Management Practices, Pay, and Pay Inequality

Abstract

We use rich Portuguese data to analyse the relationship between the use of different management practices and worker pay in a large representative sample of firms. We find that the overall score on the use of management practices is significantly associated with both higher average wages and higher within-firm wage dispersion. The positive relationship between management practices and average pay is present throughout the wage distribution and for all occupational skill groups, but is stronger for workers higher up in the wage distribution and in higher-skilled occupations. These results are driven by incentives-based management practices, in particular practices related to reward and promotion of high performers in the firm. We also identify a potential moderating role of works councils. Our results suggests that, in firms with works councils, a larger share of the gains from productivity-enhancing management practices is shared with workers, and these gains are more evenly distributed across the workforce.

Introduction

Management matters. This is the key message from an already sizable literature showing that differences in management practices play a crucial role in accounting for the large variation in firm performance across industries and countries. Firms that use more structured management practices (according to the definition of Bloom and Van Reenen 2007) have been shown to perform better along a number of different dimensions, such as productivity, profitability, growth, survival rates and innovation (see, e.g., Bloom and Van Reenen 2007, 2010, and Bloom et al. 2019). The abundance of literature on the importance of management practices in the last couple of decades has been greatly assisted by the development of the World Management Survey (WMS), first described by Bloom and Van Reenen (2007), which provides a significant amount of information on measures of management practices across a large number of countries.

There is however less empirical evidence on the relationship between different types of management practices and labour outcomes, most importantly wages. If better management practices improve labour productivity and thus firm performance, workers might benefit in the form of higher wages. But how such performance gains are linked to different types of management practices, the extent to which they are distributed to workers, and how they are distributed across the workforce, are still not extensively documented.¹

In the present paper we contribute to the literature on management practices and wages by using information on management practices in 2016 from a large representative sample of Portuguese firms, which allows us to construct a management score for each firm, measuring the extent of adoption of the different management practices included in the study. This data is linked with two other sources of data, including a matched employer-employee dataset, which allows us to relate the use of different management practices with a very rich set of firm and worker characteristics. In particular, we have data on several different wage measures, including disaggregated pay components such as base wages, bonus-related payments and overtime pay. The availability of these different wage components allows us to pinpoint the channels through which management practices affect average wages in a way that has hitherto not been done in the literature. We also have information on occupational categories, which allows us to look at the relationship between managerial practices and occupational skill composition, as well as wage effects within each skill category. The Portuguese survey of management practices is also unusually rich, and our constructed management score is based on 29 different questions in this survey. This allows us to study a larger set of management practices than what is common in the literature, and to pinpoint exactly which types of management practices that drive our overall results. As in Bloom and Van Reenen (2010), we group these questions into three broad categories of management practices, namely targeting, monitoring and incentives, and we also study the effects of each type of incentives-based management practices separately. By constructing separate management scores for each category of management practices, at different levels of aggregation, we are able to study whether different types of management practices have different effects on wages and wage inequality.

Our two main findings are that the overall management score is positively related to (i) average pay and (ii) within-firm pay inequality. Furthermore, we show that the positive relationship between the use of management practices and average wages applies throughout the wage distribution, even for workers in the bottom decile, but is much stronger towards the top of the distribution. The relationship also applies for all occupational skill groups, but is much stronger for workers in high-skilled occupations.

A more detailed analysis shows that the above-described relationships are largely driven by management practices related to incentives (rather than monitoring or targeting), and in particular two different incentives-related management practices, namely rewarding high performance and promoting high performers. Moreover, when we decompose overall pay into different pay components, we find that, for managers, the relationship between the aggregate management score and average pay stems from differences in base wages, while for non-managers it stems from differences in both base wages and overtime pay. Although incentives-related management practices are associated with higher bonus payments, as we would expect, this relationship is counteracted by a negative association between bonus payments and management practices related to targeting, which implies that the association between the overall management score and bonus payments is not significantly different from zero.

We also investigate potential mechanisms and mediating factors related to our main results. In line with previous literature (e.g., Lee 2018) we find that the overall management score is positively associated with a higher (absolute and relative) demand for workers in high-skilled occupations. However, our main results cannot be fully explained by differences in occupational skill composition, since the positive relationship between management scores and average wages also applies within each occupational category. We also explore the role of workers' collective bargaining power as a mediating factor. Using the presence of works councils as a proxy for workers' bargaining power, we find that such presence amplifies the first and dampens the second of our two main results. This suggests that, in firms with worker representation through works councils, the productivity gains from a higher management score are to a larger extent transmitted to workers in the form of higher average wages, but in a less unequal way across the workforce. The latter result is also confirmed if we instead use the presence of firm-level wage agreements as a proxy for (local) collective bargaining power.

Although our analysis is based on very detailed and rich data, it must be stressed that the main weakness lies in the cross-sectional nature of the same data, since the information on management practices is collected only for a single year. This makes it difficult to claim causality for our above described findings. Nevertheless, in an extension to our main analysis we test the robustness of our main results using propensity score matching to create a sample that is more homogeneous in all other observable dimensions than the use of management practices. Our regression estimates using the matched sample largely confirm our results from the main analysis, with one notable exception. These estimates also suggest that management practices related to monitoring are positively related to wages, but only for workers in the bottom half of the wage distribution.

Our analysis is related to three different literatures that differ in the way management practices are defined and measured. By adopting the World Management Survey methodology, our paper naturally belongs to the literature on management practices that has followed from Bloom and Van Reenen (2007). In this particular strand of the management practices literature, the evidence on wage effects is largely restricted to two papers, namely Bender et al. (2018) and Bloom et al. (2021).²

Both studies find a positive relationship between the use of management practices and average wages, as we do. Bloom et al. (2021) also find a positive relationship with within-firm wage dispersion, which is similar to what we find in the present paper, whereas Bender et al. (2018) find that management practices are negatively correlated with two different measures of wage inequality.³

Compared to this particular strand of the literature, our paper offers additional insights into (i) how management practices affect not only total pay but also each different pay component, (ii) which types of management practices matter for wages and wage inequality, and (iii) how the relationship between management practices and wages depends on labour market institutions such as the presence of works councils or firm-level wage bargaining.

The second literature our paper relates to is an earlier literature on various types of human resource management practices, often referred to as high-performance work systems or high-involvement management, which include practices such as decentralisation, teamwork, job rotation, quality programs, etc. Several papers in this literature look at wage effects. For example, Osterman (2006) finds that high-performance work systems are associated with higher wages but have no effect on wage inequality, whereas Forth and Millward (2004) find that high-involvement management is positively associated with higher pay, and more so when trade unions are involved in bargaining. Similar positive wage effects are reported by Böckerman et al. (2013), who also find that large parts of these effects are caused by worker selection.⁴

In general, the papers in this literature tend to analyse either separate effects of different types of management practices or the joint effect of a small bundle of different practices. Moreover, whereas some of these practices overlap with the set of management practices studied in the present and other papers inspired by Bloom and Van Reenen (2007), others are conceptually quite different. These differences in both the measurement and definition of management practices make it somewhat harder to perform a direct comparison with the results in the present paper.

There is also a third relevant literature with a more comparable but also much narrower focus, namely the literature on performance pay, which is one of several management practices considered in our paper. Several studies in this literature find a positive effect of performance pay on average wages (e.g., Booth and Frank 1999; Pekkarinen and Riddell 2008; Pompei et al. 2019). Increased use of performance pay has also been identified as a potential driver of increased wage inequality (e.g., Lemieux et al., 2009; Bryan and Bryson 2016). Perhaps the most closely related paper in this strand of the literature is Barth et al. (2012) who analyse the effect of performance pay on within-firm wage inequality and the potentially moderating role of collective bargaining. They find that the use of performance pay leads to higher wage inequality in non-unionised firms but not in firms with a high union density, which suggests that the inequality-inducing effect of performance pay is mitigated by collective bargaining power. This result is broadly in line with our findings regarding the mediating effects of works councils and firm-level wage bargaining. Overall, though, the main contribution of our paper in relation to the performance pay literature, which focuses on a single management practice, is that we broaden the analysis to also consider a wide range of other practices, where some are incentives-based and others are not. This allows us to assess the relative importance of performance pay in the overall package of relevant management practices.

To sum up, the main contributions of our paper are threefold. First, since we have access to very detailed wage information, we look at the relationship between management practices and a number of different wage measures, including separate wage components such as base wages, bonus payments and overtime pay. This differs from the existing literature and allows us to obtain new and more detailed insights into the relationship between management practices and wages. Our data also allows us to control for hours worked and therefore to isolate wage effects that are merely caused by workers working longer hours. Second, we perform our analysis with different levels of aggregation in the management practices dimension in order to pinpoint exactly which management practices matter for pay and pay inequality. This also differs from the existing literature, which is either based on only an aggregate management score, as in Bender et al. (2018) and Bloom et al. (2021), or only a single management practice, as in the literature on performance pay. Third, we analyse the interaction between management practices and labour market institutions in order to discover if and how the relationship between management practices and wages are influenced by the presence of works councils or firm-level wage bargaining. In this sense, our paper complements and adds to the recent literature on how the presence of works councils might affect the functioning of management practices (Jirjahn et al. 2024; Addison et al. 2023).⁵

The rest of the paper is organised as follows. In the next section we offer a brief description of some key institutional details regarding wage determination in the Portuguese labour market. In Section "Theoretical Framework" we discuss some relevant theoretical mechanisms that could explain if and how the use of different management practices affect wages and wage inequality. In Section "Data and Descriptive Statistics" we present our data and the construction of our main variables, along with some descriptive statistics on the adoption of management practices in Portugal and their relationship with wages. Section "Empirical Analysis" contains the empirical analysis. A summary and some final remarks are given in Section "Concluding Remarks".

Institutional Background

Although union density rates are quite low in Portugal, a high degree of union bargaining coverage means that the vast majority of workers in Portugal are covered by a collective wage contract.⁶

These agreements establish the minimum working conditions for each professional category or group of workers, including the monthly base wage, the overtime pay and the normal hours of work. There are three types of collective agreements: sectoral, multi-firm and firm. By far the most dominant type of collective contract is sectoral agreements, which often cover a whole industry or occupation. In our sample (which we describe in detail in the next section), around 75 per cent of workers are covered by this type of wage agreement. In contrast, multi-firm and firm-level agreements are relatively rare. In our sample, only 2.3 per cent of workers are covered by firm-level wage agreements, whereas the prevalence of multi-firm agreements is even lower.⁷

However, what is determined by the collective wage contracts is only a wage floor. Wage determination in Portugal is characterised by a two-tiered wage setting system where firm-specific arrangements result in a mark-up, usually referred to as a wage cushion (Cardoso and Portugal 2005), on top of the bargained wage floor. This wage cushion can in some cases be of a considerable magnitude.⁸

Thus, the two-tiered wage setting system leaves considerable room for local wage flexibility in response to firm-specific conditions, such as firm-level rent sharing resulting from the adoption of productivity-enhancing management practices, for example.⁹

Besides trade unions, an alternative formal channel of collective influence of workers is through works councils. The emergence of works councils is a relatively recent trend in the Portuguese labour market. Though the first experiences date back to the April revolution in 1974, the first legal piece was enacted as recently as 2009 (law 7/2009 of 12th of February) with an amendment to the Labour Code. This law establishes two channels for employee representation at the workplace, namely trade union delegates and works councils. The latter can be set up in public or private firms at the request of the employees. The formal influence of works councils is limited to information and consultation, with no formal veto power in decision making. The presence of works councils in Portugal is still relatively scarce and is found mostly in larger firms. In our sample, works councils are present in around 9 per cent of the firms.

In terms of labour market outcomes, such as wage levels and wage inequality, Portugal is somewhat of an outlier among otherwise comparable European countries. Portugal is characterised by a very low average wage level and had the highest wage inequality in Europe in the mid-2000s (Oliveira et al., 2023; Oliveira 2023). The wage inequality is particularly wide at the upper part of the wage distribution, a pattern closer to Anglo-Saxon countries, though Portuguese labour market institutions differ substantially, with a high degree of employment rigidity and fairly centralised wage bargaining, as explained above. Since the mid-2000s, though, wage inequality has reduced due to a strong compression of the lower part of the wage distribution, which is mainly caused by a 55% increase in the legal minimum wage (Oliveira 2023). Nevertheless, the average wage remains at a low level, ranking fourth from bottom among EU countries (and 8th from bottom among OECD countries) in 2022 (OECD 2024).

Finally, as we show in detail in Section "Data and Descriptive Statistics", Portugal is a country that is generally characterised by a relatively low adoption of the types of management practices that we consider in this paper. In particular, the use of incentives-based management practices is uncommonly low (see Section "Descriptive Statistics" for details). Our study is therefore well suited to analyse the effects of different management practices in the context of a managerial culture that is generally much less incentives-based than what is the case for much of the existing literature.

Theoretical Framework

To the extent that management practices affect wages, an obvious channel of influence is via labour productivity. A positive correlation between management practices and productivity is well-documented in the empirical literature (e.g., Bloom and Van Reenen 2010; Bloom et al. 2019). Theoretically, management practices might affect labour productivity in two different ways: (i) by making existing workers more productive, and/or (ii) by inducing productivity-enhancing changes in the workforce composition.

Management practices might increase the productivity of existing workers by inducing higher effort. This could be achieved through several of the different dimensions that are included in our management score, in particular practices related to monitoring and incentives. For example, higher worker effort might be induced by incentivising workers through performance pay, through rules for promotions that are based on performance rather than seniority, or through monitoring that reduces shirking. Performance-based promotions might also increase firm-level labour productivity through allocational efficiency gains if the more able workers are promoted to positions where they have a larger impact on the firm's performance.

A higher firm-level labour productivity could also result from management practices that change the firm's workforce composition, for example practices that allow the firm to better identify and retain its most productive workers. Management practices that reward individual performance–-either directly through performance pay or indirectly through a higher probability of promotion–-might also attract more productive workers to the firm and make it easier for the firm to recruit high-ability workers (Lazear 1986, 2000; Booth and Frank 1999).

On the other hand, there might also be potentially adverse productivity effects of certain types of management practices. For example, the use of explicit incentives in the forms of rewards and punishments might have adverse productivity effects due to a crowding out of intrinsic motivation (Bénabou and Tirole 2003), or because such practices are perceived as being hostile (Fehr and Falk 2002) or unfair (Heinz et al. 2020). In the same vein, there is also some evidence suggesting that excessive monitoring might harm productivity (Banker et al. 2010). Thus, although the existing literature suggests a positive relationship between management practices and firm-level labour productivity on average, it is far from obvious that all types of management practices related to incentives and monitoring, in particular, have productivity-enhancing effects in all contexts.

A positive relationship between labour productivity and wages is well-established, both theoretically and empirically. Thus, if management practices affect labour productivity, wages are likely to be affected as well. Even for a given workforce composition, the gains from an increase in firm-level labour productivity are (at least to some extent) transmitted to workers in the form of higher average wages under a number of different relevant assumptions about labour market institutions and wage-setting mechanisms, including collective bargaining, “fair wages”, and monopsony wage setting. If wages result from Nash bargaining between firms and trade unions, higher labour productivity will typically translate into higher bargained wages (e.g., Dowrick and Spencer, 1994). If wages instead result from firms' incentives to induce the desired amount of effort from its workers, as in the fair wage hypothesis of Akerlof and Yellen (1990), and if workers' notion of a fair wage is based on an internal reference which reflects the firm's ability to pay (e.g., Bastos et al. 2009), then productivity gains would be partly transmitted to workers through an increase in what is considered to be a fair wage. Or if wages are determined by firms facing upward-sloping labour supply curves due to monopsony power, for example because of market concentration on the demand side or heterogeneous job preferences on the supply side (Manning 2021), productivity increases will shift out the labour demand curve and thus lead to higher wages.

Whereas productivity-enhancing management practices are likely to result in higher average wages, exactly how these wage increases are materialised and how within-firm wage inequality is affected, is a priori more uncertain. Incentive-based management practices such as performance pay are likely to show up in the form of increased bonus payments, and possibly also more overtime pay if workers are induced to work longer hours (Green and Heywood 2023). Such payment schemes will also likely lead to higher wage inequality due to worker heterogeneity with respect to ability and the propensity to exert effort (e.g., Lemieux et al., 2009; Bryan and Bryson 2016), or if they predominantly apply to workers in the upper part of the wage distribution. On the other hand, management practices that lead to general productivity gains through a better organisation of production, for example, might instead be transmitted to all workers in the form of higher base wages.

A potentially important mediating factor in the relationship between management practices and wages is the role of labour market institutions. The presence of collective bargaining power, in the form of trade unions or works councils, is likely to affect how productivity gains are transmitted and distributed to workers in the form of higher wages. Since trade unions are often considered to have somewhat solidaristic or egalitarian objectives (e.g., Agell and Lommerud, 1993; Koskela and Stenbacka 2010), a plausible hypothesis might be that productivity gains are distributed to workers in a more equal way in the presence of such labour market institutions (e.g., Barth et al. 2012). For example, part of the productivity gains caused by incentive-based payment schemes might be redistributed to all workers in the form of higher base wages.

Works councils might also play an additional role in the way management practices affect productivity and performance. It has been suggested in the literature (e.g., Jirjahn et al. 2024) that management practices of the types considered here have a stronger productivity-enhancing effect in the presence of works councils. The idea is that formal worker representation of this kind leads to increased information sharing and strengthened employer-employee trust in a way that improves the functioning of management practices and thereby makes them more effective. The potentially moderating effects of labour market institutions such as works councils will be further explored in our empirical analysis below.

Data and Descriptive Statistics

In this section we first present our data and explain how we construct the relevant variables for our analysis. We then proceed to present some figures that describe the adoption of management practices in Portugal and how it varies with different wage measures.

Data and Variables

Our main source of data is Inquérito às Práticas de Gestão (IPG), collected by the National Statistics Institute (INE). This is a non-periodical compulsory survey collected only once (during the period between June and August 2017) at firm level, asking which management practices were in place in 2016. The survey aims to capture the perceptions of managers and top executives about the importance of their management practices for firm productivity.

IPG collects information from a representative stratified sample of all firms operating in Portugal in the non-financial private sector, excluding firms with less than five employees. The sample consists of 4469 Portuguese firms based on the following criteria: economic activity, firm age, size, and whether or not the firm belongs to a conglomerate. The final number of valid responses is 3875, which corresponds to a very high response rate of approximately 87 per cent.

The IPG survey includes questions organised in three different areas of management: (1) strategy, monitoring and information, (2) human resources and (3) management and social responsibility. It also includes detailed information on managerial delegation of decisions and leadership.¹⁰

For our purposes we select 29 (out of 55) questions from the survey, which are described in detail in Table 12 in the Appendix. The questions refer to operational aspects (2 questions), performance monitoring (10 questions), target setting (4 questions), and incentives to workers (13 questions). These questions closely follow those from the World Management Survey (WMS) and from the Management and Organizational Practices Survey (MOPS) supplements of 2010 and 2015 in the US. Previous research based on WMS has used 18 questions (e.g., Bloom and Van Reenen 2007), while more recent works based on MOPS have used 16 questions (e.g., Bloom et al., 2019). Our 29 questions include (very similar versions of) all the 16 or 18 questions used in the previous literature, in addition to some additional questions within the same general categories.

Following previous work (e.g., Bloom et al. 2019), we scale the possible responses to each question between 0 and 1, where 0 and 1 correspond, respectively, to the lowest and highest adoption of the management practices, as defined by the response options of each question. We then compute an overall measure of management practices adoption–-a management score–-which corresponds to the simple unweighted mean of the non-missing answers. In our final sample, 74 percent of the firms reply to all 29 questions, while the remaining firms answer at least 20 out of the 29 questions. This response rate is quite similar to previous studies, for example Bloom et al. (2019) or Broszeit et al. (2019) who compute the management score when the number of answers was at least 10 or 11, respectively, out of 16. However, as a robustness check we also investigate whether our main results are qualitatively similar if we instead restrict our sample to only firms that answered all questions.

We also decompose the overall management score by grouping the 29 questions into three broad categories of management practices, following Bloom and Van Reenen (2010). The first category consists of practices related to monitoring, i.e., how much of what happens inside the firm is monitored and used for improving procedures. The management score in this category is calculated as the average score across questions 1, 2 and 7–16 in Table 12. The second category consists of management practices related to targeting, i.e., the extent to which the firm defines and tracks goals over time and takes action if needed. The management score in this category is calculated as the average score across questions 3–6 in Table 12. The final category consists of practices related to incentives, i.e., the extent to which the firm rewards and promotes its workforce, and how it attracts the best workers. The management score in this category is given by the average score across questions 17–29 in Table 12. For this latter category, which turns out to play a crucial role for our results, we also perform a further decomposition into four different incentives-related management practices, namely fixing poor performers (questions 17–20), rewarding high performance (questions 21–25), promoting high performers (questions 26–27) and retaining human capital (questions 28–29).

In order to connect management practices to firm characteristics and labour outcomes, we link the IPG survey to information from two different censuses of Portuguese firms provided by INE, namely Sistema de Contas Integrado das Empresas (SCIE) and Quadros de Pessoal (QP). SCIE consists of data gathered from two compulsory financial statements (balance sheet and income statement) and gives access to a rich set of information about each firm. Key variables include gross output, value added, capital stock, exports, wage bill (for managerial and non-managerial occupations separately), industry affiliation and a firm exit indicator. In addition, the dataset includes workforce characteristics such as gender distribution and share of part-time workers.

QP, on the other hand, is a particularly large and informative dataset collected annually by the Portuguese Ministry of Labour and Solidarity since the early 80s. It consists of a matched employer-employee database for all firms with wage earners and contains a high number of variables/concepts that meet international standards about each unit (firm or employee) observed. For instance, for each firm the dataset gives the location, level of employment, economic activity, ownership and total sales. Similarly, for each employee, usual human capital variables, such as gender, level of schooling, tenure, age, occupation, full-time/part-time status, earnings, length of working time and mechanisms of wage bargaining, among others, are provided.

Our final sample used for the empirical analysis is constructed by imposing a condition that firms have non-missing observations on all variables from the three datasets (IPG, SCIE and QP). This condition implies that 259 firms are dropped from the initial 3875 firms in the IPG survey (234 of these firms do not appear in QP, whereas another 25 firms have missing information on some variables). Finally, we impose a further restriction on the sample by excluding micro firms (with less than 11 workers), leaving us with a final sample of 2792 firms.¹¹

Descriptive Statistics

Summary statistics for the main variables used in our analysis are found in Table 13 in the Appendix. Below we just show some figures illustrating how the use of management practices varies across the firms in our sample, and how management scores correlate with various wage and wage inequality measures.

Figure 1 shows the distribution of the overall management score of the 2792 firms in our final sample, where the bars represent the actual data and the dark line shows a smooth kernel fit. The average management score in our sample is 0.46, which means that, on average, firms adopt 46 per cent of the pre-defined management practices. These figures are significantly below to the ones found in the US (0.615 in 2015), Finland (0.59 in 2016) or Germany (0.57 in 2013) and which refer solely to the manufacturing sector.¹²

Fig. 1

Distribution of management scores  

This is in line with an earlier comprehensive international study by Bloom and Van Reenen (2010) for medium-sized firms, which places management in Portugal below the international average, only above Brazil, India, China and Greece.¹³

Overall, the distribution of management practices shown in Fig. 1 confirms the low level of adoption of management practices in Portugal.

In Fig. 2 we show the distribution of management scores for each of the three aforementioned subcategories of management practices: monitoring, targeting and incentives. Interestingly, Fig. 2 reveals that the distribution of the overall management score (in Fig. 1) largely reflects the shape of the distribution of monitoring practices. In contrast, the score distribution for targeting and incentives are much more left- and right-skewed, respectively. In particular, the remarkably low use of incentives-based management practices is a striking feature of the management style in Portugal. Bloom and Van Reenen (2010) and Ohlsbom and Maliranta (2021) point to strict job market regulations and high union membership rates as possible explanations for the low use of incentives. In an international context, the relatively larger emphasis on targeting and monitoring practices places Portugal closer to the profile of countries like Japan, Germany or Sweden, as opposed to the US management style that to a much larger extent is based on incentives.¹⁴

Fig. 2

Distribution of management scores by category

Our descriptive statistics also confirm that management scores tend to be higher in larger firms, which is consistent with findings from several other countries (e.g., Bloom and Van Reenen 2010; Broszeit et al. 2019; Forth and Bryson 2019), and also tend to be lower in firms with very low average skill levels of workers and managers, which is consistent with previous findings of complementarity between management practices and employee skills (Bender et al. 2018; Lee 2018).¹⁵

Finally, an indication of the relationship between management practices and wages is depicted in Fig. 3, which shows the raw association between management scores and six different measures of the level and within-firm distribution of wages. For each of these measures, the figure depicts the median level of the measure in each decile of the distribution of firms when ranked according to the management score, and we also distinguish between managerial and non-managerial pay.

Fig. 3

Management practices and firm pay

In the top row of Fig. 3 we display three measures of pay for the whole sample of workers, namely (i) annual labour costs per worker, (ii) average monthly pay (for full-time workers aged between 17 and 68) and (iii) hourly wage for the same workers. Whereas (ii) and (iii) consist only of wage payments to workers, (i) is a conceptually different pay measure in the sense that it encompasses all the firm's labour costs, including insurance, payroll taxes, perks, etc.¹⁶

However, all three pay measures show a clear positive (though not always strictly monotonic) association with management scores, and this association does not seem to be driven by variation in working hours.

In the bottom row of Fig. 3 we include one measure of within-firm wage inequality, namely the standard deviation of (the logarithm of) monthly wages, and we also split the average annual pay between managers and the rest of the workforce. Once more, we observe a positive correlation between management scores and each of these three variables. Thus, higher management scores appear not only to be associated with higher average pay, but also with higher pay inequality within the firm. It is also worth noticing that the positive correlation between management scores and average annual pay seems to be more pronounced for managers than for non-managers.

Empirical Analysis

We now turn to the empirical analysis where we use the above described data to estimate the relationship between management scores and several different measures of firm-level pay and pay inequality. More specifically, we estimate the following model:

$$ln{W}_{i}={\alpha }_{0}+{\alpha }_{1}{M}_{i}+\beta {\Phi }_{i}+\theta {\psi }_{i}+\gamma {H}_{j}+{\tau }_{k}+{\rho }_{r}+{\varepsilon }_{i}$$

(1)

where ${W}_{i}$ is a pay (or pay dispersion) measure defined at firm level. Our main pay measure is the monthly wage of full-time workers in each firm using information from our matched employer-employee dataset (QP). This wage measure is computed for the overall workforce and for narrower groups of workers, defined either according to their position in the wage distribution or by their occupational status. We also use the (logarithm of) monthly wages to compute two different measures of within-firm pay dispersion, namely the standard deviation and the coefficient of variation. In addition, we also decompose the monthly wage into three different components, namely base wage, bonus-related payments, and overtime pay. Furthermore, since the QP dataset also includes information about the number of hours worked, we also perform robustness checks using hourly wages instead of monthly wages.

In addition to all the different pay and pay dispersion measures based on the reported monthly wage from the QP dataset, we also include a conceptually different pay measure (from a different dataset) that we refer to as annual labour costs per worker. We construct this measure as payroll expenses per worker obtained from annual accounts in the SCIE dataset. These expenses encompass all the firm's labour costs, including insurance, payroll taxes, perks, etc. Although this variable is conceptually different and not just an aggregation of the monthly wage variable, the two different pay measures are clearly correlated, as indicated by Fig. 3 in Section "Data and Descriptive Statistics".

The vector ${\Phi }_{i}$ contains firm characteristics such as size (log of employment), age (years), ownership and governance (indicators for publicly owned and family-owned or family-managed firms, and an indicator for the presence of works councils), two variables measuring the global engagement of the firm (whether the firm is a multinational or an exporter), and the share of workers covered by firm, multi-firm, sectoral or other types of wage agreements. The vector ${\psi }_{i}$ contains a set of variables describing Firm i's workforce composition, including two continuous measures of the level of education and skill of the firm's workforce (share of workers with at least one degree and share of high-skilled workers defined by occupation), the average age and tenure of the workforce, the share of females, and the share of new hires in the firm,

Industry-specific characteristics are controlled for by the variable ${H}_{j}$, which measures market concentration (by the Herfindahl–Hirschman Index) in industry $j$, and an industry indicator variable ${\tau }_{k}$, where $k$ and $j\subset k$ are defined at the 2- and 5-digit industry levels, respectively. The model also includes an indicator variable ${\rho }_{r}$ for region $r$, defined at the NUTS-2 level, in addition to the residual term ${\varepsilon }_{i}$. The standard errors are robust to arbitrary heteroscedasticity.

A potential concern with the estimation of Eq. (1) is that management practices might be correlated with the error term of the model. This can occur if relevant variables are omitted from the model specification (omitted variable bias) or if the outcome variable is not only a response to management practices but also its predictor (reverse causality). Reverse causation from wages to practices adoption might arise due to differences in unobserved worker characteristics across firms. For example, it might be the case that certain management practices are more likely to be adopted in firms with more able workers, or with workers that are more willing to exert effort, either by working more hours or working better or smarter. Management practices and wages might also be simultaneously determined, for example if the implementation of practices is costly and requires the firm to hire high-skilled workers with deeper knowledge of modern management practices. In these cases, we would observe a spurious positive correlation between pay and management that is not causal.

Although the absence of convincing instruments does not allow us to fully address these concerns, we try to mitigate their potential effects in different ways. First, we control for worker and skill composition heterogeneity in two different ways. In addition to the controls for worker skills in the main regressions, we also extend our main analysis by separately estimating the relation between management practices and wages for four different skill levels (in Section "Skill Composition"). Second, we use hourly wage as an alternative pay measure to control for differences in worker preferences regarding the total number of hours worked (in the Appendix). Third, we use a complementary econometric technique–-combining propensity score matching with OLS regressions–-to test the robustness of our findings (in Section "Robustness: Propensity Score Matching").

Management Practices and Pay: Main Findings

Table 1 reports the estimated relationship between the overall management score and average pay, where we distinguish between managers and non-managers. Panel A shows estimation results using annual labour costs per worker as the dependent variable. This variable is extracted from the SCIE dataset which distinguishes between managerial and non-managerial labour costs.¹⁷

Table 1 Management practices and pay level 

Panel B, on the other hand, shows estimation results using monthly wages as the dependent variable. This variable is extracted from worker-level data in the QP dataset and therefore allows us to define managers versus non-managers according to their positions in the wage distribution (we define managers as the top 1% earners in the firm). The QP dataset also includes information that alternatively would allow us to define managers according to about occupational category. However, this implies a sizeable loss of data (due to a lot of missing observations for the occupational category variable), which arguably makes it preferable to define managers somewhat less precisely by their placement in the wage distribution. Nevertheless, despite a large drop in the number of observations, we obtain very similar results as in Panel B if we instead define managers according to occupational category in the QP dataset.¹⁸

A higher management score is significantly associated with higher average pay, regardless of which pay measure we use (annual labour costs per worker or monthly wages), and the magnitude of this relationship is also sizeable. Furthermore, we find that this relationship is considerably stronger for managers than for non-managers. The only qualitative difference in the estimation results for the two different pay measures is that, for non-managers, the relationship between the management score and annual labour costs per worker is not statistically significant, though the point estimate has a positive sign. This lack of statistical significance could potentially be explained by the sizeable reduction in the number of observations (compared to Panel B), but it might also indicate some degree of substitution between wage and non-wage benefits to workers, since the latter are included in the annual labour costs variable but not in the monthly wage variable.

In Table 1 we also report regression estimates without controls for firm characteristics and/or workforce composition, and we can see that the magnitudes of the estimates drop considerably when including these controls. Interestingly, controlling for firm characteristics seems to be more important when estimating the effects on managerial pay, while controlling for workforce composition appears to be relatively more important when estimating the effects on average wages for the remaining workforce. The drop in the magnitude of the coefficient estimates in column (3), relative to column (2), suggests that the estimated effects of management practices on wages in the regression without workforce controls (column (2)) are partly capturing the effect of workforce composition, as discussed in Section "Theoretical Framework".

In Table 2 we show the estimated relationship between the management score and different measures of wage dispersion for all workers (managers and non-managers).¹⁹

Table 2 Management practices and pay inequality

These measures are all based on the monthly wage variable, which is measured at worker level and therefore allows us to construct within-firm wage dispersion measures. In the first two columns we estimate the relationship between management scores and two aggregate measures of wage dispersion, namely the standard deviation and the coefficient of variation. In the remaining four columns, we show estimates for workers in different percentiles of the wage distribution.

The results in Table 2 reveal that a higher management score is not only associated with higher average pay, but also with a significantly higher within-firm wage inequality, as evidenced by a significantly positive association between the management score and two different aggregate measures of wage dispersion.²⁰

Furthermore, the results in columns (3)-(6) indicate that the higher wage inequality is not only caused by a larger pay difference between managers and non-managers. Although the positive relationship between the management score and the average wage is statistically significant throughout the wage distribution, even for workers in the bottom decile, the magnitude of the coefficient decreases monotonically as we move down the wage distribution.

In Table 3 we re-estimate the relationships shown in Tables 1 and 2, but instead of using the overall management score as the dependent variable, we use management scores based on each of the three aforementioned subcategories of management practices, related to monitoring, targeting and incentives. The results are quite striking, showing that only management practices related to incentives are consistently associated with higher wages and increased wage inequality. For the other categories of management practices, the relationships with average wages and wage inequality are almost exclusively not statistically significant. Thus, the main results shown in Tables 1 and 2 seem to be almost entirely driven by management practices related to incentives.

Table 3 Categories of management practices and pay

We also check the robustness of the results reported in Tables 1, 2 and 3 along two different dimensions. First, we re-estimate the results using a more restricted sample consisting only of firms that answered all the 29 questions described in Table 12. These results are reported in Tables 14 and 15 in the Appendix and are qualitatively quite similar to the results shown in Tables 1, 2 and 3, particularly for outcome variables based on monthly wages. We also re-do the analysis using a set of dependent variables based on hourly wages, in order to eliminate potential effects related to differences in the number of hours worked. These results are shown in Table 16 in the Appendix and also reveal a picture that is very similar to the one emanating from Tables 1, 2 and 3.

The significant effects of incentive-related management practices shown in Table 3 are broadly in line with previous findings of a positive effect of performance pay on average wages and wage inequality (e.g., Booth and Frank 1999; Lemieux et al., 2009; Barth et al. 2012; Bryan and Bryson 2016). However, performance pay is only one of four different sub-categories of management practices related to incentives in our data (see Table 12 in the Appendix). In Table 4 we therefore perform a further disaggregation of the management score by re-estimating the relationships in Tables 1 and 2 for each of these four sub-categories.²¹

Table 4 Incentives-related management practices and pay

The results in Table 4 are quite illuminating and reveal that the main results in Tables 1 and 2 are basically driven by two different management practices, namely rewarding high performance, which is essentially the use of performance pay, and promoting high performers (i.e., to base promotions on performance instead of other criteria such as seniority). And out of these two incentive-based management practices, the use of performance pay seems to be the quantitatively most important one. Management practices related to hiring and retaining young talented workers is also significantly associated with average wages for non-managers, but in a negative way. Thus, even when controlling for the average age and tenure of the workforce, these particular management practices seem to have a negative wage effect.

We also perform a more disaggregated analysis with respect to the average wage measure, where we estimate the relationship between management scores and wages for each of the three components of workers' monthly wage, namely (i) base wage, (ii) bonus-related payments, and (iii) overtime pay. As before, we perform this analysis both for the entire workforce (Panel A) and for managers (Panel B) and non-managers (Panel C) separately. The results are displayed in Table 5, where we show the estimated coefficients for the overall management score and for the score on each of the three dimensions of management practices separately.

Table 5 Management practices and pay components

We see that a higher overall management score is significantly associated with higher base wages for both managers and non-managers. For the latter category of workers, there is also a significantly positive association between the overall management score and overtime pay. Of course, the absence of a significant effect on overtime pay for managers is not surprising, given that managers often are paid according to contracts where overtime is not a relevant concept. For the remaining wage component, namely bonus-related payments, we do not find any significant association with overall management scores, although the point estimates are positive and quite sizeable both for managers and non-managers. At first glance, the lack of statistical significance for this variable is perhaps a bit surprising, given our results in Table 3 showing that the positive relationship between management scores and average wages is mainly driven by management practices related to incentives.

A more illuminating and comprehensive picture is revealed when we disaggregate the overall management score into three separate dimensions. Here we see that management practices related to incentives are significantly and positively associated with a higher value of each of the three wage components, both for managers and non-managers (with the intuitive exception of overtime pay for managers). Other types of management practices appear to have less significant and somewhat mixed effects. The most consistent pattern here is that practices related to targeting are negatively related to bonus payments, which explains why the estimated relationship between the overall management score and this particular pay category is not statistically significant.

Finally, in Table 6 we also estimate the relationship between each type of incentive-based management practices and each monthly wage component.²²

Table 6 Incentives-related management practices and pay components

Consistent with the results shown in Table 4, only practices related to rewarding and promoting high performers yield statistically significant (and positive) coefficient estimates, with the use of performance pay being the most important management practice, particularly for the average remuneration of non-managers.

We would like to highlight two interesting patterns from the results shown in Table 6. First, the use of performance pay is not only positively related to bonus payments and overtime pay, but also to base wages. We can think of two possible (and not mutually exclusive) interpretations of this result. One possibility is that this is caused by workforce composition effects that are not fully controlled for in our regression estimates. If more use of performance pay attracts higher-skilled and more productive workers to the firm, as discussed in Section "Theoretical Framework", this would likely also be reflected in higher base wages on average. In Subsection "Skill composition" below we further explore the role of skill composition in explaining our main results. Another possibility is that this result reflects a redistribution of productivity gains within the firm, where parts of the gains from higher effort induced by performance pay for some workers are redistributed to other workers in the form of higher base wages. In Subsection "The Role of Labour Market Institutions" below we explore the role of labour market institutions (more specifically works councils and firm-level wage bargaining) as a potentially mediating factor in the way gains from productivity-enhancing management practices are distributed across the workforce.

The other pattern worth noticing is that management practices related to rewarding high performance and promoting high performers, respectively, seem to have significantly different effects for managers and non-managers. More specifically, for non-managers the use of performance pay translates into higher bonus payments, while for managers bonus payments are instead significantly related to the use of performance-based promotions. This might suggest that, for managers, performance-based incentives are more effective if they are also tied to promotions, perhaps because promotion opportunities are higher and more valuable among managers than for workers further down the job hierarchy.

Skill Composition

Whereas management practices might make existing workers more productive, for example through a better use of incentives, they might also enable firms to hire more skilled workers, thus leading to changes in the skill composition of the workforce. Although we control for worker skills in our main regressions, some unobserved heterogeneity might still remain. In order to explore this further, we introduce an alternative way of controlling for skill composition by classifying workers into four different skill groups according to the ISCO-88 classification (ILO 1990). A description of each skill group and a list of occupational categories in each group are provided in Table 17 in the Appendix.

We estimate different versions of a regression equation similar to (1) for each of the skill groups, using monthly wage, employment and employment share as dependent variables. These regressions are estimated on the full sample of firms and on a sub-sample consisting only of firms in which all four skill groups are present. The results are presented in Table 7 and reveal a relatively clear picture. First, if we consider the full sample of firms, higher management scores are significantly associated with lower demand, both in absolute and relative terms (Panel B and C), for the lowest-skill occupation (skill level 1). The results in columns (5)-(8), which are based on firms with workers in all skill groups, and therefore makes it easier to compare relative demand across these groups, also suggest that a higher management score is associated with a monotonic shift in demand towards higher-skilled occupations. In other words, firms with a higher management score tend to have a different occupational skill composition with a higher share of workers in high-skilled occupations. This result resembles the result reported by Lee (2018), who finds that modern management practices increase the relative demand for skilled (in particular technical) workers. Given that workers in higher-skilled occupations are on average paid more, our results in Table 1 could therefore to some extent be explained by differences in occupational skill composition that are not fully controlled for in our regression estimates.²³

Table 7 Management practices and occupational skill levels

However, the results in Panel A also reveal a significantly positive relationship between management practices and average wages within each occupational skill group. This relationship is considerably stronger for high-skilled than for low-skilled occupations, which is consistent with our results in Table 2 showing that the positive relationship between management scores and wages is stronger for high-wage than for low-wage workers. Moreover, the magnitudes of the coefficient estimates for the four skill levels in Panel A of Table 7 are roughly comparable with the magnitudes of the coefficient estimates for the four partitions of the wage distribution in the last four columns of Table 2. Overall, this suggests that neither the relationship between management scores and firm average wages, nor the relationship between management scores and within-firm wage inequality, can be fully explained by differences in occupational skill composition. A significant part of these relationships might instead be explained by productivity-enhancing effects of management practices for the firms' existing workers, for example through incentives for exerting higher effort, as discussed in Section "Theoretical Framework".

The Role of Labour Market Institutions

In this subsection we explore the potential role of labour market institutions as a mediating factor in the relationship between management scores and wages. If management practices improve firm performance, the extent to which these gains are transmitted to workers in the form of higher wages, and exactly how these wage increases are distributed across the workforce, are likely to depend on the collective bargaining power of workers. In the absence of a direct measure of collective bargaining power at firm level, we will use two different proxies, namely the presence of (i) works councils or (ii) firm-level wage agreements. All else equal, worker representation in the form of works councils, which operate at firm level, are likely to increase workers' influence over the distribution of productivity gains.²⁴

A similar effect might be present for firms in which wage bargaining takes place at firm level. As explained in Section "Institutional Background", the vast majority of the firms in our sample are subject to wage bargaining at a relatively centralised level, which implies that the bargained wage floor is not likely to reflect characteristics at individual firm level to any significant degree, and that wages to a lesser extent reflect local collective bargaining power. However, some firms bargain with trade unions at firm level, which implies that the bargained wages are likely to be more responsive to local productivity changes and that the distribution of productivity gains are more influenced by local collective bargaining power.

In Panel A of Table 8 we reproduce the main results from Tables 1 and 2 showing the separate effect of the presence of works councils. We see that, all else equal, firms with works councils have both higher annual labour costs per worker and lower wage inequality (due to higher wages in the lower parts of the wage distribution) than firms with no such worker representation, which is consistent with our underlying assumption that the presence of works councils is a proxy for workers' collective bargaining power.

Table 8 Management practices, works councils and pay

The more interesting results emerge in Panel B, where we interact the management score variable with the works council indicator variable. The estimated coefficients associated with this interaction term indicate that the relationship between management practices and worker pay is indeed mediated by the presence of works councils. More specifically, the presence of works councils seems to (i) strengthen the relationship between management scores and annual labour costs per worker while simultaneously (ii) weaken the relationship between management scores and wage inequality. The latter conclusion follows from the estimated coefficients reported in columns (3)-(4) and (6)-(7), which show that, in firms with works councils, the relationship between management scores and wages is stronger for workers in the bottom half of the wage distribution (columns (6)-(7)), which in turn weakens the positive relationship between management scores and wage dispersion (columns (3)-(4)). It is also worth noting that the amplifying effect of works councils on the relationship between management scores and average pay only applies to the variable measuring annual labour costs per worker (column (1)). This might suggest that, in firms with works councils, a larger share of the gains from productivity-enhancing management practices are distributed to workers in the form of non-wage benefits (e.g., perks).

Overall, the results in Table 8 suggest that, in firms with works councils, a larger share of the gains from productivity-enhancing management practices is shared with workers, and the gains are more evenly distributed across the workforce. These results are also consistent with the hypothesis that the productivity-enhancing effects of management practices are larger in the presence of works councils (Jirjahn et al. 2024).

In Table 9 we reproduce the analysis behind the results shown in the previous table, but using instead an indicator of firm-level wage agreements as a proxy for collective bargaining power at firm level.²⁵

Table 9 Management practices, wage bargaining and pay

The results in Panel A of Table 9 show that the effects of firm-level wage bargaining are qualitatively similar to the effects of works councils shown in Table 8, namely higher average wages (but only for non-managers) and lower within-firm wage inequality. In Panel B, where we interact the firm-level wage agreement indicator with the management score, the signs of the estimated coefficients also follow much the same pattern as for the case of works councils, though the precision of the estimates is generally lower, with most of the coefficients being statistically insignificant. The lack of statistical significance might be caused by the fact that the number of firms with firm-level wage agreements is very low. Besides, more than half of these firms also have works councils, and it might be the case that the separate effect of wage-level bargaining is smaller for firms that already have worker representation in the form of works councils.²⁶

However, we still find a (weakly) significant negative effect on our two measures of wage inequality (columns (3)-(4)), suggesting that productivity gains generated by firms with a higher management score are more evenly distributed among the workforce in firms with local wage bargaining.

Robustness: Propensity Score Matching

An undeniable limitation of our study is the cross-sectional nature of our data, since we only have information about management practices in a single year (2016). This obviously makes it hard to claim causality for our findings. In a tentative step towards mitigating this problem, we perform a robustness check where we combine OLS regressions with propensity score matching, classifying firms as being treated if the management score is above some threshold level. The aim of this approach is to re-create an ex post random sample by pairing treated with control firms based on observed firm characteristics that determine the use of management practices. A key underlying assumption is the conditional independence assumption, which states that treatment assignment (sufficient use of management practices), conditional on firm attributes, is independent of the potential wage outcomes. If this assumption holds, we can (after controlling for selection into treatment) use firms that adopt less management practices to approximate the counterfactual wage outcomes of firms that adopt more managements practices. In other words, the key matching assumption is that selection occurs only on observables. Thus, by construction, matching eliminates two of the three sources of selection bias identified by Heckman et al. (1998), namely the bias resulting from having different ranges of the matching covariates for treated and control samples, and the bias resulting from having different distributions of these covariates across their common support. The remaining source of bias, differences in unobservables across groups, is ruled out by the matching assumptions.

Since our management score is a continuous variable and propensity score matching is based on a dichotomous categorisation of treated versus non-treated firms, we need to transform the continuous management score into a binary variable. We do this in two alternative ways, by defining two treatment indicator variables that are equal to one if the management score is larger than either the median or the 75th percentile of the distribution of management scores. Our treatment variable thus splits firms according to a high- or low-level use of management practices. We then estimate a logit regression to model the probability of a firm being treated (high-level use of practices) in our sample. In the selection equation we use a wide set of firm characteristics variables regarding age, size, ownership and governance (government-owned, family-owned, family-managed, presence of works councils), market concentration (HHI defined at the 5-digit level), global engagement (multinational or exporter), and human capital (education and skill-level of the workforce). Most of these variables have previously been shown to systematically relate to the use of management practices (Bloom and Van Reenen 2007, 2010). We also include industry indicators at the 2-digit level. Table A7 in the Appendix shows the results of the logit estimation using the two alternative management indicators. The variables that significantly determine selection into the treated group, regardless of choice of threshold for the management indicator, are firm size and workforce skills (which are associated with a high use of management practices) and public ownership (which is associated with a low use of management practices).

In the next step, we estimate the propensity score of each firm using the predicted probability from our logit model. We then match each treated firm to a control firm using the nearest neighbour matching technique (with replacement). Figure 4 in the Appendix illustrates how matching succeeds in reducing the variability between the two groups of firms for the two treatment indicators. Furthermore, column (2) and (4) in Table 18 show the estimates when we run the logit model on the matched sample in order to assess whether significant differences between control and treated firms persist.

The main results from this robustness check are shown in Tables 10 and 11.²⁷ In the former table we show the estimated relationships between the overall management score and our different measures of pay and pay inequality. These estimates serve as robustness checks of the results reported in Tables 1 and 2 in the main analysis. In order to ease the comparison with the main analysis, in Panel A we re-estimate the main results in Tables 1 and 2 on the full sample using (the two versions of) the dichotomous management practices variable. In qualitative terms, the estimates displayed in Panel A almost perfectly mirror the estimates shown in Tables 1 and 2, although the magnitudes of the estimates are naturally smaller when using the binary management score variable.

Table 10 Management practices and pay – results based on propensity score matching

Table 11 Categories of management practices and pay – results based on propensity score matching

In Panel B we show the corresponding OLS estimates using the matched sample. For the most part, these results are qualitatively similar to the estimates in Panel A, though some of the estimates lose statistical significance if we use the 50-percentile threshold instead of the 75-percentile threshold in the definition of the dependent variable. What persists in all estimations is a statistically significant relationship between management scores and average wages that is particularly strong for workers at the top end of the wage distribution. In Panel C we also present an alternative set of estimates based on the matched sample, but instead using the continuous version of the dependent variable, which makes these estimates directly comparable to the estimates from our main analysis (in Tables 1 and 2). Again, the overall qualitative picture is the same, and the magnitudes of the estimates are also fairly similar to the ones in the main analysis.

Finally, in Table 11 we show the estimation results from the matched sample that correspond to those in Table 3, where we look at the effects of different types of management practices (using a continuous management score variable for each of the three main categories of management practices). As in our main analysis, management practices related to incentives are clearly the most important ones in terms of explaining the relationship between management scores and wages. However, the results in Table 11 also show that monitoring plays a significant role, but only for a subcategory of workers. More specifically, these results suggest that management practices related to incentives are more important for workers in the top half of the wage distribution, whereas management practices related to monitoring are more important for workers in the bottom half of the distribution. These results are qualitatively similar for both definitions of the dependent variable.

Concluding Remarks

In this paper we have used rich Portuguese data to study the relationship between management practices and pay, focusing both on average earnings and within-firm pay dispersion. Our two main results are that a higher use of management practices is significantly associated with both (i) higher average pay and (ii) higher pay dispersion within the firm. Although the positive relationship between management practices and average wages applies throughout the wage distribution, even for workers in the bottom decile, the magnitude of this relationship is monotonically larger for workers higher up in the distribution, thus contributing to higher within-firm wage inequality.

Whereas our two main results are broadly in line with the existing literature, our detailed analysis produces a further set of new insights regarding the relationship between management practices and pay. Perhaps the main new insight from our paper is that, although the management score is positively correlated with average wages and wage inequality, these results are mainly driven by a very narrow subset of all the management practices that are included in the overall management score, namely management practices related to incentives. In this sense, our results are in line with some of the existing literature on performance pay, as discussed in the Introduction, but what we additionally show is that other forms of management practices seem to have little or no effects on wages and wage inequality. More specifically, our main results are to a large extent explained by the adoption of two different incentives-related management practices, namely rewarding high performance and promoting high performers. In an extension to our main analysis, where we produce regression estimates from a more restricted sample created by propensity score matching, we also show that management practices related to monitoring might have a positive impact on wages, but only for workers in the bottom half of the distribution.

Another insight we would like to emphasise is the potentially moderating effect of labour market institutions, in particular the presence of works councils. We find that the presence of works councils strengthens the relationship between management scores and annual labour costs per worker, while it weakens the relationship between management scores and wage inequality. Thus, this result can be seen as giving some support to the hypothesis that formal worker representation improves the functioning of incentives-based management practices (Jirjahn et al., 2024).

The main limitation of our study is the absence of more than one observation per firm for the management practices variable, which makes it hard to claim causality for the above described relationship between management practices and wages. Although our main results are largely robust to alternative approaches based on propensity score matching, a more solid confirmation of causality could potentially be established in a panel data analysis, which would require new waves of the management practices survey. This will hopefully be available in the future.

Appendix

Management Practices

Table 12 shows a description and categorisation of the 29 questions from the IPG survey that are used to construct the management scores that we use in our analysis.

Table 12 Information used from IPG to compute the management scores

Summary Statistics

Table 13 shows summary statistics for the variables used in our main analysis.

Table 13 Summary statistics

Sample Restrictions Based on Survey Response Rates

Table 14 shows regression estimates corresponding to Tables 1 and 2, whereas Table 15 shows regression estimates corresponding to Table 3, when we restrict the sample to firms that answered all the 29 questions in the IPG survey.

Table 14 Management practices and pay – using only firms that answer all questions

Table 15 Categories of management practices and pay – using only firms that answer all questions

Using outcome measures based on hourly wages

Table 16 shows regressions estimates corresponding to Tables 1 and 3 when measuring wages and wage inequality in terms of hourly wages.

Table 16 Management practices and hourly wages

Skill groups and occupational categories

Table 17 describes the four different skill groups according to the ISCO-88 classification and lists the occupational categories belonging to each group.

Table 17 Skill levels and occupational categories

Propensity score matching

Table 18 shows the estimates from logit regressions (before and after matching) explaining the adoption of high- versus low-level use of management practices, for each of the two management score thresholds.

Table 18 Logit regression estimates for high-level use of management practices

Figure 4 illustrates the propensity score matching quality in the form of covariate balance plots that show how propensity score matching reduces the variability between treated and untreated firms, for each of the two management score thresholds.

Fig. 4

Quality of matching on the propensity score