Introduction

The widespread use of Information and Communication Technologies (ICT) is one of the main distinguishing features of today’s economic activity (Jovanovic and Rousseau 2005; Jorgenson and Vu 2007). The reason for this is twofold: first, their direct contribution to increased productivity and economic growth; second, their indirect contribution resulting from the generation of complementary innovations that improve the economy’s Total Factor Productivity (TFP) (Pilat 2006; Jorgenson et al. 2011; Ceccobelli et al. 2012). From the perspective of the impact analysis of ICT investment on productivity and economic growth, empirical evidence shows the following: 1) the rates of return on digital investment are relatively much higher than those on investment in other physical components; and 2) the reason is that the digital investment and use often go hand in hand with other endeavours, usually human capital improvement and organisational and institutional change (Bresnahan et al. 2002; Arvanitis 2005). The link between innovation and ICT has been identified in literature as a set of internal knowledge externalities to explain company productivity (Venturini 2015). The accelerating point of this literature is established with the consolidation of ICT as the general purpose technologies (Venturini et al. 2013). However, ICT does not give rise to generalised productivity improvements until companies and their workers have achieved the required technological, educational/training, strategic, organisational, labour and cultural competencies. In other words, the role of ICT as a general-purpose technology needs organisational and business process changes to fully exploit its growth opportunities. In this context, the effects of ICT on company productivity are indirect. Complementary relationships (co-innovation) are established with other dimensions, in particular with human capital and workplace innovation. These spillovers are widely demonstrated in research using company data (for a review of this literature see Cardona et al. 2013; Díaz-Chao et al. 2015). Transition countries of Eastern Europe (EE) face considerable challenges in adapting their economies to effectively compete in regional and global markets. It is a key issue for them to find a way to increase their productivity and adapt their economies’ structure to global-knowledge competition, promote co-innovation and develop new goods and services that respond to the changing domestic and international demands. Thus, the impact of digital technological changes and their co-innovation processes on productivity is an important aspect in the region’s economic performance.

The main motivation behind this study is to evaluate the relationship among ICT, management practices, innovation and human capital, and to evaluate the influence of these variables on productivity in a sample of manufacturing enterprises in Eastern European countries. The following are the main questions underpinning the research: 1) Does the existence of new co-innovative productivity sources (the usage of ICT, workplace organisation and human capital) affect the performance of manufacturing enterprises in Eastern European countries? 2) What are the interrelationships in these complementary factors? 3) What are the differences in the relationships of complementarity (co-innovation) as the sources of productivity in Eastern European manufacturing enterprises?

The remainder of this paper is organised as follows: The literature review section presents a survey of empirical literature on the relationship among ICT, innovation and company productivity. The data section describes the Management, Organisation and Innovation (MOI) dataset and presents descriptive analysis of the sample. The empirical results section reports the results of the ordinary least squares (OLS) model, the structural equation modelling (SEM) model and the empirical findings. Finally, the conclusion provides a summary and policy implications based on those findings.

Literature Review - ICT, Innovation and Company Productivity

Much effort has been put into research to understand the so-called Solow Paradox concerning the limited evidence of the positive productivity impact of ICT (Jorgenson and Stiroh 1999). The importance of ICT is a much debated question with extensive literature focused on explaining and understanding their role in economic growth, productivity and efficiency. Significant progress has been noted since 1990 in the analysis of ICT and productivity. Most empirical studies have been performed at the microeconomic level of company and industry, examining the relationship among ICT and economic growth and productivity. At the macroeconomic level, fewer studies have been conducted because of a shortage of datasets specifically related to the ICT investment and usage and other relevant national characteristics.

Company-level analysis complements analysis conducted at the macro-level and enables researchers to better understand ICT diffusion effects and, especially, to more adequately reflect those quality changes brought about by ICT. Regarding the relationship between ICT and productivity at the company level, there are findings that ICTs alone are not enough to affect productivity (Bresnahan et al. 2002; Arvanitis 2005). Ignoring other complementarities may introduce bias to the analysis and overestimate or underestimate the effect of ICT on productivity. Significant changes within the company structure, such as a shift in the employment structure from low to high skills, the diffusion of ICT and the redesign of a company workplace organisation, can be observed over recent years and present new challenges for companies. Many authors have pointed out the importance of these interrelated changes as a shift towards a ‘new company paradigm’, which they have characterised using different labels: from a ‘mechanistic’ to an ‘organic’ company structure (Burns and Stalker 1994), from the ‘mass production model’ to the ‘flexible multiproduct company’ (Milgrom and Roberts 1990), or from a ‘tailoristic’ to a ‘holistic’ organisation of work (Lindbeck and Snower 2000). ICT potential will not be realised without business model changes and an increase in human capital and ICT skills (Bresnahan et al. 2002; Arvanitis 2005; Díaz-Chao et al. 2013). The main results that the international empirical evidence has suggested in relation to the new co-innovative sources of company productivity are presented in this section. The results are presented by country and date of publication, with special attention paid to the Eastern European countries.

Most of the company-level studies have been focused on highly developed countries. The empirical study for the United States (Bresnahan et al. 2002) formulated and confirmed the new theory of skill-biased technical change (SBTC). The authors have shown evidence of a positive correlation between ICT use and investment, workplace organisation and skilled labour which have affected productivity. Moreover, they have concluded that with the growing spread and access to ICT, investment in complementarities is crucial, particularly in skilled labour. Brynjolfsson and Hitt (2003) identified a set of new organisational practices in companies (freedom of information and communication, decision-making rights, performance-related incentives, and investment in education and training) that, together with digital innovation, are determinants in the explanation of productivity growth. Furthermore, with regards to the United States, there are Black and Lynch’s (2001, 2004) studies of manufacturing establishments showing that productivity growth during the 1990s has its source in workplace organisation changes and innovations (employee involvement, team work, incentive pay and decision-making autonomy) along with the diffusion of computers.

Furthermore, Autor et al. (2003) reinterpreted the role of technical progress in shaping labour markets and the argument of skill-biased technical change theory has been questioned. Routinization-biased technical-change (RBTC) states that ICT —and computers in particular— can replace labour in “routine” tasks (those tasks that involve step-by-step procedures and can therefore be codified) but not in “non-routine” tasks. This leads to job polarization of the labour force, when the middle-class jobs disappear more than the jobs which require a few skills and expert jobs, which require many skills. The recent studies have confirmed the routinization-biased technical-change relation to productivity growth (Jung and Mercenier 2014; Cortes 2016).

Investigations conducted in other countries followed the path of analysis initiated in the United States. In a sample of companies in Australia, Gretton et al. (2004) found interactions among the use of ICT and companies’ human capital, history of innovation, the use of advanced business practices and intensity of organisational restructuring (such as changes in the range of products and services, advertising, technical and on-the-job training, and business structure); they also observed a positive influence on productivity. Grimes et al. (2012) have confirmed that in New Zealand, adoption of broadband communications boosted company productivity by 7-10%. ICT effects appear consistent across the urban and rural locations, and across the high- and low-knowledge-intensive sectors. An analysis of panel data from British and French companies (Caroli and Van Reenen 2001) revealed that skilled workers adapt more easily to organisational changes. Keeping the above in mind, the authors presented empirical evidence of a relationship between workplace innovation and human capital, and its influence on productivity. Another comparative study of Swiss and Greek companies (Arvanitis and Loukis 2009) showed the positive effects of physical capital, ICT, human capital and new organisational practices on labour productivity. However, Swiss companies are more efficient in combining and implementing those factors, while in Greek companies, physical capital still plays a crucial role in relation to labour productivity. Research on Catalan companies (Torrent-Sellens and Ficapal-Cusí 2010) confirmed the role of new co-innovative sources in technology and knowledge-intensive companies. Among the remaining 80% of companies, no evidence was found to show that those sources had any impact. In Japan, Miyazaki et al. (2012) classified ICT applications into four stages of sophistication (non-performing ICT assets, section-wide system applications, company-wide system applications and inter-corporate system applications) and found that the impact of ICT on company productivity increases with a successive stage of ICT use sophistication. In Germany, a set of studies (Hempell 2005; Hempell and Zwick 2008) based on several time samples of its industry’s companies also confirmed the existence of relationships of the dependence among labour productivity, digital technology and organisational innovation processes. The authors found that ICT fosters product and process innovations by facilitating employee participation and communication, and ICT investment is most productive in companies with experience gained from earlier innovations. In recent work, Hall et al. (2013) examine the company-level relationships among product, process, organisational innovations, productivity, research and development (R&D) and ICT, using data from Italian manufacturing enterprises. They found that R&D and ICT are both strongly associated with innovation and productivity, with ICT investment being more important for productivity. ICT and R&D contribute to productivity both directly and indirectly through the innovation equation, but are neither complements nor substitutes. However, in individual terms, they both appear to have large impacts on productivity, suggesting some underinvestment in these activities by Italian companies.

Although the transition of Eastern European economies from centrally planned economies to market-driven systems occurred more than two decades ago and those countries now actively participate in the global economic community, publications on ICT in transition economies are sparse (Roztocki and Weistroffer 2011). According to Roztocki and Weistroffer (2008, 2011) there are several explanations for the scarcity of published research: Firstly, there is a lack of funding for this type of research. Therefore, much of the published research dealing with ICT in transitional economies has been carried out by researchers employed at institutions in the developed countries. However, the transitional economies that joined the European Union can now access more funds and more research should be expected from these countries in the future. Secondly, in the communist period, research was directed to disciplines other than ICT, such as physics and chemistry, and was mostly theoretical in nature, with hardly any market implications. Moreover, the effect of many administrative structures and procedures that were implemented in the past still remain. Furthermore, reforms have been concentrated on economic changes rather than academics, with the existing structures at many universities still inhibiting research productivity. Economies that were closed blocked international linkages, which affected innovation, knowledge spillovers and technology adoption. The character of the command economic system resulted in low competitiveness and technological obsolescence (Winiecki 2004). Another crucial problem was connected with the different institutional milieu and cultural backgrounds within CEE countries at the beginning of the transition, in contrast to those within Western Europe, which had an influence on implemented changes on the micro and macroeconomic level (Theodoulides and Kormancova 2013).

The first publications concerning Eastern European countries evaluated the impact of ICT on growth at the aggregate level. Van Ark and Piatkowski (2004) compare the productivity performance of 10 Central and Eastern European countries (CEE-10) and 15 European Union (EU-15) countries during the 1990s, examining productivity and income convergence hypothesis. Their investigation gives more support to the convergence hypothesis. Moreover, they show that ICT capital in the CEE countries has contributed as much to labour productivity growth as it has in the EU-15 countries and that ICT capital on its own has not been an important source of convergence. They emphasise the importance of consistent progress in the economic, institutional and regulatory environment, in the creation of modern institutions, the implementation of market-oriented policy reforms, and the increases in innovation and improvements in the quality of human capital. There are some pioneers of company-level analysis, such as Stare et al. (2006), who explored a link between ICT and the performance of service companies in Slovenia. They confirmed the positive impact of ICT use on productivity; however, due to the absence of data on complementary expenditures for training and organisational change, the results might overestimate the impact of ICT.

Table 1 summarises the main results for a broad set of studies. Most of the international empirical evidence has confirmed the complementarities of new co-innovative company productivity sources: ICT investment and usage, human capital and new forms of work organisation. However, more empirical studies are still needed in this field. Eastern European countries are noticeable by their scarcity of studies on ICT, complementarities and productivity.

Table 1 Literature Review Summary
Full size table

Data

Our empirical descriptive and econometric analysis is based on data from the Management, Organisation and Innovation (MOI) Survey 2009, a joint initiative of the European Bank for Reconstruction and Development (EBRD) and the World Bank Group (the World Bank). The MOI survey was undertaken for the first time in 2008-2009, covering 1870 manufacturing establishments with between 50 and 5000 employees from 10 Eastern European countries: Belarus, Bulgaria, Kazakhstan, Lithuania, Poland, Romania, Russia, Serbia, the Ukraine, Uzbekistan, with Germany as a developed country benchmark and India as a developing country benchmark. MOI interviews were conducted face-to-face with interviewers recruited by local survey companies and took place between October 2008 and April 2010. The interviews were conducted with managers and lasted on average 50 minutes. The response rate was 44%. The questionnaire comprised seven sections organised by topic. Initially, questions were posed about the characteristics of the company, such as legal status, ownership and number of years in operation. This was followed by sections on company management practices, organisation, innovation and R&D, and the degree of competition and labour. The MOI questionnaire was developed and tested in two pilot surveys prior to its implementation in the field. The two main objectives of the sample are to measure and compare management practices across the countries and to conduct a company performance analysis focusing on determining how management practices affect productivity and job creation in manufacturing. The MOI survey was used to determine if the quality of management practices is positively associated with various measures of company performance in Eastern European countries (Bloom et al. 2012; Schweiger and Friebel 2013). The survey uses a standardised survey instrument and a uniform sampling methodology to minimise measurement error and generate a sample representative of the manufacturing sectors in each country. Data are comparable across the countries and the sample size is large enough to conduct statistically robust analysis with levels of precision at a minimum of 7.5% for 90% confidence intervals (EBRD and World Bank 2008).

Data from the MOI survey was complemented by company performance data (balance sheets and income and loss statements) from the Bureau van Dijk’s Orbis database. Given that the output variables from the Bureau van Dijk’s Orbis database are not available for all countries or even for all companies in a particular country, we run the risk that results are driven by the specific country. Performance data: the operating revenue and cost of employees are winsorised at 1% to limit the impact of outliers on the result (this means that all the data below the 0.5th percentile are set to the 0.5th percentile and all the data above the 99.5th percentile are set to the 99.5th percentile as in Bloom et al. (2012)).

For the present study, we have included 444 complete cases from five Eastern European countries: 95 from Bulgaria, 41 from Poland, 84 from Romania, 110 from Serbia and 114 companies from Ukraine. Other countries were excluded from the analysis due to lack of financial data available in 2008.

The companies operate mainly in low technology industries (52% of companies), where 37% produce textiles, textile products, leather and foot wear, and 33% produce food products, beverages and tobacco. 22% of companies belong to medium-low technology industries, where 53% produce basic metals and fabricated metal products. Furthermore, 21% of the sample are medium-high technology companies, where 40% produce machinery and equipment. Lastly, only 5% of companies operate in a high-technology industry, where 64% produce medical, precision and optical instruments. The companies in the sample are medium-sized (below 250 employees) manufacturing companies (78% in total) and have on average 207 employees. Table 2 gives summary statistics of variables of interest for the sample of 444 companies included in econometric analysis as well as additional data to conceptualise.

Table 2 Summary statistic
Full size table

Most countries of Eastern Europe, during the transition from a command-driven to market-driven economy, have restructured and reallocated resources to foster great efficiency in the use of resources. Despite a significant growth in the service sectors after the transition, those countries’ manufacturing sectors still contribute significantly to their GDP. The manufacturing value added as a share of GDP ranged from 15% for Bulgaria to 21% for Romania. There is a significant gap between Poland and other EE countries in labour productivity (measured as a company’s operating revenue in 2008 divided by the number of full-time employees). Considering this indicator, on average the companies in Poland have more than two and half times higher productivity than those in Serbia and more than three times higher productivity than those in the EU countries of Bulgaria and Romania. Regarding ICT indicators, an average 80% of manufacturing enterprises in Serbia have a high-speed internet connection, whereas in Poland all companies are connected. Moreover, in Polish and Ukrainian manufacturing companies, an average of more than 32% of employees regularly use personal computers on the job, while less than 15% do so in Romania and less than 20% in Bulgaria and Serbia. Polish companies are the most innovative in comparison with the countries in the sample. 80% of Polish companies, compared to 60% of companies in other countries introduced a new product or service in the last three years. Furthermore, a strong patent protection, which protects new innovations, is associated with higher levels of total factor productivity. Patents can also be used as a measure of the output of innovation. Patents are registered in almost half of the Ukrainian companies, in 34% of Bulgarian companies and in 29% of Polish. Companies in Ukraine have the highest share of employees with university degrees, as 33% of the total number of workers possess a university degree. In the rest of the countries, the share of employees with higher education varies from 12% to 20%. The final variable included in our analysis is the quality of management. On the average, the management practices are better in the Bulgarian companies included in the sample, followed by the Ukrainian companies.Footnote 1

Empirical Results

The OLS Estimation

Our methodology for the estimation of co-innovative sources of company productivity is an extension of the well-established traditional growth and productivity-accounting approach, based on the Solow growth model (Solow 1957) and its subsequent elaboration by Jorgenson and Griliches (1967). The co-innovative productivity sources are incorporated in the efficiency component (Total Factor Productivity, TFP). This is an important contribution in the analysis of the determinants of company productivity because this methodology allows us to incorporate several variables simultaneously (e.g. ICT use, work organisation or human capital). This allows the use of explanatory elements that go beyond pure investment, which contemplate the management and effective transformation of business activity. For example, in the case of ICT, investment in such technologies is not an automatic determinant of efficiency improvements. For improvements to be made, ICT goods and services must be used effectively by a company’s value elements, which entails the need to capture indicators of use. Our study is a one of the first empirical studies using this methodology for Eastern European manufacturing enterprises.

In order to examine labour productivity sources in Eastern European manufacturing enterprises, we used the following as dependent variables: wage, ICT usage and infrastructure, product innovation, patents, human capital and organisation. The company production function of the Cobb-Douglas type takes the form:

$$ {\mathrm{Y}}_{\mathrm{i}}={\mathrm{A}}_{\mathrm{i}}{\mathrm{K}}_{\mathrm{i}}^{\upalpha}{\mathrm{L}}_{\mathrm{i}}^{\upbeta}{\mathrm{I}}_{\mathrm{i}}^{\upgamma} $$

where, for any given company i, Y is the level of operating revenue; A is the production efficiency (Total Factor Productivity); K is the input of physical capital; L is the input of labour; and I is the input of intermediate production costs. The coefficients α, β and γ represent the elasticities of physical capital, labour and intermediate production costs over the level of company operating revenue.

In line with the literature (Bresnahan et al. 2002; Arvanitis 2005; Timmer et al. 2010), innovative sources of productivity are incorporated into the production efficiency indicator. This element shows the effects of company innovation that are not associated directly with factors of production. Thus, the indicator of efficiency Ai takes the following functional form:

$$ {\mathrm{A}}_{\mathrm{i}}= \exp \left({\updelta}_0+{\updelta}_1{\mathrm{ICTU}}_{\mathrm{i}}+{\updelta}_2{\mathrm{ICTU}\mathrm{I}}_{\mathrm{i}}+{\updelta}_3{\mathrm{INNOV}}_{\mathrm{i}}+{\updelta}_4{\mathrm{PATENT}}_{\mathrm{i}}+{\updelta}_5{EDU}_{\mathrm{i}}+{\updelta}_6{ORG}_{\mathrm{i}}\right) $$

After the logarithmic transformation of the first equation and after incorporating the production efficiency indicator model, innovative sources of productivity take the form:

$$ {lnY}_{\mathrm{i}}-{lnL}_{\mathrm{i}}={\upbeta}_0+{\upbeta}_1\left({lnK}_{\mathrm{i}}-{lnL}_{\mathrm{i}}\right)+{\upbeta}_2{\mathrm{ICTU}}_{\mathrm{i}}+{\upbeta}_3{\mathrm{ICTI}}_{\mathrm{i}}+{\upbeta}_4{\mathrm{INNOV}}_{\mathrm{i}}+{\upbeta}_5{\mathrm{PATENT}}_{\mathrm{i}}+{\upbeta}_6{EDU}_{\mathrm{i}}+{\upbeta}_7{ORG}_{\mathrm{i}}+{\upvarepsilon}_{\mathrm{i}} $$

where, β0 (constant) incorporates the logarithmic difference of intermediate costs per worker, βi for i=1…7 represents the elasticities (coefficients) of the explanatory components of company productivity and εi is the estimation error.

Finally, the labour productivity function of the Eastern European manufacturing enterprises, which would be estimated by the OLS method, takes the following form:

$$ {LP}_{\mathrm{i}}={\upbeta}_0+{\upbeta}_1{\mathrm{WAGE}}_{\mathrm{i}}+{\upbeta}_2{\mathrm{ICTU}}_{\mathrm{i}}+{\upbeta}_3{\mathrm{ICTI}}_{\mathrm{i}}+{\upbeta}_4{\mathrm{INNOV}}_{\mathrm{i}}+{\upbeta}_5{\mathrm{PATENT}}_{\mathrm{i}}+{\upbeta}_6{EDU}_{\mathrm{i}}+{\upbeta}_7{ORG}_{\mathrm{i}}+{\upvarepsilon}_{\mathrm{i}} $$

Regarding the specific indicators and variables used in the estimation, the following comments need to be made: The dependent variable, company labour productivity (LP), has been approximated by the logarithm of operating revenue in 2008 in thousands of USD divided by the number of full-time employees. This definition of LP as the dependent variable is similar to Hall et al.’s (2009) definition.

Considering the independent variables, the procedure described below has been followed. Firstly, it should be noted that the logarithmic difference between the intermediate production costs and full-time employees, which is required for the conversion of the operating revenue indicator into added value, has been incorporated into the constant of the model to be estimated. Secondly, the effect of physical productive capital on company productivity has been captured by the logarithmic difference between the cost of employees in 2008 and the number of full-time employees (WAGE).

The existing empirical literature on causal relationship between labour productivity and wages of workers presents opposite directions with respect to the flow of causality. The link between the variables is established through two routes – the marginal productivity theory and efficiency wage theory. In our analysis, we follow the efficiency wage theory developed by Alfred Marshall and supported by Katz (1986), Shapiro and Stiglitz (1984). Moreover, five additional arguments of the direction from the wage growth to labour productivity are analysed in Nayak and Patra (2013). In line with the recent empirical evidence (Lallemand et al. 2009; Faggio et al. 2010; Mahy et al. 2011) the WAGE variable allows us to determine the relationship from the wage structure to company productivity. Regarding the use of wage as an indicator of productive physical capital in companies, it is important to make two points. First, we should note that in the context of the companies analysed, the wage cost is configured as a very important component of productive physical capital. Within these contexts of analysis, wage is an indicator that partially illustrates human capital investment, in addition to the components attached to individuals in the process of company activity. Second, note that when dealing with companies, financial information on total productive investment is not always available.

Given this limitation, it was considered appropriate to introduce wage as an indicator of productive physical capital. Firstly, and as noted by economic theory, this introduction was done because of the evident relationship between wage and productivity and, secondly, because wage is an indicator of the capitalisation available with data sets. Thirdly, the set of independent variables represents ICT and its complementarities. Regarding the ICT, we included two variables: ICTU and ICTI. The first variable, ICTU, represents ICT usage in the company and is the percentage of company employees that regularly use personal computers in their jobs. The second variable ICTI is an indicator of ICT infrastructure and takes two values: 1 when the company has a high-speed internet connection on its premises and 0 if otherwise. We have two hypotheses related to the direct effect of ICT on productivity.

To show a company’s innovatory dynamics, we used two dummy variables: product innovation (INNOV) which takes value 1 when the company has introduced new products or services in the last three years and 0 if otherwise; and the variable PATENT which is 1 when the company has any patents registered abroad or at home and 0 if otherwise. Moreover, we have one indicator of human capital, the variable EDU which is a percentage of the company’s full-time employees (including production and non-production workers) with a university degree.

Lastly, the variable ORG is an indicator of management quality. To compose this indicator, we followed the procedure used by Bloom et al. (2012). Management practices were grouped into four areas: operations, monitoring, targets and incentives. The operations question focused on how the establishment handled a process problem, such as machinery breakdown. The monitoring questions covered collection, monitoring, revision and use of the production performance indicators. The target question focused on the time-scale of production targets and the incentives questions covered promotion criteria, practices for addressing poor employee performance and rewarding production target achievement. As the scaling varied across management practices, the scores were converted to z-scores by normalising each practice (i.e. each question) to the mean of zero and the standard deviation of one. To avoid putting the most emphasis on the monitoring aspect of management practices (which had the most underlying questions), an unweighted average was first calculated across z-scores for a particular area of the four management practices and was then normalised. Furthermore, an unweighted average was taken across the scores for the four practices and was then normalised. This means that the average management practices across all the companies in all the countries in the sample are equal to zero, and the actual management practices of the companies deviate from zero either to the left (‘bad’ practices) or to the right (‘good’ practices).

Results of the OLS Estimation

The results of the OLS estimation of the productivity of manufacturing establishments in Eastern European countries are presented in Table 3. We estimated the model for all the countries and sample divisions for every country: Bulgaria, Poland, Romania, Serbia and Ukraine. The models include a set of two-digit industry-fixed effects; also, the model for all countries includes country-fixed effects as the additional controls that will affect productivity. All the models estimated are significant (p-value<0.001) and the level of adjustment (adjusted R2) is satisfactory and varies from 44% to 69%.

Table 3 Influence of ICT and complementarities on labour productivity: Cross-country comparison
Full size table

Regarding the coefficients of the determinants of labour productivity, the variable representing physical capital (WAGE) (p<0.001) is significant in each subsample and, as expected, has a positive sign. For the whole sample (EE), ICT infrastructure (β=0.397, p<0.001) and management quality (β=0.166, p<0.001) are considered to be important determinants of productivity. In the sample of Bulgarian companies, two variables are significant: ICTU (β=0.01, p<0.05) and ICTI (β=0.904, p<0.05). Furthermore, in Poland only the education variable (β=0.019, p<0.05) appeared to be relevant. In Romania, two indicators of company innovation: INNOV (β=0.521, p<0.001) and PATENT (β=0.481, p<0.1), and EDU (β=0.01, p<0.1) are significant in terms of positively affecting productivity. Lastly, in manufacturing companies from Ukraine, the existence of good organisation practices (β=0.185, p<0.001) is the relevant factor for labour productivity. The estimated models show that different variables influence productivity depending of the country.

However, the results are disappointing, considering that only a few of the six reviewed explanatory variables of ICT and its complementarities turned out to be significant. Moreover, we have tried to include complementary relations (co-innovation) in the model by including the interactions among ICT, education, innovation and organisation, but they were not significant for company productivity. As shown in the previous studies (e.g. Bresnahan et al. 2002; Arvanitis 2005; Arvanitis and Loukis 2009), we expected to have a significant positive impact on the productivity of human capital (EDU) and workplace organisation (ORG). Moreover, we expected to have complementary relations by including interactions, as in, for example, Brynjolfsson and Hitt (2003), Gretton et al. (2004), or Hempell and Zwick (2008). The reason for the different results could be the relatively small sample size (n = 444). The sample size makes statistical significance hard to interpret. In small studies, the confidence intervals are narrower and the estimates can be less precise. Moreover, unobserved heterogeneity or simultaneity issues can appear. However, these data have descriptive and analytical value, as it was proved by the previous studies based on MOI survey (Bloom et al. 2012; Schweiger and Friebel 2013). It was worth attempt to analyse a sample from Eastern Europe.

The SEM Estimation

The presence of complementary relations (co-innovation) was not found using OLS analysis. Therefore, we introduced the causal effects of the interaction between factors of productivity into the analysis. We aimed to use structural equation modelling (SEM) with observed variables and measurement errors to examine the direct and indirect effect of ICT and its complementarities on labour productivity sources in Eastern European manufacturing enterprises.

Structural equation systems are formal mathematical models; they are a set of linear equations that encompass, as particular cases, various types of model, such as regression models, simultaneous equation systems, factor analysis and path analysis. The equation system’s variables can either be directly observed or they can be measurable, theoretical or latent variables representing concepts that are not directly observed. While latent variables must be continuous, observed dependent variables can be continuous, censored, binary, ordered, categorical (ordinal) or combinations of these variable types.

General SEM model comprises two sub-models: a structural model that relates latent variables to each other, and a measurement model that relates each latent variable to the respective variables measuring it; these are generally called indicators. A causal structure between latent variables is usually assumed to exist.

SEM models have certain desirable distinctive features that 1) allow for the explicit inclusion of measurement error in the estimation process into as many variables as is seen fit; 2) allow for the simultaneous estimation of the parameters of a series of relationships of dependence, where a variable can act as dependent in some equations and independent in others; 3) are able to show reciprocal causes, recursive and non-recursive models; and 4) enable them, with new developments, to be used for exploratory analysis, albeit with a confirmatory technique.

Following the most commonly used notation (Jöreskog and Sörbom 2004), it is possible to formalise SEM models by a system of linear structural equations whose metrical representation is:

$$ \eta =\alpha +B\eta +\varGamma \xi +\zeta $$

where: η (m x 1) and ξ (n x 1) are random vectors of latent dependent and independent variables; α is a vector (m x 1) representing the intersections of axes; В (m x m) is the matrix of coefficients of latent endogenous variables representing the effects of variables η on other variables η; Г (m x n) is the matrix of coefficients of latent exogenous variables representing the direct effects of variables ξ on variables ξ; and ζ is a vector (m x 1) indicating the random perturbations in the equation. It is assumed that E(η) = 0, E(ξ) = 0 and E(ζ) = 0.

The observed (measurable) variables are represented by vectors y (p x 1), where p is the number of indicators of η; and x (q x 1), where q is the number of indicators of ξ. Both formulations are related to the latent variables by the following equations:

$$ y={\uptau}_{\mathrm{y}}+{\Lambda}_{\mathrm{y}}\upeta +\upvarepsilon $$
$$ x={\uptau}_{\mathrm{x}}+{\Lambda}_{\mathrm{x}}\upxi +\updelta $$

where ε (p x 1) and δ (q x 1) are the vectors of the error terms. It is assumed that ε is not correlated to η, ξ or δ; and that δ is not correlated to η, ξ or ε. Λy (p x m) and Λx (q x n) are matrices containing the structural coefficients λij that relate the latent and observed (measurable) variables, and τy (p x 1) and τx (q x 1) are the vectors of the constant terms of intersection.

The fundamental hypothesis of structural equation systems is Σ = Σ(θ), where Σ is the population covariance matrix and Σ(θ) is the model covariance matrix, written as a function of a parameter vector of θ. The estimation of the parameters of θ is obtained by minimising a function of adjustment:

$$ F\left(\theta \right)=F\left[S,\Sigma \left(\theta \right)\right] $$

Once the model’s parameters have been estimated, the resulting covariance matrix is compared with the data covariance matrix and, if the difference between the two matrices is statistically acceptable or zero, then the proposed SEM model is recognised as a plausible explanation of the reality.

This methodology for explaining productivity sources in manufacturing enterprises from Eastern European countries allows us to establish a better explanatory model because of the use of a multi-equation system. Furthermore, this method enables us to introduce specific measurement errors to each of the variables. This improves the specification of parameters, meaning they can be unbiased, consistent and have lower variance.

As in the previous part, we have used the 444 complete cases for manufacturing enterprises from Eastern European countries from the MOI survey. The variables are approximated similarly to the OLS estimation. The general analysis model includes 10 hypotheses. The direct explanatory factors of the labour productivity of Eastern European manufacturing enterprises are wage, ICT usage, ICT infrastructure and quality of management. The company productivity is higher when the wages of employees are higher (H 1 ). H 1 implies that a company's capacity to increase turnover per worker relies on finding better-paid and probably higher-quality labour. Furthermore, regarding ICT, we included two variables: ICTU and ICTI. We have two hypotheses related to the direct effect of ICT on productivity. The productivity is higher when more workers regularly use computers in their jobs (H 2 ) and when a company has a high-speed internet connection (H 3 ). Digital intensity makes a company's value generation process more competitive; therefore, it is a lever of productivity for a company. Lastly, regarding its direct influence on labour productivity we stated in hypothesis 4 (H 4 ) that better quality of management in a company can be explanation for higher productivity. Afterwards, we have established six hypotheses related to the indirect factors of productivity and their interrelationships. In our attempts to show that companies with innovatory dynamics increase the need for ICT use, we introduced two hypotheses. H 5 and H 6 that postulate that in companies with innovative practices (new product or service introduction, or patent obtention) the ICT use needs to increase. The next examined hypothesis is that the higher the level of human capital, the greater the presence of the use of technology during their tasks (H 7 ). The last hypothesis (H 8 ) is related to the causal relationships between ICT usage and ICT infrastructure. The better the ICT infrastructure, the higher the ICT usage in the company will be. Furthermore, we examine the relationship between education and patents with H 9 : a higher level of human capital in a company boosts innovation. Lastly, H 10 is that better organisation and management in a company improve ICT infrastructure development.

The SEM model represented by the following system of equations with observed variables and measurement errors is to be formally estimated as a simplification of the formulation considered earlier:

$$ \left(\begin{array}{l}LP\\ {} ICTU\\ {} PATENT\\ {} ICTI\end{array}\right)=\left(\begin{array}{l}{\beta}_{10}\\ {}{\beta}_{20}\\ {}{\beta}_{30}\\ {}{\beta}_{40}\end{array}\right)+\left(\begin{array}{ccccccc}\hfill {\beta}_{11}\hfill & \hfill {\beta}_{12}\hfill & \hfill {\beta}_{13}\hfill & \hfill {\beta}_{14}\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill {\beta}_{23}\hfill & \hfill 0\hfill & \hfill {\beta}_{25}\hfill & \hfill {\beta}_{26}\hfill & \hfill {\beta}_{27}\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill {\beta}_{37}\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill {\beta}_{44}\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \end{array}\right)*\left(\begin{array}{l} WAGE\\ {} ICTU\\ {} ICTI\\ {}ORG\\ {} PATENT\\ {} INNOV\\ {}EDU\end{array}\right)+\left(\begin{array}{l}{\varepsilon}_1\\ {}{\varepsilon}_2\\ {}{\varepsilon}_3\\ {}{\varepsilon}_4\end{array}\right) $$

The model shows direct and indirect factors that explain labour productivity in Eastern European manufacturing enterprises. The first equation is related to the direct explanatory factors of labour productivity (LP) and includes hypotheses 1-4. The second equation is related to the indirect explanatory factors of the ICT usage of employees and includes hypotheses 5-8. The third equation is related to the indirect explanatory factor of the innovation indicator PATENT and includes hypothesis 9. Finally, the fourth equation is related to the indirect explanatory factor of ICT infrastructure and includes hypothesis 10.

Results of the SEM Estimation

The results of the structural equation model estimation are presented in Figure 1 and Table 4. Figure 1 is a path diagram with hypotheses and standard coefficients estimated. Table 4 shows the SEM results and goodness of fit statistics.

Fig. 1
figure 1

SEM model results. Source: The authors’ own elaboration. Orbis database and MOI survey.

Full size image
Table 4 Results of the structural equations model (n=444)
Full size table

Firstly, it should be noted that, as proposed in the literature (Hu and Bentler 1999; Hooper et al. 2008; Kline 2011), the fit indices illustrate the model's goodness of fit (CFI=0.974, NFI=0.947, IFI=0.974, TLI=0.951, SRMR=0.042). In addition, the value of CMIN/DF=1.904, below 2, and the value of RMSEA=0.045, below 0.05, prove the adequate fit of the developed model. These values suggest that the model has acceptable reliability, validity, and unidimensionality, thereby confirming the constructs reliability and scale validity.

Secondly, it is confirmed that all coefficients obtained are significant at the 5% level and the direction of relations are as expected. It should be noted that the main direct determinant of labour productivity in Eastern European manufacturing enterprises is the wage of full-time employees (H 1 : β=0.649, p<0.001). This is followed by the quality of management (H 4 : β=0.152, p<0.001), ICT infrastructure (H 3 : β=0.13, p<0.001) and ICT usage (H 2 : β=0.087, p<0.010).

Furthermore, the set of indirect determinants of productivity is confirmed, firstly, by the relationship between the ICT usage and the variables representing innovation: INNOV (H 5 : β=0.105, p<0.011) and PATENT (H 6 : β=0.156, p<0.001), human capital (H 7 : β=0.425, p<0.001) and ICT infrastructure (H 8 : β=0.097, p<0.019). Another indirect impact on productivity is established by the human capital on the presence of patents (H 9 : β=0.154, p<0.001). Moreover, organisation (ORG) has an indirect effect by influencing ICT infrastructure (H 10 : β=0.214, p<0.001).

Our results partially confirm the evidence from the previous study of Díaz-Chao et al. (2015), using the SEM model for SMEs in Spain. They confirmed that wage per worker is the main determinant of higher productivity. Moreover, their study established an indirect relationship between co-innovation and productivity by the capacity of companies to export goods to the EU-markets. In our study, due to the lack of available data on exports, we could not capture this dimension. However, our study also identifies similar indirect sources of productivity by ICT usage. Similar to Díaz-Chao et al. (2015) , where ICT usage is affected by ICT capital and innovation, in our results determinates are patents, education, the presence of innovation and ICT infrastructure.

Conclusions

In recent years, a variety of international research has demonstrated the existence of co-innovative sources of company productivity: more precisely, the complementarity among ICT usage and investment, innovation, human capital and work organisation. Using the MOI Survey 2009 data for a representative sample of manufacturing enterprises in Eastern European countries (i.e. Bulgaria, Poland, Romania, Serbia and Ukraine), we analysed the determinants of company labour productivity. We aimed to extend the existing literature with new empirical evidence regarding manufacturing enterprises from Eastern European countries. The OLS method was used to identify factors explaining levels of labour productivity. Moreover, the developed and tested structural equation model added evidence for the direct and indirect determinants of labour productivity.

The results of the OLS are the estimation of the growth model extended by ICT and its complementarities. In line with Solow’s growth model, labour productivity is explained by physical capital represented by the wage of full-time employees. Models for country subsamples show that different factors appear to be significant depending on the country: mainly ICT infrastructure, human capital and management quality. However, the results of the OLS estimation have not allowed any causal relationships to be established to explain the labour productivity in Eastern European manufacturing companies.

The SEM estimation has allowed relationships to be established which explain the labour productivity in Eastern European manufacturing companies. The results of the SEM estimation show that the relationships among ICT and its complementarities and productivity, have been entrenched both directly and indirectly. As was in the case of the OLS model, the main determinant of company productivity is the wage of full-time employees. In this respect, it was found that productivity is basically associated with labour quality. Lower direct effects have indicators of ICT usage, infrastructure and management quality. Furthermore, we identified a set of indirect effects. Firstly, the variable for the ICT usage of workers in an enterprise is affected by patents, education, the presence of innovation and ICT infrastructure. Secondly, an indirect effect on labour productivity is established from education to obtained patents. Lastly, the quality of management is significant for ICT infrastructure in an enterprise.

Obtained results present some policy implications. First, it is important to emphasise the need to coordinate efforts in the joint promotion of ICT use, with innovation activities, organisational changes and human capital improvement. For example, partial public policies to promote ICT use or ICT investment, without considering other ICT complementarities which affect labour productivity, may not produce the desired effects. Moreover, due to indirect effects on labour productivity. companies that invest in human capital, for example by internal training, increase ICT usage and then productivity. In addition, the promotion of innovative activities not only has a direct effect but can also raise ICT usage. All in all, enterprises in which ICT investment go hand in hand with other determinants of productivity will be better off than those which just invest in one of the factors. Knowledge of such relations is useful for policy makers because it can lead to the more efficient choice and combination of policy initiatives and measures.

The main limitation of this research is the relatively small sample of enterprises included in the MOI survey database and the high number of missing financial data. We have in mind existing disparities in labour productivity caused by company heterogeneity across countries and across industries. This implies that more factors influence labour productivity, which we may not have taken into consideration. However, such a discussion is out of the scope in the current research. Regarding the importance of this topic, especially for transitional economies, there is a need for more data: more countries need to be involved; indicators need to be improved, data needs to be collected from service enterprises and small and medium enterprises; and longitudinal data needs to be obtained. This study has a preliminary character and suggests conducting further research into transition economies.