The Role of Innovation and Management Practices in Determining Firm Productivity

Abstract

In this paper, we compare the impacts of management practices and innovation on productivity, using data from a unique firm-level survey covering 30 countries in Eastern Europe and Central Asia in the period 2011–2014. We estimate a structural model linking productivity to innovation activities and management practices. Results suggest that both returns to innovation and returns to management practices are important drivers of productivity. However, productivity in lower-income economies is affected to a larger extent by management practices than by innovation, while the opposite holds in higher-income economies. These results imply that firms operating in less favourable business environments can reap large productivity gains by improving the quality of management practices, before engaging in innovation through imitating and adapting foreign technologies.

Introduction

There are many ways in which firms can increase their productivity (see Syverson 2011, for a review) and thereby contribute to the improvement of aggregate productivity. However, the most common and most important driver of change within firms—particularly in advanced industrialised countries—is the introduction of new products, new processes or new ways of conducting business—in other words, innovation (see, for example, Geroski 1989, and Geroski et al. 2009). The link between innovation and productivity is generally found to be positive and significant (see, for example, Mohnen and Hall (2013), for an overview and studies by Hall and Sena (2014), for the UK, and Raymond et al. (2015), for Dutch and French manufacturing firms, Crespi and Zuniga (2012), for six Latin American countries, Masso and Vahter 2008, for Estonian firms). Governments and policy makers regardless of the country’s level of development are keen to foster innovation (see, for example, European Commission 2014), typically of the high-tech variety (EBRD 2014).

While the concept of innovation as a driver of productivity is widely assessed in the literature, the role of management practices—that is, the core practices the firm follows in operations, monitoring, targets and incentives—has remained empirically unexplored for a long time, thanks to the absence of high-quality data. Since the seminal Bloom and Van Reenen study (2007), we know that the quality of management practices is correlated with firm performance measures, such as productivity and profitability, in developed (Bloom and Van Reenen 2007, 2010; Bloom et al. 2017) and developing countries (Bloom and Van Reenen 2010; Bloom et al. 2013). Better managed firms are also more likely to innovate (Bloom et al. 2017). An analysis of firm-level productivity should thus account for the quality of firms’ management practices in addition to innovation. While the two are correlated with a certain degree, they are not identical; every firm uses management practices, but not every firm innovates.

This paper combines both strands of research by assessing the impact of innovation as well as management practices on firm productivity. More specifically, we answer the question whether both channels affect firm productivity significantly, with management practices having a direct impact on innovation and productivity, as well as an indirect impact on productivity via innovation. Moreover, we explore whether the importance of innovation and management practices varies according to the status of economic development. The catch-up growth literature suggests that firms in developing countries can imitate or adapt technologies introduced elsewhere in order to catch up with firms in advanced countries, while the latter need to innovate at the frontier to progress further (Acemoğlu et al. 2006; Aghion 2016). However, there might be an even easier strategy for the firms in least developed countries: before they start imitating foreign production processes, they can reap large productivity gains by improving their management practices.

We contribute to the literature in three important dimensions. First, our paper is one of the first to include both innovation and the quality of management practices in the model, rather than just one of them. We use a variation of the three-stage model devised by Crépon et al. (1998, known as the “CDM model”). In addition to R&D we focus on management quality since in lower-income countries, as some of the ones we have in our sample, technological change is more likely to be driven by imitation and assimilation without formal R&D, whereas every single firm has either good or bad management practices.

Second, we use data from a unique firm-level survey, the fifth round of the European Bank for Reconstruction and Development (EBRD)—World Bank (WB) Business Environment and Enterprise Performance Survey (BEEPS V). For the first time, BEEPS V included an Innovation Module, with the aim to obtain a better understanding of innovation in its various forms (product, process, organisational and marketing), R&D and management practices. Our sample covers 30 countries in Eastern Europe and Central Asia¹ in the period 2011–2014, with a wide variety in terms of economic and institutional development. Within each of these countries, the sample of firms is representative, with a large variation in productivity levels. Due to data availability we focus on manufacturing enterprises with at least 20 employees.

Large differences in productivity across both firms and countries continue to exist (see, for example, Griffith et al. 2006; Arnold et al. 2008, for OECD countries; and Hsieh and Klenow 2009, for China and India) and can be found even in industries producing very homogenous goods (Foster et al. 2008). In that respect, the countries in our sample are no exception. There are highly productive firms in lower-income economies and poorly performing firms in higher-income economies. Moreover, Akcigit et al. (2016) show that managerial delegation is important for firm selection: in developing countries, where managerial human capital is scarce and managerial delegation less efficient, firms with growth potential are not expanding enough to replace firms with little growth potential. Having such a diverse sample allows us to assess under which conditions innovation versus the quality of management practices are most successful in boosting firm productivity.

Third, we improve the measures of product and process innovation typically available in surveys by analysing the verbatim descriptions of the firms’ new products and processes and comparing them with the definitions in the Oslo Manual (Eurostat and OECD 2005), which contains the guidelines for the collection and the use of data on innovation activities. To our knowledge, this is the first paper to do so for a large number of countries.²

We find that both management practices and innovation (regardless of the type) are positively and significantly associated with labour productivity. However, the importance of each varies with the level of development. In lower-income economies, the economic impact of high-quality management practices is stronger than the effect of introducing product and process innovation, whereas the opposite holds for higher-income economies. Furthermore, in lower-income economies, high-quality management practices primarily affect labour productivity directly, rather than indirectly via product or process innovation. In higher-income economies, the primary channel is indirect via process or technological innovation. We interpret these findings as evidence that economic progress can be achieved by improving management practices despite an unfavourable environment for innovative activities.

The remainder of the paper is organised as follows. “Data and Descriptive Statistics” section describes the data. “Estimation Model” section presents the underlying model, while “Estimation Results” section contains the estimates and “Sensitivity Analysis” section a number of robustness checks. “Conclusion” section concludes.

Data and Descriptive Statistics

Our main data source is BEEPS, a firm-level survey conducted by the EBRD and the WB. BEEPS is based on face-to-face interviews with managers of registered firms with at least five employees to examine the quality of the business environment. The fifth round of the survey (BEEPS V) took place in 2011–2012 in Russia and 2012–2014 in all other countries and included an Innovation Module with a section on management practices.³

In this paper, we focus on the subset of manufacturing firms with at least 20 employees (50 in Russia⁴), for which both measures of innovation and management practices are available. Table 1 provides an overview of our sample including the geographical region and income level of the countries that firms belong to as well as a few descriptive statistics.⁵ The number of observations per country ranges from 16 in Montenegro to 380 in Ukraine, 479 in Russia and 693 in Turkey.⁶ There is also significant variation in the average quality of management practices and occurrence of innovation across countries.

Table 1 Sample breakdown.

Measurement of Innovation

The Innovation Module of BEEPS V builds on the established guidelines published in the third edition of the Oslo Manual (Eurostat and OECD 2005), covering product and process innovation, organisational and marketing innovation, R&D spending and the protection of innovation. Examples for each type of innovation were given to generate a common understanding of the definition of innovation, and eligible firms were asked to provide a detailed description of their main product or process innovation (in terms of impact on sales or costs, respectively). This detailed information was used to obtain the so-called cleaned measures of product and process innovation.⁷

All in all, the BEEPS V methodology and the efforts made to cross-check and reinterpret individual responses go a long way towards achieving comparability of results across countries and firms. However, as any existing innovation survey, they cannot ensure a common understanding of innovation across all survey respondents.

Management Practices

Besides innovation, this paper highlights the role of management practices in determining productivity at the firm level. In order to examine their impact, we use survey responses to measure management practices. BEEPS V included a selection of questions from the U.S. Census Bureau’s Management and Organisational Practices Survey (MOPS) (Bloom et al. 2017). The questions concerned four aspects of management—operations, monitoring, targets and incentives—and requested unordered categorical responses. The section on operations focused on how the firm handled a process-related problem, such as a machinery breakdown, while the question on monitoring covered the collection of information on production indicators. The timescale for production targets, as well as their difficulty, and the awareness of them are part of the section on targets. Lastly, the questions on incentives covered criteria governing promotion, practices for addressing poor performance by employees and the basis on which the achievement of production targets was rewarded. These questions were directed to all manufacturing firms with at least 20 employees (50 in the case of Russia). The median number of completed interviews with sufficiently high response rates to the management practices section was just below 55 per country, with totals ranging from 15 in Montenegro to 626 in Turkey. On the basis of firms’ answers, the quality of their management practices can be assessed and assigned a rating, which can then be used to explain productivity levels.⁸

As the scaling varies across management practices, we first standardise the scores of each management practice (i.e. each question) to having a mean of zero and a standard deviation of one (as in EBRD 2009; Bloom et al. 2012; EBRD 2014). We then use the z-scores to calculate unweighted averages making use of the z-scores for each individual section of the respective management practice, in order to prevent accentuating the target and incentive sections, which included multiple questions. Finally, we compute an unweighted average across the scores for the four management areas and standardise once more this unweighted average.⁹

As a result, the average management score across all firms for which the underlying variables are available for all countries is equal to zero. Management practices of individual firms in turn deviate either left or right from zero. While the former indicates below average management practices, obtaining a positive overall z-score refers to a higher quality of management practices. As Bloom et al. (2012) put it, “indicators of management practices can be thought of as indicators for the quality of management (a latent variable, which cannot be observed directly)” (p. 601).

Differences Between Management Practices and Innovation

There is potentially some overlap between management practices and organisational innovation, which deals primarily with people and the organisation of work. For example, the first introduction of quality management systems or lean production is an organisational innovation in business practices as well as an improvement in management practices. However, not every improvement in management practices is an organisational innovation—once a firm has introduced a quality management system, its further improvements are not organisational innovations anymore.

Furthermore, the survey measures the quality of management practices over the last complete fiscal year and not improvements in management practices. This is further reflected in the fact that the correlation coefficient between the quality of management practices and organisational innovation is only 0.198 (even if it is statistically significant at p = 0.000). This means that above median quality of management practices is more often than not associated with occurrences of organisational innovation. But the correlation is far from 1, which would imply that organisational innovation occurs every time that the quality of management practices is above the median. Correlation coefficients with other measures of innovation (product, process and marketing) are even lower.¹⁰ Hence, we are confident that our results on the impact of management practices on productivity are not confounded with innovation activities.

Estimation Model

Identifying the effect of management practices quality on innovation and productivity is challenging because they are likely to affect productivity directly and indirectly. Management practices can be an important input for innovation. For example, target setting can be important to manage the introduction of new production processes, and these in turn are likely to impact productivity.

To estimate these direct and indirect effects, this paper uses the workhorse empirical model relating innovation and performance by Crépon et al. (1998). The CDM model generally consists of three sets of relationships. The first set explains the firm’s decision to invest in R&D as a function of firm and industry characteristics. The second set characterises various innovation outcomes as a function of R&D and other firm and industry characteristics. The third set links innovation outcomes to firm productivity. We amend this original CDM model by adding an equation explaining the quality of management practices to the first stage and allowing the firm’s quality of management practices to affect firm’s innovation output as well as productivity.

More concretely, our model is composed of three sets of equations shown below and graphically represented in Fig. 1:

$$\begin{aligned} {\text{Md}}_{i} & = 1\left[ {M_{i}^{*} \ge 0} \right],{\text{and}}\,R_{i} = 1\left[ {R_{i}^{*} \ge 0} \right] \,{\text{where}} \\ M_{i}^{*} & = X_{i1} \beta_{1} + X_{i2} \beta_{2} + X_{i3}^{1} \beta_{3} + \varepsilon_{i1} \\ R_{i}^{*} & = X_{i1} \gamma_{1} + X_{i2} \gamma_{2} + X_{i3}^{2} \gamma_{3} + \eta_{i1} \\ \end{aligned}$$

(1)

$$\begin{aligned} {\text{Innov}}_{i} & = 1\left[ {{\text{Innov}}_{i}^{*} > 0} \right],{\text{where}} \\ {\text{Innov}}_{i}^{*} & = \delta_{M} M_{i}^{*} + \delta_{R} R_{i}^{*} + X_{i4} \delta_{4} + X_{i2} \delta_{2} + X_{i1} \delta_{1} + \varepsilon_{i2} \\ \end{aligned}$$

(2)

$${\text{Prod}}_{i} = \theta_{I} {\text{Innov}}_{i}^{*} + \theta_{M} M_{i}^{*} + X_{i5} \theta_{5} + X_{i1} \theta_{1} + \varepsilon_{i3}$$

(3)

Fig. 1

Source: Authors’ representation of the model

Variation of the CDM model used in this paper. Note: Based on Crépon et al. (1998). $X_{2}$, $X_{3}^{1}$ and $X_{3}^{2}$ also contain indicators for “don’t know” values of the number of years of manager’s sector experience, having an internationally recognised certification and percentage of employees with a completed university degree, respectively.

The first set of equations, Eq. (1), is a bivariate probit model describing the binary variables of performing R&D and having management practices above the median of the overall distribution in the sample. $\varepsilon_{i1}$ and $\eta_{i1}$ follow a bivariate standard normal distribution.

The second equation of the model, Eq. (2), determines the probability of a firm implementing innovation, taking into account its management practices. The latent variables $M_{i}^{*}$ and $R_{i}^{*}$ derived from (1) are used to explain the effect that management practices and R&D exert on innovative activities. $\delta_{M}$ and $\delta_{R}$ denote the impact that the quality of management practices and R&D performance have on the probability to innovate. ${\text{Innov}}_{i}$ refers to the occurrence of one or a combination of the various types of innovation mentioned earlier.

The final equation of the model, Eq. (3), relates the firm’s innovative activities—or more precisely the latent variable that determines whether or not the firm innovates—to labour productivity (measured as sales per employee, converted into US dollars, in natural logarithmic terms).¹¹ $\theta_{I}$ captures the marginal effect of innovation and $\theta_{M}$ the direct marginal effect of management quality on labour productivity.

In each of the above equations, the choice of explanatory variables was dictated by data availability while at the same time adhering as closely as possible to the literature. The variables that are likely to affect R&D, management practices quality, innovation and labour productivity are included in vector $X_{1}$, which enters all equations. We control for the age of the firm, its size, ownership (whether a foreign company or the state have at least a blocking minority in the firm—a stake of 25 per cent or more) and direct exporter status. Start-ups or young firms are often assumed to be more innovative and/or productive, although survey evidence shows that this is not necessarily the case in the transition region, where large and old firms are more likely to engage in innovation activities (EBRD 2014). Foreign owners may be an important source of information about new products, processes, organisational and marketing methods (EBRD 2014), as well as have superior management practices and human capital (Girma and Görg 2007; Kumar and Aggarwal 2005). Moreover, evidence for transition countries indicates that foreign-owned firms tend to be more productive. In contrast, managers of state-owned firms may have weaker incentives to achieve efficiency savings and improve productivity, and state-owned and formerly state-owned firms tend to have lower productivity than always private firms (Estrin et al. 2009). Exporting firms may be more willing to use best practice management techniques in order to be able to compete on the international market, and they also learn about new products and processes through exporting. There is also widespread and longstanding evidence that exporters tend to be more productive and larger than their non-exporting counterparts (see Wagner 2007, 2012, for thorough reviews).

Acemoğlu et al. (2017) suggest that openness to disruption, proxied by the manager’s age, is associated with more creative innovations. Our dataset does not include the manager’s age; instead, we control for the length of the manager’s experience in the sector ($X_{2}$) and argue that it directly affects the quality of management practices, the likelihood of engaging in R&D activities and the probability to innovate. Balsmeier and Czarnitzki (2014) show that managerial experience (the number of years the manager has worked in the same industry) increases innovativeness, especially in institutionally less developed economies.

The productivity Eq. (3) includes variables contained in vector $X_{5} :$ information on whether the firm is located in the country’s capital or main business centre and capacity utilisation. As a robustness check we also add the log of fixed assets per employee, the insertion of which, however, significantly reduces the sample size (see Table 2 and Online Appendix OD for more details). We also explicitly control for the quality of management practices, thus productivity is affected by management practices not only indirectly via innovation, but also directly by including the latent variable of management practices as an explanatory variable in the productivity equation. In this paper, we are primarily interested in coefficient $\theta_{I}$, which reflects the impact that innovation has on labour productivity, and the expression $\theta_{M} + \theta_{I} \times \delta_{M}$, which reflects the accumulated direct and indirect impact that the quality of management practices has on labour productivity.

Table 2 Descriptive statistics by subsample underlying each equation.

The remaining (sets of) variables—$X_{3}^{1}$, $X_{3}^{2}$, and $X_{4}$—are related to our identification assumptions. In order to obtain an internationally recognised certification (such as ISO, HACCP or similar), firms are required to implement a quality management system. We expect that having such a certification ($X_{3}^{1}$) and thus following international best practice standards improves the quality of management practices at the firm level (see, for example, Subba Rao et al. 1997). It is not, however, directly related to R&D, innovation or productivity.

A high share of employees with a completed university degree ($X_{3}^{2}$) is likely to be correlated with the existence of an R&D division in the firm. We argue that it affects innovation output only through R&D, as the quality of management requires essentially a good manager and not necessarily a high percentage of highly educated workers. Moreover, most firm innovations, especially in lower-income countries, stem from the adoption and adaptation of existing technologies that have been developed elsewhere (EBRD 2014). If it is produced inside the firm, it is likely to be due to the firm’s R&D activities.¹²

We include access to finance (whether a firm has a loan or a line of credit), the firm’s level of geographical expansion, i.e. whether the firm’s main product is mostly sold in the local market, and the firm’s level of ICT usage (in other words, whether it uses email to communicate with its clients) among the variables that affect innovation outcomes only ($X_{4}$). Access to finance is known to be an obstacle to innovation, more so to innovation output than to R&D. While banks do not necessarily finance R&D or introduction of new products, processes or other types of innovation directly—especially when innovation is of the more risky, frontier moving type—having access to a loan or a line of credit means that the firm can use its internal sources for financing R&D or innovation, rather than using them for working capital or fixed assets purchases (EBRD 2014; Bircan and Haas 2015). R&D is generally financed internally because of asymmetric information between the firm and the finance provider and the public good nature of knowledge. If access to finance affects productivity, it is likely to be through the possibility to innovate rather than directly.

Firms that sell their products mostly in the local (i.e. municipal or regional) market are less likely to innovate as their ability to spread the cost associated with innovation is low. However, their level of geographical expansion is not likely to directly affect the firm’s productivity, more so since all equations in our model already control for direct exporter status. Firms that use ICT have better access to information about innovations appearing elsewhere and the needs of their clients. The most controversial assumption is not including ICT in the productivity equation, since many studies have documented a positive effect of ICT on productivity through capital deepening (Jorgenson et al. 2008). Some studies, however, have also shown that ICT affects productivity through innovation rather than directly (Kleis et al. 2012; van Leeuwen and Farooqui 2008).¹³

In summary, we propose a recursive system of simultaneous equations, where we use identification assumptions based on theoretical considerations or empirical evidence to identify the drivers of our endogenous variables.¹⁴ The selectivity of R&D, management practices and innovation is explicitly modelled and explains the complexity behind the observed correlations between these variables and productivity. For instance, the correlation observed between innovation and productivity may be weaker than the true underlying impact that innovation has on productivity. Indeed, if poorly performing firms find themselves under greater pressure to innovate, innovation may appear to be linked to poor short-term performance, although it improves firms’ productivity in the longer run.

We could have used continuous variations in the quality of management and R&D and get more informative estimates of their impact on labour productivity. We decided to work with binary variables instead, first to handle all three variables in the same way, as BEEPS only conveys binary data on innovation. Secondly, as illustrated by Mairesse et al. (2005), measurement errors, which tend to drive the estimates to zero or to make them insignificant, might be more important with quantitative than with binary variables.

We estimate the model by asymptotic least squares, as in the original CDM paper (Crépon et al. 1998). That is, we first estimate the reduced form of the model by a bivariate probit for the management quality and R&D equations, a simple probit for the innovation equation, and an OLS for the productivity equation. In a second stage we minimise the distance between the reduced form and the structural form parameters using the identification conditions. We winsorise labour productivity at 1 per cent to reduce the impact of outliers on the results and use cleaned measures of innovation, which are based on the actual descriptions of new products and processes introduced and comply with the definitions in the Oslo Manual.

Table 2 shows the number of observations, mean and standard deviation for the main variables in the various subsamples that correspond to our estimating equations. It also indicates the identifying assumptions of the structural parameters of the model.¹⁵ The β and γ coefficients are identified; for the δ coefficients to be identifiable, we need and have two exclusion restrictions; for the θ coefficients to be identifiable, we need 2 exclusion restrictions and we have 6.

Estimation Results

We now turn to the estimation results, more precisely the total marginal effects (direct and indirect) of all our control variables on the various endogenous variables of our model. We explore possible sources of heterogeneity depending on the country's level of development.

Baseline Specification

Table 3¹⁶ shows that the estimated semi-elasticities of labour productivity with respect to the quality of management practices ($\theta_{M} + \theta_{I} \delta_{M} )$ and innovation ($\theta_{I} )$ are economically and statistically significant, but not with respect to R&D ($\theta_{I} \delta_{R} )$.¹⁷ On average, high quality of management practices is associated with higher labour productivity more than the occurrence of any type of innovation. It should be noted that, whenever they appear as explanatory variables in Eqs. (2) and (3), it is the latent variables of management practices, R&D and innovation that enter and not the observed binary variables. This has the advantage of providing a measure for these variables even when they are actually reported as being equal to zero. Indeed, small values for R&D and innovation may not be reported and therefore the observed value may mis-measure the innovation effect on productivity (Crépon et al. 1998; Raymond et al. 2015).

Table 3 Average marginal effects on Ln(labour productivity) by type of innovation, baseline model.

Since the latent variables are unobserved and their magnitude has no particular meaning, a more informative way to interpret the marginal effects is to compare the means of the latent variables for firms that engage in R&D or innovation, or that have above median quality of management practices, with those for firms that do not engage in R&D, innovation or that have below median quality of management practices. The estimated differences in the means of latent variables of these two groups of firms for the subsample of observations used to estimate the productivity equation (2139 observations) are 0.754 for R&D, 0.475 for management practices, 0.682 for product innovation, 0.590 for process innovation and 0.609 for technological innovation.¹⁸ Hence, switching from below median to above median quality of management practices is associated with a 45.3 per cent ((e^0.786*0.475−1) × 100) higher labour productivity, whereas switching from not engaging in product innovation to introducing a new product is associated with a 27.5 ((e^0.356*0.682−1) × 100) per cent higher labour productivity (column 1).

The association between labour productivity and process innovation is stronger than for product innovation: engaging in process innovation is associated with a 46.2 per cent ((e^0.644*0.590−1) × 100) higher labour productivity (column 2). In the absence of complementarity or substitutability and no differences in the sample size, the marginal effect of technological innovation should be a linear combination of the marginal effects of product and process innovation. This is indeed the case: the effect is estimated at 37.8 per cent ((e^0.526*0.609−1) × 100). In other words, we estimate that switching from not engaging in technological innovation to introducing a new product or process (or both) is associated with a 37.8 per cent higher labour productivity. This is close in magnitude of the average of the magnitude of the effects associated with the introduction of a new product (27.5 per cent) and the introduction of a new process (46.2 per cent). These marginal effects of innovation are somewhat higher than those found for OECD countries (see Mohnen and Hall 2013), but they are comparable to those reported for Latin America by Crespi and Zuniga (2012). Doing R&D does not have a significant effect on labour productivity. As we noticed, few firms in these economies do R&D, and those that do are likely to also have a high quality of management. The latter seems to dominate between the two.¹⁹

The association between labour productivity and the quality of management practices is driven primarily by direct effects when coupled with product innovation and by the indirect effects, when coupled with process or technological innovation in general (see Table 3). This suggests that management practices have to be sufficiently good for firms to be able to introduce new products (the indirect effect), but even more importantly, for firms to actually be able to sell them (the direct effect; our measure of labour productivity being based on sales per employee). Indeed, the direct effect of switching from below median to above median quality of management practices is associated with a 31.9 per cent ((e^0.583*0.475−1) × 100) higher labour productivity, whereas the indirect effect of doing so (via product innovation) is associated with a 10.1 per cent ((e^0.203*0.475−1) × 100) higher labour productivity (column 1). Both effects are statistically significant—indirect at 10 per cent and direct at 1 per cent level of significance.

Management practices also have to be sufficiently good for the firms to be able to introduce new processes. However, since process innovation does not necessarily have an impact on sales per employee (it may save on labour but also involve adjustment costs and learning to use the new technology efficiently), the quality of management practices does not have much of a direct impact on labour productivity beyond its effect via process innovation. Indeed, only the indirect effect is statistically significantly different from zero in this case. The direct effect of switching from below to above median quality of management practices is associated with a 14.1 per cent ((e^0.277*0.475−1) × 100) higher labour productivity, whereas the indirect effect of doing so (via product innovation) is associated with a 27.5 per cent ((e^0.511*0.475−1) × 100) higher labour productivity (column 2).

Labour productivity is also affected by some other factors, though only the magnitude of its association with ICT usage comes close or is comparable to the magnitudes of its associations with innovation and management practices. Estimates using technological innovation in column (3) suggest that being located in the capital or the main business city is associated with an almost 19 per cent higher labour productivity than being elsewhere in the country. This may be due to better infrastructure and other resources available in the capital or the main business city. A higher labour productivity is also associated with having access to external finance (18.4 per cent), ICT usage (more than 30 per cent), obtaining an internationally recognised certification (10.2 per cent), and a higher capacity utilisation (0.5 per cent).²⁰, ²¹

The exclusion restrictions—internationally recognised certification in the management quality equation, percentage of full-time permanent employees with a university degree in the R&D equation and ICT usage and access to finance in the innovation equation—are statistically significant, indicating that those are strong variables to instrument the endogenous variables with. Following Duguet and Lelarge (2012), we have performed a test of over-identifying restrictions. As discussed, we have four over-identifying exclusion restrictions. The value of the $\chi_{4}^{2}$ statistic is 11.67, which is below the critical value of 13.28 at a significance level of 1 per cent. This result shows that the over-identifying exclusion restrictions do not significantly increase the distance between the structural and the reduced form coefficients, in other words that the way we instrument the endogenous variables is valid.

Overall, the results suggest that labour productivity benefits from both a higher quality of management practices and the introduction of innovation. This finding holds regardless of the type of innovation. Additionally, the results point to the discrepancies in the availability of infrastructure, external funding and other resources available in the capital or main business city versus other locations in the country.

The Role of Economic Development

Given the high heterogeneity of the countries in our sample in terms of income level, we run our model for subsamples by gross national income (GNI) per capita (calculated using the World Bank Atlas method) in 2007.²² The results in Table 4 for the sample split into two groups according to GNI per capita, higher-income (high-income and upper-middle-income) economies and lower-income (lower-middle-income and low-income) economies, reveal significant differences of how both channels work across these groups.^23,24

Table 4 Average marginal effects of R&D, innovation and management practices on Ln(labour productivity) by type of innovation and GNI per capita.

To begin with, the results confirm our baseline findings of a positive and significant association between innovation and labour productivity. This relationship in general also holds for management practices. It is only when we restrict ourselves to process innovations in the higher-income economies that we do not obtain a significant marginal effect for management quality.

Having said this, two major differences across the groups stand out. In the higher-income group the marginal effects of innovation are in general larger than those of management practices. Switching from not being a technological innovator to becoming one, for example, is associated with a 36.0 per cent ((e^0.501*0.614−1) × 100) higher labour productivity, while improving the quality of management practices from below to above median quality is associated with a 25.7 per cent ((e^0.490*0.467−1) × 100) higher labour productivity (column 3).²⁵ The exception is product innovation, suggesting that introducing new products may be more difficult than improving management practices in raising labour productivity even in higher-income economies. To a large extent, new products replace old products and this substitution dampens the effect on productivity. Moreover, R&D is also positively and significantly associated with labour productivity in combination with process and technological innovation (columns 2 and 3). As shown in Table 3, the association between labour productivity and the quality of management practices is driven primarily by the direct effects when coupled with product innovation, and by the indirect effects, when coupled with process or technological innovation.

In the lower-income group, the association between innovation and labour productivity is still positive and statistically significant. However, the marginal effect of management quality on labour productivity is at least three times as high as the marginal effect of innovation. Switching from not being a technological innovator to becoming one, for example, is associated with a 65.4 per cent ((e^0.753*0.668−1) × 100) higher labour productivity, while improving the quality of management practices from below to above median quality is associated with a 200.8 per cent ((e^2.047*0.538−1) × 100) higher labour productivity (column 3). Contrary to the results for higher-income countries, the association between labour productivity and the quality of management practices is driven primarily by the direct effects regardless of the type of innovation. This underscores the importance of the quality of management practices in lower-income economies. Moreover, the marginal effects of R&D are no longer statistically significant.

Overall, the point estimates suggest that while firms in higher-income economies benefit more from introducing process innovation than following high-quality management practices, firms operating in lower-income economies can improve labour productivity to a greater extent by improving the quality of their management practices. We could also interpret the result the other way round, namely that in higher-income economies, high productivity firms invest more in process innovation than in management practices, whereas in lower-income countries productivity is more related to the adoption of high-quality management practices. However, at a 95% confidence level the marginal effects of innovation and management are not statistically different from each other.

Sensitivity Analysis

Our results could be affected in four additional ways. First, while we argue that cleaned measures of innovation are more reliable than self-reported measures, the former might be affected by the cleaning effort made. Second, the estimations have so far not taken into account capital intensity. This may undermine our results as for instance the effect of innovation in determining productivity improvements could be overstated if a firm increases its capital base at the same time.²⁶ However, we refrained from including the variable in the first place as the sample size reduces to just over a third of the available sample (Table 1). Third, as shown in Table 1, our sample covers 30 mostly developing economies and the number of observations per country varies significantly, with Russia, Turkey and Ukraine making up almost half of the total sample. This runs the risk that results are affected by the inclusion of a specific country in the sample. Fourth, our estimates are based on the structural model and could thus be model-specific. We address these four issues in turn.

Self-reported Innovation Measures

To test for the robustness of results to using self-reported rather than cleaned measures of innovation, we re-estimate the baseline model (Table 3) using self-reported measures of innovation. This allows us to additionally check whether the impact is different for organisational or marketing innovation, i.e. non-technological innovation.

The results in Table 5 show that the estimated marginal effects of management practices and innovation are positive and significant regardless of the measure of innovation used and slightly higher in magnitude compared with the estimates in Table 3. For instance, engaging in technological innovation is now associated with a 38.9 per cent ((e^0.569*0.577−1) × 100) higher labour productivity, compared with a 37.8 per cent ((e^0.526*0.609−1) × 100) increase when using the cleaned measure of technological innovation. Introducing a non-technological innovation is associated with a 42.9 per cent ((e^0.688*0.519−1) × 100) higher labour productivity (column 4). Similarly, management practices also impact productivity to a slightly larger extent in this specification. As shown in Table 3, the direct effect of management practices is larger than the indirect effect in combination with product innovation. In combination with process and non-technological innovation, the direct effect of management practices on productivity is not statistically significant and is also smaller than the indirect effect, which is statistically significant. In combination with technological (product or process) innovation, the direct effect is slightly larger, but not statistically significantly different from the indirect effect.

Table 5 Average marginal effects of R&D, innovation and management practices on Ln(labour productivity) with self-reported innovation variables.

The marginal effect of R&D on labour productivity remains the same in magnitude as in the baseline model: engaging in R&D is associated with a 28.2 per cent ((e^0.329*0.754−1) × 100) higher labour productivity for process innovators (column 2). Contrary to the baseline model the marginal effect of R&D is now statistically significant, at least for process and technological innovation. This could be because with the self-reported innovation measures the sample increases by about a quarter and because those responses were more correlated with the answers to the R&D question than the cleaned responses.

Total Factor Productivity

Our results are also broadly robust to the inclusion of capital per employee in the baseline model as an additional control variable in the labour productivity equation, despite the significant reduction in sample size (Online Appendix OD). Controlling for capital intensity is equivalent to analysing total factor productivity instead of labour productivity. The marginal effects of management practices and innovation remain positive and statistically significant at least at the 10 per cent level of significance. To compare the results with and without correction for capital intensity, we run the baseline regression only for the sample for which capital per employee is available (Table OD.1 in Online Appendix OD, columns 4–6). The results indicate that some of the decline in the marginal effects of innovation is a consequence of the reduction in sample size when capital per employee is included in the regression and some is due to controlling for capital intensity. In the version with capital intensity, management practices continue to have a higher marginal effect than innovation on total factor productivity and R&D remains insignificant.

Changes in the Sample

To test for the robustness of results to changes in the sample, we re-estimate our baseline specification (Table 3), removing one country at a time from the sample. The results in Figure OD.1 in Online Appendix OD show a remarkable stability of the estimated marginal effects of the quality of management practices, R&D and technological innovation on productivity to the exclusion of one country at a time. The marginal effects of the quality of management practices and technological innovation are always positive and statistically significant. The marginal effects of R&D and management practices are somewhat sensitive to the exclusion of Turkey, but they keep their sign and significance. The results are also robust for product, process and non-technological innovation.²⁷ We thus conclude that our results are not driven by any country in particular.

OLS Results

In order to test whether our findings are an outcome of our model setup, we run a simple OLS regression with labour productivity as the dependent variable. The results in Table OD.2 in Online Appendix OD are now to be interpreted as discrete shifts of innovation, R&D or management practices quality from 0 to 1 and no longer as continuous variations. They show that both the quality of management practices and innovation are positively and significantly associated with labour productivity. Introducing a technological innovation, for example, is associated with an almost 23 per cent higher labour productivity than not being a technological innovator, while having a high quality of management practices is associated with an about 14 per cent higher labour productivity than having a low quality of management practices (column 3). Performing R&D is also associated with an approximately 22 per cent higher labour productivity. Overall, the OLS results confirm the importance of all three variables without favouring one over the others.

When interpreting these findings, it is important to remember that the OLS estimates are likely to be biased because they do not take into account the endogeneity of management practices and innovation activities. Unobservable factors such as the manager’s competence or dynamism may affect productivity, the adoption of management practices and innovation activities at the same time. Nevertheless, the OLS results do not contradict those derived from our structural equations.

Conclusion

As the contribution of firm productivity to economic growth is widely acknowledged, both researchers and policy makers are interested in the drivers of productivity. In particular, innovativeness is found to be crucial in determining firm performance (see, for example, Rosenbusch et al. 2011). Because of institutional obstacles (access to credit, corruption, poor intellectual property rights) and because of their distance to the technological frontier, firms in lower-income countries have less of an incentive to invest in R&D and innovation. Improving management practices requires less of an investment and may be more rewarding in the short term. Moreover, as Bloom et al. (2013) have shown, there is a causal relationship between management quality and firm performance.

In this paper we explore the relationship between innovative activities and management practices in determining firm-level productivity. Moreover, we analyse this relationship in different economic environments to investigate whether potential effects are dependent on the level of development. We use data from a unique firm-level survey on innovation and management practices to estimate, for the first time in the same model, the associations between the quality of management practices, innovation and manufacturing firm productivity in mostly developing countries in Eastern Europe and Central Asia, while controlling for capacity utilisation and other firm characteristics. These countries range from low-income economies such as Tajikistan to high-income economies such as Slovenia.

We find that management practices and any type of innovation are significant drivers of firm productivity. Moreover, these two factors work differently within higher- and lower-income countries. More specifically, above-median-quality management practices of firms operating in lower-income economies are associated with a stronger positive impact on labour productivity than the introduction of product, process or technological innovation. In other words, firms can achieve higher labour productivity by improving their management practices than by introducing new products and processes, with the management practices affecting labour productivity primarily directly, rather than through their impact on innovation. By contrast, in higher-income countries, firm-level management practices play a somewhat less important role in boosting firms’ labour productivity; in line with catch-up growth literature, firms need to engage in innovation instead. With the exception of product innovation, management practices are associated with labour productivity primarily indirectly, via process or technological innovation. These findings suggest that firms operating in less favourable environments are able to overcompensate non-existent innovation activities by improving the quality of their management practices, thereby overcoming potential institutional barriers and contributing to aggregate productivity.

Our findings raise the question of why firms in low-income economies do not adopt better management practices. The recent management field experiment looking at large Indian textile firms suggests that this may be due to information barriers. Firms might not have heard of some management practices, or they may be sceptical regarding their impact (Bloom et al. 2013). Improvements to certain management practices—particularly those relating to underperforming employees, pay or promotions—may also be hampered by regulations or a lack of competition (since competition could force badly managed firms to exit the market).

Training programmes covering basic operations (such as inventory management and quality control) could be helpful, but suitable consultancy or training services offering such products may not exist in a given market or may be geared towards large firms, making them too expensive for SMEs.²⁸

Policy makers in less developed countries should focus their attention on providing more basic business education and improving the quality of education in general, as well as improving the general business environment, rather than aspiring to create new Silicon Valleys.