribution of labor incomes: a survey with special reference to the human capital approach’ in March 1970 and the volume on Schooling, Experience and Earnings in 1974. Among his more recent works we can find ‘Human capital: a review’, published in 1994, ‘Investment in US education and training’, also published in 1994, ‘Economic development, growth of human capital and the dynamics of the wage structure’ and ‘Changes in wage inequality, 1970–1990’, both published in 1996, and ‘The production of human capital and the life cycle of earnings’, published in 1997. Mincer did publish his first works earlier than Becker, even if he acknowledges his intellectual debt to Becker for the works published after 1957. Moreover, both these authors owe Gregg Lewis and T.W. Schultz a great intellectual debt. In this respect reference can be made also to Psacharopoulos (1987).

  • 8

    This is a definition dealing with a notion of virtual capital, rather than with real or physical capital.

  • 9

    We refer, in particular, to the notion of dexterity in his theory of the division of labour, to his perception of man as a costly and complex machine and to his compensating differentials theory.

  • 10

    Among them Nassau Senior and Arthur Cecil Pigou.

  • 11

    Lovaglio (2010), in this issue, develops a convincing methodology for the measurement of human capital. In his contribution, human capital is conceived as a non-directly observable multidimensional variable which can be measured as a latent variable. In this way both demand and supply factors can be taken into account.

  • 12

    See, for instance, Antonelli (2000).

  • 13

    To be precise, the method used in making knowledge relevant for economic utilization and in improving the skills of employees (Antonelli and Pegoretti, 2008).

  • 14

    Which is based on knowledge and skills actually used, given the prevailing organization of knowledge.

  • 15

    More specifically, over a certain span of the employee's life-cycle. In this respect, the unit of analysis is the employee.

  • 16

    More specifically, over a certain span of years. In this respect, the unit of analysis could alternatively be the firm, the sector, the filière, the territorial or network context (for instance, an industrial district or cluster).

  • 17

    Which in some circumstances could be approximated through the differential in labour incomes and profits.

  • 18

    Even if the different versions of the Frascati manual are relevant in this respect (see, for instance, Antonelli and Montresor, 2000).

  • 19

    For instance, the availability of a ‘dual system’ for vocational training, on the one hand (Ghisla et al. 2008), or the role of tripartite agreements in support of ‘high-performance work organization’ (Leoni, 2008).

  • 20

    It includes, for instance, activities like on-the-job training; off-the-job training; work-based learning; learning by doing, by interacting, by using.

  • 21

    Formerly Mediocredito Centrale.

  • 22

    Following ATECO 1991 classification.

  • 23

    In this work, however, we consider only the manufacturing industry.

  • 24

    Managers, executives/clerks, plant operators.

  • 25

    Following the Pavitt (1984) taxonomy.

  • 26

    Table A1 in the Appendix, instead, reports the structure of the merged sample by industry.

  • 27

    Tables A2 and A3 in the Appendix, for instance, show respectively some international data on the distribution of firms that exported goods and services in the period 2002–2005 and some data on the distribution of the value of exports within Italian manufacturing.

  • 28

    The variance of training expenditures is also increasing over time: the standard error for total average training expenditures is 902.27 in 2004 and 1430.06 in 2005, while for private training expenditures it increases from 875.94 in 2004 to 1415.81 in 2005.

  • 29

    Due to the presence of a high number of zeros (80%) in the dummy variable measuring top managers' training, we aggregate top and middle managers in a unique variable (56% of zeros).

  • 30

    Tables A1, A4 and A5 in the Appendix provide their description and summary statistics.

  • 31

    Information is also available on the amount of public and private funds used for financing training in 2003, but, due to the high number of missing data, we are not able to provide reliable estimates.

  • 32

    Services include transport, insurance, communication, financing, informatics, R&D and design.

  • 33

    Both total and private training costs are divided by the 2001-based GDP deflator.

  • 34

    The same approach is valid also for the share of trainees.

  • 35

    Unfortunately we do not have information on training hours.

Appendix

Table A1.  Sample Structure by Industry.
IndustryNo.%
15 – Food products and beverages1167.51
17 – Textile1187.64
18 – Wearing apparel442.85
19 – Leather, luggage, shoes624.01
20 – Wood (except furniture)362.33
21 – Paper and paper products392.52
22 – Publishing, printing and recorded media382.46
23 – Coke, petroleum products and nuclear fuel90.58
24 – Chemicals and chemical products905.83
25 – Rubber and plastics835.37
26 – Non-metallic mineral products925.95
27 – Basic metals664.27
28 – Fabricated metal products (except machinery)18211.78
29 – Machinery and equipment27217.61
30 – Office, accounting and computer machinery50.32
31 – Electrical machinery and apparatus664.27
32 – Radio, TV and communication equipment342.20
33 – Industrial process control equipment362.33
34 – Motor vehicles, trailers and semi-trailers332.14
35 – Other transport equipment251.62
36 – Other manufacturing, furniture, etc.996.41
Total1545100.00
Table A2.  Exporting Firms Distribution by Firm Size and Macro-regions.
 SmallMediumLarge
200020052000200520002005
  1. Notes: aManufacturing industry. Elaborations from ICE on ISTAT data.
    Source: ICE-ISTAT (2007), Tables 8.3 to 8.4. pp. 383–384.

World20.218.927.027.641.543.3
Europe20.318.927.427.940.842.9
Italya18.716.729.430.446.447.7
Table A3.  Value of Exports (millions of Euros) by Firm Size, Italian Manufacturing.
Firm size2002200320042005
Value%aValue%aValue%aValue%a
  1. Source: Our calculations on ICE-ISTAT (2007) data, Table 5.1.1, p. 323.

  2. Notes: aPercentage of the total value of exports; bpercentage of S+M+L.

Small (S)52.22619.851.15919.853.16819.155.05618.9
Medium (M)70.82626.971.45927.676.75827.580.35727.6
Large (L)112.41542.5109.85442.4121.51543.6126.03843.3
M + L183.24169.4181.31370.0198.27371.2206.39571.0
 (77.9)b (78.0)b (78.9)b (78.9)b
S+M+L235.46789.2232.47289.8251.44190.2261.45189.9
Total264.093100.0258.888100.0278.625100.0290.889100.0
Table A4.  Variables Description.
VariableDescription
AreaNorth West: Liguria, Lombardia, Piemonte, Valle d'Aosta
North East: Emilia-Romagna, Friuli Venezia-Giulia, Trentino Alto Adige, Veneto
Centre: Lazio, Marche, Toscana, Umbria
South: Abruzzo, Basilicata, Calabria, Campania, Molise, Puglia, Sardegna, Sicilia
Capital intensityReal net technical assets over total employees (natural log, average 2001–2003) (Log_K/L)
Industry14 industry dummies according to the ATECO 1991 classification (see Table A1): 15 – food products and beverages; 17 + 18 – textile and clothing; 19 – leather; 20 – wood; 21 + 22 – paper and printing; 23 – oil refining; 24 – chemicals; 25 – rubber and plastics; 26 – non-metal minerals; 27 + 28 – metals and metal products; 29 – non-electric machinery; 30 + 31 + 32 + 33 – office equipment, electric machinery, medical apparel; 34 + 35 – vehicles and other transportation; 36 – furniture and other manufacturing industries
Innovation- Capital-embodied technology: dummy = 1 if the firm invested in new machines and equipment in 2001–2003 (investment); real investments in new machines and equipment (natural logs, average 2001–2003) (Log_inv)
- Technological innovation: dummy = 1 if the firm introduced a product and/or a process innovation in 2001–2003 (inno_tech)
- Techno-organizational innovation: dummy = 1 if the firm introduced new organizational practices after the introduction of a new product or a new process in 2001–2003 (inno_org)
- Innovation intensity: number of innovations (product, process, techno-organizational) introduced in 2001–2003 (inno_comb)
Internationalization- Dummy = 1 if the firm exported goods in 2001–2003 (export)
- Dummy = 1 if the firm moved production abroad in 2001–2003 (off-shoring)
- Dummy = 1 if the firm purchased business services from abroad in 2001–2003 (services)
- Foreign expansion index: 0 if the firm is purely domestic; 1 if the firm engaged only in export activity; 2 if the firm exported and purchased business services from abroad; 3 if the firm exported, purchased services and off-shored production in 2001–2003 (FEI)
Out-sourcing- Dummy = 1 if the firm contracted out production or service activities in 2001–2003 (out-sourcing)
- Number of phases out-sourced in 2001–2003 (out_int)
Previous recruitment- Dummy = 1 if the firm recruited new workers in 2003 (hire 2003)
- Share of newly recruited personnel in 2003 (natural log) (Log_hire 2003)
Previous training financing- Dummy = 1 if the firm utilized own private funds for financing training in 2003 (private)
- Dummy = 1 if the firm utilized public funds for financing training in 2003 (public)
Skill intensityShare of white collars, computed as entrepreneurs + managers + executives + clerks (natural log, average 2001–2003) (Log_WC/L)
SizeNumber of employees (natural log, average 2001–2003) (Log_employee)
Training- Dummy = 1 if the firm provided any form of employee training in 2004 (WBT)
- Dummy = 1 if the firm trained managers and executives in 2004 (WBTM)
- Dummy = 1 if the firm trained plant operators in 2004 (WBTO)
- Dummy = 1 if the firm provided in-house training in 2004 (WBT in-house)
- Dummy = 1 if the firm provided training from outside in 2004 (WBT outside)
- Dummy = 1 if the firm provided coaching in 2004 (WBT coaching)
- Number of training forms provided: in-house, outside, coaching (WTI)
- Real private training costs per employee in 2004 (natural log) (Log_TPC)
- Real total training costs per employee in 2004 (natural log) (Log_TTC)
- Share of trainees in 2004 (natural log) (Log_trainees)
Table A5.  Summary Statistics.
VariableNo. obs.MeanSt. dev.Min.Max.
K/L142049109.3559044.12468.1653823142.8
Log_K/L142010.368540.9650616.14882113.62088
investments15450.91844660.273771701
inv13139174.85448132.3301526533
Log_inv13050.256241.3413652.05701114.23851
inno_tech154564919090.477377801
inno_org15450.37993530.477377801
inno_comb15451.3501621.11674803
export15450.82135920.383175301
off-shoring15450.10679610.308953901
services15450.21941750.41398601
FEI15450.95210360.763673103
out-sourcing15450.22006470.414424201
out_int15450.34563110.752987603
hire 200315450.73398060.442017601
Share_hire 200315450.07286970.135788202.955882
Log_hire 20031134−2.7423560.9197016−5.9295891.083797
private15450.41229770.492407601
public15450.1922330.394182801
WC/L15450.32112860.185477201.094254
Log_WC/L1534−1.289720.5932657−4.5027690.0900732
employee1545283.0928623.946612279.67
Log_employee15454.8377681.2062951.7917599.4157
WBT15450.48025890.499771901
WBTM15450.44336570.496943101
WBTO15450.38252430.486160901
WBT in-house15450.38252430.486160901
WBT outside15450.3611650.480493801
WBT coaching15450.29255660.455083801
WTI15450.81229771.15998903
TPC732198.504245.20912.8411373518.898
Log_TPC7324.7409241.1263361.0442048.165903
TTC742213.0186256.90832.8411373518.898
Log_TTC7424.8095861.1328931.0442048.165903
Trainee 200415450.16326330.259298401
Log_trainee742−1.5250961.06815−5.2870040

Ancillary