Employee representation and hierarchical depth

We begin by considering the following baseline regression model:

(1)$$\eqalign{Y_{ijc} = & \beta _0 + \beta _1{\rm E}{\rm R}_{ijc} + \beta _2{\rm Medium\ fir}{\rm m}_{ijc} + \beta _3{\rm Large\ fir}{\rm m}_{ijc} \cr & \quad + \beta _4{\rm E}{\rm R}_{ijc} \times {\rm Medium\ fir}{\rm m}_{ijc} + \beta _5{\rm E}{\rm R}_{ijc} \cr & \quad \times {\rm Large\ fir}{\rm m}_{ijc} + {\bf b}{\bf X}_{ijc} + \varepsilon _{ijc}} $$

where subscripts i, j and c denote the establishment, industry and country, respectively; Y _ijc. is the natural logarithm of the number of layers in 2013 for establishment i in industry j and located in country c; ER_ijc. is a dummy variable for the presence of ER at the establishment level; Medium firm_ijc and Large firm_ijc. are dummies for medium (50–249 employees) and large firms (250+ employees), the small firm category (10–49 employees) being the benchmark; ${\bf X}_{ijc}$,. is the vector of controls; $\varepsilon _{ijc}$ are the residuals.

Table 2 reports the results of a series of OLS estimates. In column 1, we estimate a parsimonious model in which we only include a dummy variable that takes value one for establishments in which there is an ER structure in place, three dummy variables to control for establishment size (small firms being the benchmark category) and interaction terms to capture the interplay between ER and plant size. Estimates reported in column 2 also control for industry and country fixed effects. The industry dummies capture differences in ER prevalence due to sector-specific industrial relations structures as well as cross-sectoral differences in technology, which in their turn may involve differences in the predictability of the production process. The country dummies absorb cross-country heterogeneity in labour institutions, including those regulating ER formation and functioning, and culture (e.g. trust) that may affect the structure of corporate hierarchies. The country dummies also control for differences in average income, judicial efficiency, development of financial markets and other macro-level institutional factors that may have an impact on the firm structure. The presence of ER is positively associated with the number of layers. As expected, larger establishments tend to have more layers. Interestingly, the interaction between ER and size is significantly negative, suggesting that the impact of ER on firm hierarchies tends to vanish as firm size increases.

Table 2. Depth of hierarchy: OLS estimates

Notes: Estimation by ordinary least squares with robust standard errors in parentheses. The dependent variable is the number of layers (in logs). Establishment-level controls: workforce composition (gender, age, education, fraction of part-time, permanent employees), multi-plant, change in employment, productivity, organizational structure and ownership in the last three years, subsidiary site, share of workers involved in complex tasks. Estimates reported in column 4 control for outsourcing of production activities. In column 5, we add controls for the use of information systems. In column 6, we add controls for manager's characteristics (gender, position, tenure). The small firm category (10-49 employees) is the benchmark category for size dummies. ***P < 0.01, **P < 0.05, *P < 0.1.

In columns 3–6, we sequentially add more controls to see the robustness of the results. In column 3, estimates control for establishment-level differences in workforce composition (gender, age, skills, fraction of part-time and permanent contracts) and share of workers performing complex tasks, reported change in productivity and employment since 2010, and dummy variables identifying multi-site firms, subsidiary sites, recent changes in ownership and organizational changes. In column 4, we also account for differences in the prevalence of outsourcing of production activities that may also affect the hierarchical structure of firms. In column 5, we additionally control for the use of information systems oriented to minimize supplies or work-in-process (e.g. just-in-time, lean production systems). Finally, in column 6, we add a series of ‘noise controls’ on respondents' characteristics (gender, position and job tenure of the manager) in order to increase the precision of our estimates and reduce concerns about measurement error in the organizational variables. None of the described modifications alters the basic finding.Footnote ⁷

The estimated positive effect of ER on the number of layers can be interpreted as compatible with the fact that on average the reduction in communication costs associated with the presence of ER is sufficiently large to offset any reduction in knowledge acquisition costs. This would also indicate that the net effect of ER on communication costs is negative, i.e. the skip-level reporting effect dominates the cost of delayed decisions resulting from the operation of ER (e.g. information and consultation process). The fact that the effect of ER is heterogeneous across establishment size categories suggests the trade-off between communication costs and information acquisition costs may be size-contingent. This evidence is consistent with the idea that larger organizations are less flexible and are less able to accommodate the layer increasing effects of ER, thereby leading to the negative coefficients of interacted terms observed above.

To give a sense of scale of the impact of ER on the hierarchical depth, we also estimate an ordered probit model which makes the interpretation of the magnitude of the coefficients easier. We recoded the dependent variable into five categories, with the first four categories corresponding to the reported number of layers and the last category including all establishments where the number of layers is five or more. Then we calculated the marginal effects for each value of the dependent variable. The marginal effects of interest are reported in Table 3. We obtain that the probability that establishments have five or more layers is 3.2 p.p. higher in establishments with ER compared to those without ER. This is equivalent to roughly a fifth of the difference between small and medium-sized establishments in the same probability.

Table 3. Depth of hierarchy: Marginal effects from ordered probit model estimates

Notes: Marginal effects from ordered probit estimation. The full model is considered, including country and industry dummies, establishment-level controls, and controls for outsourcing, information systems and manager's characteristics. Log likelihood null model: −26,849.783. Log likelihood full model: −23,405.669. χ² test: 6,513.07. Reported values are the differences between the probabilities that result when the dummy variable under scrutiny takes its five different values. The small firm category (10–49 employees) is the benchmark category for size dummies. ***P < 0.01, **P < 0.05, *P < 0.1.

Endogeneity

Tables 2 and 3 present conditional correlations that are broadly consistent with our theoretical intuition. Nevertheless, we are concerned about the potential endogeneity bias of our estimates, as due for example to unobservable variables correlated with the organizational outcome and our measure of ER.

To deal with this we consider an IV strategy. We ground the identification of a viable instrument on the analysis of the institutional determinants of unionization. According to standard cost–benefit analysis of union determination (e.g. Berkowitz, Reference Berkowitz1954; Hirsch and Addison, Reference Hirsch and Addison1986; Pencavel, Reference Pencavel1971), ER can be modelled as though it were an asset available to utility-maximizing workers that provides a flow of services. In this framework, the costs and benefits of unionization (as well as the propensity and the opportunity to organize) are affected by institutional variables such as the centralization of collective bargaining (Schnabel, Reference Schnabel, Addison and Schnabel2003). Sectoral or regional coverage of collective agreements influences, in particular, the collective action costs needed to establish ER and the benefits that the workers can obtain from it. For instance, when employment conditions are determined by a collective agreement, workers have incentive to be active proponents of these conditions through union action. Setting up an ER structure at the workplace level may also require expert knowledge and operational support which is more likely to be available when there are higher level union confederations involved in collective bargaining (Devicienti et al., Reference Devicienti, Naticchioni and Ricci2018). Depending on the labour legislation, it is also possible that sectoral collective agreements cannot be extended to workers if it is absent at the firm an ER body acts as a signatory party of the agreement.

Previous empirical research has showed that the coverage by centralized collective agreements is an important determinant of the degree to which unions can successfully pursue an individual service or insurance strategy. In line with this, Scheuer (Reference Scheuer2011) finds that coverage by a collective agreement actually triples the likelihood of union membership. Moreover, comparative legal analysis clarifies that extension of collective agreements to third parties at the sectoral or regional level is mostly subject to regulatory institutions and labour laws that are clearly exogenous in our study (Adams et al., Reference Adams, Bishop and Deakin2016). On the other hand, however, even if the literature does not indicate explicitly that collective agreements coverage should exert a direct effect on hierarchical depth, there may be instances in which the exclusion restriction is violated in this empirical context, for example, when collective contracts discipline minimum pay thereby inducing the firm to introduce certain internal labour reorganizations regardless of ER presence. Keeping this limitation in mind, we use information on whether the firm is covered by a collective wage agreement negotiated at the sectoral or regional level as an exogenous factor that shifts the probability of establishing an ER at the establishment level. We build a dummy variable (sectoral bargaining,- i.) coded 1 if the firm is covered by a sectoral wage agreement and 0 otherwise, and use it as an instrument for ER in equation (1). Also the interaction terms capturing ER effects at different firm size classes are instrumented. The results are collected in Table 4. The first-stage results show that coverage by sectoral or regional agreements is a strong predictor of ER presence at the firm level. Moreover, in the second stage, we find again that ER positively correlates with the number of hierarchical layers and that the positive firm size–layers relationship is tempered by the presence of ER. The interaction term is significantly negative for both large and medium-sized establishments, suggesting that the positive association between ER and depth of the hierarchy holds especially for small firms. This intuition is confirmed by running the same empirical exercise while splitting the sample across establishment size classes. Reassuringly, usual IV diagnostic tests for instrument relevance and exogeneity are passed.Footnote ⁸

Table 4. Depth of hierarchy: IV results

Notes: Estimation by 2-stage least squares with robust standard errors in parentheses. The dependent variable is the number of layers (in logs). The instrumental variable is a dummy coded 1 if the wages are determined at the sectoral level through collective bargaining agreements. We report the 1st-stage coefficient associated with instrumenting ER. Also the interaction terms are instrumented (but the coefficients are not reported for reasons of space). Establishment-level controls are the same as those in Table 2. Estimates reported in column 3 control for outsourcing of production activities. In column 4, we add controls for the use of information systems. In column 5, we add controls for manager's characteristics (gender, position, tenure). The small firm category (10–49 employees) is the benchmark category for size dummies. ***P < 0.01, **P < 0.05, *P < 0.1.

Additional results

In this section, we enrich our basic findings by analysing the relationship between training activities, coordination of work and staff meetings on the one side and ER on the other. We want to see if the sign of the correlations among such variables is consistent with our theoretical framework.

First, we use training as a proxy of firm-specific activities related to the acquisition and accumulation of non-codifiable knowledge (Garicano and Wu, Reference Garicano and Wu2012). The survey contains extensive information on training activities at the workplace level, including the fraction of employees entitled to paid time off for training and information on the purpose of training activities. Results are reported in columns 1–2 of Table 5.Footnote ⁹ The presence of ER positively correlates with our measures of training. Interestingly, ER is positively associated with the extent of training activities oriented not only to develop workers' skills at the current jobs but also with those aimed at enabling workers to take different job positions and rotate tasks with colleagues. This suggests that employees in ER establishments are able to deal with and solve a wider variety of production problems.

Table 5. Additional results: training, coordination and meetings

Notes: In column 1, we report estimates from ordered probit models with robust standard errors in parentheses, where the dependent variable (see definition in footnote 9) is the fraction of workers entitled to paid time off for training. In columns 2–6, we report estimates from linear probability models with robust standard errors. In column 7, we report estimates OLS estimates with robust standard errors. Control variables are the same as those in column 6 of Table 2. ***P < 0.01, **P < 0.05, *P < 0.1.

Hence, the effect of ER (if any) would be to reduce knowledge acquisition costs.Footnote ¹⁰ Within our theoretical framework, this ‘skilling’ effect of ER should contribute to push down the number of hierarchical layers. Therefore, our finding that ER is associated with a higher number of vertical layers should be driven by an even stronger reduction in communication costs in establishments under ER. To corroborate such insight, in columns 3–5 of Table 5, we show that the presence of ER at the establishment level is associated with reported changes in coordination and allocation of work to employees and meetings between managers and employees. It is worth noting that the positive correlation between ER and the frequency of staff meetings is an empirical fact that conflates both delay effects and skip-level reporting. Interestingly, according to estimates reported in column 6 of Table 4, managers in establishments with ER are not more likely than their counterparts in establishments without ER to agree with the statement that ‘employee involvement causes delays in the implementation of decisions’. Our interpretation is that arguably skip-level reporting dominates so the net effect of ER on communication costs is negative. This is consistent with the results reported in column 7, where we show that the magnitude of the effect of trade unions on the depth of hierarchy is higher with respect to other types of ER (including works councils). The latter result may reflect the fact that unions are typically more structured forms of ER using consolidated communication channels with the top management, whilst works councils and other forms of shop-floor ER tend to represent workers' voice through more informal bottom-up processes involving relatively higher communication costs.