Employee collective voice and short-term sickness absence in Europe

Abstract

This article analyses the relationship between employee collective voice, measured by union density and institutionalized forms of employee representation at enterprise level, and short-term sickness absence rates in 24 European countries over the period 1996–2010. It relies on individual-level data on sickness absence from the European Labour Force Survey combined with country-level data on employee collective voice. There is a small but significant and non-trivial, negative relationship between employee collective voice and short-term sickness absence. Regression analysis suggests that if union density had remained at the 1996 level, short-term sickness absence would have been, on average, 2.5 hours lower per year than in 2010.

Keywords

Collective voice comparative employee representation Europe exit-voice sickness absence unions

Introduction

Sickness absence involves considerable costs to individuals, companies and society at large. This study highlights how institutionalized and collective channels for employees to voice their opinion about workplace matters can help explain differences in sickness absence levels and trends across countries. I first discuss the theoretical relationship between employee collective voice and sickness absence and, second, empirically test the relevance of this exit-voice perspective for understanding trends in short-term sickness absence in 24 European countries over the period 1996–2010. To my knowledge, this is the first study that has analysed sickness absence from an exit-voice perspective within a comparative framework.

A number of theoretical perspectives have been proposed for understanding differences in sickness absence levels and trends across countries (Osterkamp and Röhn, 2007). Explanations rooted in neo-classical economic theory have emphasized the economic incentives to which employees are exposed. A generous benefit system will, according to this perspective, result in higher sickness absence since individuals do not bear the full cost of reporting sick. The role of sickness compensation systems has also been linked to so-called absence cultures, where sickness absence is viewed as a form of group behaviour with shared beliefs and practices within an organization or indeed a country (Chadwick-Jones et al., 1982).

In contrast, epidemiological research has tended to view sickness absence primarily as a manifestation of the health status of individuals rather than as a result of calculative decision making. Such research has largely focused on analysing how different risk factors (at the individual, organizational or population level) influence sickness absence, and studies have shown a strong association between morbidity and sickness absence (Kivimäki et al., 2003). A related line of research has viewed sickness absence as a result of unfavourable working conditions. In the well-known demand-control model (Karasek and Theorell, 1990), the psycho-social work environment is seen as crucially determined by two dimensions: job demands (such as workload) and job control (the employee’s control over decision making relevant to his or her work and the use and development of skills). Stress, possibly resulting in sickness absence, is seen as a result of a combination of high demand and low control, or what are termed high-strain jobs.

The state of the economy, and above all unemployment, has also been identified as crucial for understanding trends in sickness absence, as high unemployment rates might discipline employees to be absent from work less often (Osterkamp and Röhn, 2007). This hypothesis hinges critically on the assumption that sickness absence increases, or is seen by employees to increase, the probability of job loss. Moreover, if employees with high absenteeism lose their jobs as part of cutbacks in staff, the remaining workforce will consist of employees who will be less likely to take sickness absence. Unemployed workers with a record of absenteeism might also find it more difficult to find new jobs, thus exacerbating the effect of unemployment on sickness absence. However, low unemployment is often the result of high economic growth, which can raise the demands on employees, thereby elevate the risk of sickness absence.

Finally, theories originating from psychology tend to view sickness absence as a form of withdrawal behaviour and a consequence of low work satisfaction. Steers and Rhodes (1978) argue, however, that low work satisfaction in itself is a poor predictor of absence. They argue that employee attendance (or its obverse, absence) is a function of both the ability to attend and the motivation to do so, which in turn, to an important extent, is determined by job satisfaction. Whether or not low attendance motivation manifests in absence is contingent upon the employee’s ability to attend, as well as the pressure to attend placed on her by a number of factors both inside and outside the workplace.

This short review highlights that multiple factors, ranging from proximate and concrete features of the working environment to more abstract changes in norms and attitudes, might be important for understanding why levels of sickness absence differ across countries. Below, I outline an exit-voice model that relates to both the demand-control perspective and approaches that view sickness absence as a form of withdrawal behaviour. At the heart of this model are the opportunities for employees to voice their opinions and influence both the immediate work environment and more strategic issues relating to organizational goals and production processes. For this form of control to be effective, however, it must be collective and institutional, which implies that the strength of trade unions and employee representation at the enterprise level are central to understanding differences and trends in sickness absence across countries.

Sickness absence from an exit-voice perspective

Freeman and Medoff (1984), using the framework developed by Hirschman (1970), argued that workers have two basic ways of dealing with workplace problems: exit, ranging from temporary absence to permanently quitting a job, and voice, through speaking up to management. Both options, however, come with costs. The costs associated with exit are, for example, the economic penalties associated with absence (or unemployment) as well as forgone opportunities for wage and career development. The costs associated with voice are also the risks of worse wages and career development and, at the extreme, of being fired if employees are penalized for voicing their discontent. Moreover, many benefits from negotiations between individual workers and employers, such as improvements in working conditions, are public goods, which mean that individuals who do not voice their opinion cannot be effectively excluded from benefiting from these improvements. This may give rise to the well-known free-rider problem. Effective voice, therefore, has to be collective and institutionalized, since only then will workers have an incentive to express their opinion and preferences, and have the assurance that they will not suffer for it. When employees have a collective or union voice to express their discomfort to managers and employers, they are offered an important and tangible alternative to temporary absence or quitting.

An important consequence of employee collective voice is to change employers’ attention from the marginal to the average worker (Freeman and Medoff, 1984). With individual bargaining, the employer may be primarily concerned to retain the worker who is on the verge of leaving the company. This marginal worker is likely to be young, single and mobile. In contrast, the preferences of the average worker, who is more likely to be older, have a family and, therefore, be relatively immobile, may be taken for granted or even ignored. Working conditions that appeal to the average and to the marginal worker are likely to differ, and moving from individual to collective bargaining therefore implies, through a median voter process, that working conditions will come closer to those desired by an average worker or the majority of the workforce (Bennett and Kaufman, 2004).

This exit-voice perspective leads to the hypothesis that stronger employee collective voice, in the form of unions or institutionalized representation at enterprise level, will lead to lower sickness absence, net of the effect of other factors influencing sickness absence. The mechanism translating employee collective voice into lower sickness absence is that it provides workers with a collective and institutionalized means to address workplace problems. This theoretical framework relates to approaches linking control in the workplace to mental and physical illness, and to accidents at work (Spector, 2009). Most studies have, however, defined control in the workplace to denote employees’ control over specific tasks: Control has been measured by workers’ subjective assessment of their control over a specific task, a conceptualization that is open to subjectivity bias. In the original formulation of the demand-control model, Karasek and Theorell (1990) emphasized the role of workplace democracy and workers’ involvement in initiating, implementing and changing existing organizational structures for the health of employees. As argued by Theorell (2004), ‘the individual’s possibility to exert control over his or her own situation is of fundamental importance to health’ (pp. 323–324).

A number of important qualifications must, however, be made to this approach. First, sickness absence is, to an important extent, determined by actual illness, and it is therefore essential to recognize that there are a range of medical conditions that effectively rule out work attendance, no matter the extent of workplace control (however defined). Second, temporary withdrawal from work might also be viewed as a form of voice behaviour, a signal to the employer that something is wrong at the workplace. However, the existence of collective voice would still reduce aggregate absence through its effect on absence as exit behaviour (Mastekaasa, 2013). Moreover, it could be argued that absence as voice is an inefficient signal to employers, since they will not know the exact reasons for the absence (Freeman, 1976). Third, there are other channels, besides unions, that employees may use as voice mechanisms, such as autonomous work groups. Concerns have therefore been raised about the limitations of using union membership as a proxy for collective voice. Fourth, an important implicit assumption of the exit-voice model is that collective voice provides an effective alternative to exit by safeguarding employees from reprisals from management (Luchak and Gellatly, 1996). However, there are often issues over which management retains decision-making power even in the case of strong employee collective voice, such as evaluation of employee performance or assessment of ability in matters of promotion.

Studies analysing exit behaviour, including sickness absence, within an exit-voice framework have used individual-level data from one country and based their conclusions on observed differences between union and non-union members. The general conclusion is that union membership appears to be associated with lower quit rates and increased job tenure but positively associated with sickness absence. Studies analysing the association between union membership and sickness absence have tended to focus on more long-term absence. Leigh (1986) found that a union contract is positively related to absence, defined as annual hours of absence that respondents attribute to illness, and according to Tompa et al. (2010), union members in Canada have a higher probability of spells of sickness absence of at least 1 week. Leigh (1984) found that the effect of unionization is to increase absenteeism among blue- but not among white-collar workers. These results are in line with Mastekaasa (2013), who also found a stronger effect of union membership for workers in lower-grade jobs. In terms of the definition of the dependent variable (sickness absence), Veliziotis (2010) comes closest to this study. Defining sickness absence in the United Kingdom as hours lost through illness, he finds that union members have higher weekly expected absence.

This study, focusing on the macro-level and comparing countries over time, offers an important complement to these individual-level studies. At the heart of the employee collective voice framework is the proposition that individual bargaining is replaced by (more efficient) collective bargaining. However, many of the benefits generated by such employee collective voice, such as better working conditions, are properly viewed as collective goods. This also implies that channels and structures for interest representation in themselves are collective goods from which both members and non-members can benefit (Traxler, 2000). This suggests that evaluation of the relevance of the exit-voice perspective for sickness absence should focus on institutionalized forms of employee representation in the economy rather than on individual membership in unions. Important evidence in this context is provided by Furåker and Bengtsson (2013), who found that it is unions’ collective power, rather than individual membership, that matters for the occurrence of regular workplace meetings (arguably an important voice mechanism) and for the impact of these meetings on organizational decisions. The focus on individual membership in unions also overlooks the fact that other forms of employee collective voice might provide representation at workplace level and in the economy. Here, Visser (2009) has made a useful distinction between two ideal-typical forms of employee representation: the single channel model, where workplace representation is based on election by and/or appointment from union members, and the dual channel model, where employee representation is formally independent from, but often exists in addition to, the union. Perhaps, the best known form of such employee representation is work councils, which are usually elected by and from all employees, are held accountable to union and non-union members alike and operate within powers and competences defined by law.

Data, operationalization and method

I use data from the European Union Labour Force Surveys (EU-LFS) for the period 1996–2010 for 24 countries; these provide harmonized cross-sectional information on individuals compiled from National Labour Force Surveys, and the comparability of data is high (Mazzuco and Suhrcke, 2010).

Respondents were asked their usual (or contracted) hours ( $C_{i}$ ) and the actual hours worked in the reference week ( $A_{i}$ ); if there was a difference, they were asked the reason. If the stated reason was own illness, injury or disability, the respondent is considered absent through illness during the reference week. Following Barmby et al. (2002), short-term sickness absence $S_{i}$ for an individual (i) is defined as the difference between usual hours and fewer hours actually worked because of sickness (equation (1))

S_{i} = r_{i} (C_{i} - A_{i})

(1)

where $r_{i} = 1$ if absence is due to illness during the reference week, and $r_{i} = 0$ otherwise. Since the focus is on short-term absence, respondents who were absent the whole reference week are excluded. Only respondents aged 20–54 years are included in the analyses, and self-employed are excluded.

Several limitations of this measure should be noted. First, there is no information on the severity of the health problems. Second, if usual and actual hours worked differed because of sickness in the reference week, we have no way of knowing whether this week marked the beginning or end of a longer-term sickness absence. Third, to some extent, this measure may include respondents who combine reduced working hours with sick leave.

Since the EU-LFS consists of individual-level surveys for each country and year, the data can be represented as a hierarchical, three-level structure. Two alternatives exist for analysing this form of data structure (Lewis and Linzer, 2005): First, simultaneous modelling of individual outcomes as a function of both individual-level and country-level characteristics, which means estimating a three-level hierarchical model. Such models are very complex, and in this study, the large number of respondents (over 7 million) in a nested data structure meant that hierarchical models often did not converge. The second option is to employ a two-step approach, in which individual-level parameters are estimated separately for each country and year and then, in a second stage, used as dependent variables in macro-level regression where the individual-level parameters are related to country-level characteristics that vary over time. The two-step approach provides a very flexible specification because all individual-level effects are allowed to vary across countries and time without imposing any further distributional assumptions. This approach is equivalent to estimating a model with both random intercepts and random coefficients for all independent variables, which is especially attractive in the present analysis since our interest lies in the second-stage (country-level) parameters.

In the two-step approach employed in this article, a separate ordinary-least-squares (OLS) regression for each country (j) and year (t) was first estimated (equation 2). For notational simplicity, we assume only one covariate at level 1 ( $X_{1}$ ) and level 2 ( $X_{2}$ )

S_{i j t} = β_{0 t j} + β_{1 t j} X_{1 i t j} + v_{i t j}

(2)

In the second step, the predicted average values for each country (j) and year (t) obtained from equation 2, where the individual-level covariates set at their average values were pooled and used as dependent variables in a pooled time-series cross-section analysis (equation (3))

{\hat{S}}_{j t} = α + υ_{j} + β_{2} X_{2 j t} + μ_{j t}

(3)

Given that sickness absence, to an important extent, is related to unobserved differences between countries, it is necessary to include unit-fixed effects ( $α + υ_{j}$ , with $υ_{j}$ estimated by N-1 country dummies) to account for this unobserved heterogeneity in the second step. An important issue in this approach is that the dependent variable in the second stage is estimated from the first stage, and we must, therefore, account for the uncertainty of these estimates. Different weighting procedures have been proposed, but Lewis and Linzer (2005) have shown that OLS with robust standard errors is the best approach. When estimating the fixed-effect pooled time-series regression, robust standard errors mean that we have to take into account the fact that errors may be heteroscedastic (different variances across units) and autocorrelated (not independent from one time period to another).

Fisher-type unit root test (Choi, 2001) indicates that although both short-term sickness absence and union density (as well as temporary employment and gross domestic product (GDP) growth per capita) are stationary in levels, we cannot reject the null hypothesis (or the test is not clear-cut) that all panels contain unit roots for the other independent variables. Models using the first difference form of all variables (where the unit root test indicates that all variables are stationary) is therefore also estimated.

The EU-LFS does not record individual trade union membership. Employee collective voice is therefore operationalized and measured along two different dimensions. The first is union density, defined as net union membership as a proportion of the wage-earning population in employment (Visser, 2013). The second focuses on organizational strength and is based on a composite measure developed by Visser (2011: 43). This indicates (a) whether a provision for information and consultation in the workplace exists (score between 0 and 2), (b) whether it possesses strong or weak powers delegated to it by the unions, or acts independently of the unions (0–4), (c) whether it has strong or weak rights of intervention against management over a narrow or wide range of issues (0–3), and (d) whether it is directly involved in negotiations over pay, working hours and conditions of work (0–4). Since dimensions b, c and d require the existence of provision for information and consultation in the workplace (dimension (a) must take value 1 or 2), an index of the following form was constructed: employment representation = a * (b + c + d).

The multivariate models control for a number of covariates at the country level when testing the relationship between employee collective voice and sickness absence (unless otherwise stated, these country-level variables have been retrieved from Eurostat, 2015). Unemployment levels will capture the disciplinary effect of unemployment on sickness absence. The generosity of sickness benefit schemes is taken from the welfare state entitlement dataset (Scruggs et al., 2014) and calculated as the sum of two central characteristics of sickness benefit schemes: (1) average replacement rate (percent of previous wage) and (2) coverage (the share of the labour force insured for benefits). Employment protection (OECD, 2014) is also included as a control variable in the multivariate models. GDP growth per capita (purchasing power parity (ppp)-adjusted) and average hours worked by full-time employees is included to account for the fact that increased workloads and demands on employees in conjunction with economic booms might lead to increased sickness absence. Following Osterkamp and Röhn (2007), structural differences in employment and production are controlled for by including the share of employment in goods-producing industries and the share with only primary education in the labour force. Although it could be argued that low-educated employees and employees in the manufacturing sector on average have worse working conditions, many preventive policies in the area of occupational safety and health have in fact targeted ‘high-risk’ sectors and workers (Belin et al., 2011). To control for health-related selection into and on the labour market, we control for temporary employment (percentage of the total number of employees). The models also control for female and older-age (45 years and above) labour market participation rates. Finally, working conditions may affect the relationship between employee collective voice and absence through at least two channels. First, unions may provide better working conditions in general and make work presence more attractive, and second, hazardous working conditions resulting in the absence from work may trigger collective awareness and action, leading to increased unionization.

The EU-LFS contains no comparable measures of working conditions over the period studied here.¹ The strategy I adopt is, first, to include a measure of fatal accidents at work (standardized by the size of the labour force) as a measure of working conditions at the country level in the regression models. Though a crude measure of working conditions, it is the only comparable measure available for all 24 countries over the time period studied here. Second, in the regression models, I test for the possibility of a curvilinear relationship between employee collective voice and sickness absence. This specification tests whether the relationship between unionization (and employee representation) and short-term absence depends on the level of unionization. Intuitively, it seems more likely that high absence would be associated with increased unionization (suggesting a positive relationship between absence and unionization) at low levels of unionization, or when there the channels for employee collective voice are relatively weak or underdeveloped.

A number of individual-level variables that have previously been shown to be of importance for both ill-health and sickness absence (Beemsterboer et al., 2009) are also included in the multivariate models at level 1 (i.e. equation 2). Only individual-level variables that are available for all countries are included. In many studies, age has been shown to be related to work ability, with older workers not only absent less often but also having longer sickness absence spells than younger workers. Women have generally been found to have a higher likelihood of reporting sick, and the gendered division of labour might also mean that there is a higher likelihood of women reporting sick when, in reality, they are taking care of sick dependents. For this reason, the models also control for the family situation of the respondent (single, married and divorced). A measure of work contract, differentiating between respondents with temporary and permanent contracts, is also included in the multivariate model since it can be assumed that workers with temporary contracts enjoy less job security and larger penalties for reporting sick. Previous research has also found education to be an important determinant of sickness absence, with more highly educated individuals being less prone to absence, possibly reflecting not only better health status overall but also greater control over working conditions as compared to individuals with lower levels of education. Industrial sector (based on the nomenclature of economic activities (NACE) classification) is also included in all models.

Sickness absence and employee collective voice in Europe

The sample of countries studied here can fruitfully be divided into three groups in terms of average predicted short-term sickness absence according to equation (2) (Table 1). The first group (Estonia, Ireland, Greece, Hungary, Latvia, Lithuania, Poland, Portugal, Spain and Slovakia) displays low sickness absence, especially in the period 2006–2010. Some of these countries also show a decreasing trend in sickness absence. Many continental-European countries display medium levels of sickness absence, but some of these have shown rather substantial increases in sickness absence, such as France and Slovenia, and to a somewhat lesser extent also Italy, and also Germany, which was not represented in the EU-LFS until 2002. Sickness absence in the Czech Republic appears to have decreased somewhat, whereas no clear trends can be observed in Belgium, Austria and Switzerland. The highest levels of sickness absence are found primarily in northern Europe, and several of these countries also appear to have experienced a clear upward trend in sickness absence. Sickness absence in the Netherlands appears to have fluctuated at relatively high levels, whereas sickness absence in the United Kingdom displays a clear downward trend over the whole period 1996–2010.

Table 1.

Sickness absence, union density and employee representation, 1996–2010.^a

	Sickness absence			Union density			Employee representation
	1996–2000	2001–2005	2006–2010	1996–2000	2001–2005	2006–2010	1996–2000	2001–2005	2006–2010
GR	0.017	0.007	0.007	28.2	25.1	24.6	5.0	5.0	5.0
LT	0.027	0.004	0.006		15.7	9.4	0.0	1.4	7.0
PL		0.014	0.024	19.8	17.8	15.2	3.0	3.0	8.0
LV	0.036	0.0151	0.024		19.8	16.2	0.0	0.0	5.0
PT	0.045	0.032	0.030	23.6	21.4	20.3	8.0	8.0	8.0
ES	0.027	0.041	0.045	16.2	15.5	15.0	16.0	16.0	16.0
IE	0.048	0.050	0.053	41.6	35.4	33.5	0.0	0.0	3.0
HU	0.088	0.051	0.045	30.3	17.9	16.9	12.4	10.8	10.0
SK	0.075	0.071	0.038	38.5	26.1	18.1	4.8	12.0	12.0
EE	0.104	0.052	0.041	20.0	11.8	7.7	1.2	3.0	10.2
IT	0.073	0.086	0.091	35.9	33.9	34.1	18.0	18.0	18.0
CZ	0.092	0.092	0.077	33.1	21.8	17.8	0.0	0.0	8.0
BE	0.094	0.126	0.108	53.2	51.7	52.2	12.0	12.0	12.0
FR	0.110	0.138	0.192	8.2	7.9	7.7	12.0	13.6	16.0
SI	0.111	0.146	0.188	43.5	41.6	27.6	12.0	12.0	12.0
DE		0.068	0.249	26.1	22.8	19.4	16.0	16.0	16.0
AT	0.164	0.151	0.168	38.3	34.5	29.4	14.0	14.0	14.0
CH	0.267	0.302	0.262	21.3	19.7	17.9	5.0	5.0	5.0
DK	0.258	0.285	0.379	75.5	72.6	68.7	14.0	14.0	14.0
UK	0.596	0.465	0.318	31.3	28.6	27.3	3.0	4.0	8.0
FI	0.412	0.480	0.657	77.8	73.3	69.3	14.0	14.0	14.0
NL	0.662	0.544	0.666	23.8	21.5	19.9	12.0	12.0	12.0
NO	0.627	0.550	0.929	55.3	54.7	54.2	14.0	14.0	14.0
SE	0.754	1.068	1.033	82.2	76.6	70.3	16.0	16.0	16.0

Countries ranked (in ascending order) by average sickness absence.

Union density and employee representation in firms, together with bargaining coordination and the regular involvement of social partners in consultation over social and economic policies, are often considered the institutional pillars of the industrial relations systems in Europe (Visser 2009, 2011). Based on these characteristics, Visser (2009) has proposed a classification of industrial relations systems that also captures essential characteristics of employee collective voice. Organized corporatism, found in the Nordic countries, involves encompassing and centralized trade unions with high union density and collective bargaining coverage. The unions are normally the main representatives of employees in the workplace, although other, parallel channels of employee representation may also exist. The four Nordic countries display the highest levels of union density, but the rates have declined markedly in Denmark, Finland and Sweden. The Nordic countries also rank high according to our measure of employee representation.

The social partnership model, primarily found in continental (western) Europe, displays not only intermediate levels of union density but also relatively high levels of collective bargaining coverage. In this model, works councils offer a second channel for worker representation and consultation. Sometimes, these work councils function as a workplace union, but at other times non-union councils may offer a cheaper alternative to workplace representation than union membership (Ebbinghaus and Visser, 1999). This is reflected in high scores found on the employee representation index. Most countries in this model such as Austria, the Netherlands, Germany and Slovenia have, however, experienced a more or less continuous decline in union density rates. The highest union density rate in this model is found in Belgium.

In the liberal model, in this study represented by the United Kingdom and Ireland, bargaining is decentralized to the company level and bargaining coverage is low. Employees are primarily represented by unions at workplace level, but this form of employee representation is less protected under the law. Both Ireland and the United Kingdom score very low on the employee representation index. Trade union density is at an intermediate level.

In the state-centred regimes found in southern Europe, there are highly institutionalized forms of employee representation (Ebbinghaus and Visser, 1999). Whereas all these countries score high on the employee representation index, union density varies greatly in this model but has been relatively stable over the period studied here in France, Spain and Italy while declining in Portugal and Greece.

Finally, the transition economies of central and eastern Europe display strong statist features, and both collective bargaining and employee representation are fragmented. The enforcement of employee rights is generally weak (Kohl and Platzer, 2007), as are institutionalized forms of employee representation. Many of these countries have experienced drastic decreases in union density.

Results

Table 2, column 2 summarizes the results of a series of regression models where the country-level variables are entered into the models one at a time. The regression coefficients of both union density and employee representation are negative and significant (at the 1 and 5% level, respectively), thus indicating support for the hypothesis that employee collective voice is associated with lower short-term sickness absence. The regression coefficient associated with the unemployment rate is negative and highly significant, providing support for the argument that there is a disciplinary effect of unemployment on sickness absence. Also, the coefficient of employment protection legislation is negative and significant (p < 0.1), suggesting that stricter protection against dismissal decreases the likelihood of reporting sick. The coefficient associated with the share with primary-only education in the labour force is positive and significant (p < 0.01), as is the coefficient associated with the share of employees with temporary employment contracts (p < 0.01). Fatal accidents are associated with lower short-term absence (β = −0.872, p < 0.01), possibly indicating that more severe accidents prompt occupational health and safety arrangements that lower short-term absence.

Table 2.

Associations between employee collective voice and control variables and average predicted short-term sickness absence. Pooled cross-section time series models.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Union density (*100)	−0.383*** (0.120)	−0.708*** (0.233)	−0.565*** (0.169)	−1.080*** (0.375)	−0.881*** (0.271)	−0.919** (0.411)	−0.548*** (0.157)
Union densitysquared (*100)							−0.046** (0.022)
Employeerepresentation (*100)	−0.288** (0.147)	−0.567*** (0.197)	−0.628*** (0.204)			−1.340** (0.617)	−0.672*** (0.197)
Unemployment (*100)	−0.468*** (0.158)	−0.862*** (0.281)	−0.532*** (0.202)	−0.511 (0.338)		−0.650 (0.396)	−0.554*** (0.198)
Industrial employment share	−0.411 (0.253)	−0.845*** (0.326)	−0.475 (0.297)	−0.272 (0.365)		−0.368 (0.430)	−0.473 (0.297)
Share with primary-only education	0.375*** (0.133)	1.037*** (0.242)	0.960*** (0.193)	1.785*** (0.565)	1.631*** (0.523)	0.564*** (0.236)	1.009*** (0.198)
Fatal accidents	−0.872*** (0.323)	−2.086*** (0.697)	−2.091*** (0.533)	−1.716* (0.888)	−2.187*** (0.573)	−1.824** (0.883)	−2.256*** (0.530)
Temporaryemployment	0.796*** (0.186)	0.667*** (0.192)	0.749*** (0.194)	1.428*** (0.548)	1.509*** (0.366)	0.778*** (0.276)	0.748*** (0.190)
Average hours worked (*100)	0.098 (0.486)	0.012 (0.008)		3.110** (1.270)	0.343 (0.766)
Proportion women in labour force	−0.459 (0.773)	−0.157 (1.075)		3.962*(2.256)	2.769 (1.899)
Share working part-time	0.438 (0.312)	0.436 (0.436)		1.562** (0.665)	0.594 (0.656)
Employment protection	−0.051* (0.028)	−0.012 (0.061)		−0.066 (0.091)
GDP per capita growth (*100)	0.006 (0.120)	−0.134 (0.165)		−0.070 (0.179)
Sickness benefit generosity	−0.040 (0.049)	0.096 (0.073)		−0.034 (0.145)
Share elderly in labour force	−0.178 (0.297)	−0.746 (0.612)		0.254 (0.844)
Constant		0.285 (0.760)	0.540*** (0.127)	−0.016 (0.014)	−0.009 (0.015)	0.791*** (0.280)	0.361*** (0.120)
Country-level observations	283–328	258	297	218	265	207	297
Number of countries	24	24	24	24	24	14	24
Individual-level observations		6,372,451	7,250,435	5,425,235	6,598,355	2,031,074	7,250,435
R²		0.868	0.821	0.240	0.170	0.857	0.825
Unit-specific effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Period-specific effects	No	No	No	No	No	Yes	No
Differenced data	No	No	No	Yes	Yes	No	No
Hausman test(p)		51.46 (0.000)	42.64 (0.000)				47.79 (0.000)

Non-standardized regression coefficients with AR1-correction and panel-corrected standard errors (in parentheses).

p < 0.10, **p < 0.05, ***p < 0.01.

Model 2 includes the full set of independent variables. The coefficient for union density is negative and significant (β = −0.708, p < 0.01), as is the coefficient for employee representation (β = −0.567, p < 0.01). Notably, the (absolute) size of these two coefficients increases considerably when we control for the full set of independent variables. Also, the coefficients for unemployment and fatal accidents are negative and highly significant (p < 0.01). In this full model, the variable measuring the industrial employment share is negative and significant (β = −0.845, p < 0.01), which might indicate that the manufacturing sector is often targeted for occupational safety and health arrangements and initiatives. Models where only union density or employee representation are included (not shown) together with the control variables do not alter the main conclusions from model 2.²

Model 3 includes only the significant variables from model 2. Overall, there are rather small differences in the size of the coefficients between models 2 and 3, with the possible exception of that associated with the industrial employment share, which in this model is not significant. Excluding this variable has, however, no substantive effects on the coefficients associated with union density and employee representation. Including working-time arrangements as covariates at the individual level does not have any substantive effects on the results from any of these models. Excluding employee representation from model 3 reduces the size of the coefficient associated with union density slightly (β = −0.473, p < 0.01), whereas the exclusion of union density results in a somewhat more sizeable reduction of the coefficient associated with employee representation (β = −0.417, p < 0.05).

Models 4 and 5 are estimated with all variables in first-difference form. Since the variable measuring employee representation is constant for the period analysed here for a large number of countries (see Table 1), this variable is excluded from the models. These models confirm that union density is negatively associated with short-term sickness absence: in both model 4, which includes the full set of independent variables in first-difference form, and model 5, which includes only the significant variables from model 4, the coefficients associated with union density are negative (β = −71.080 and β = −0.881, respectively) and significant at the 1 percent level.

Although the purpose of this article was to analyse the relationship between employee collective voice and short-term sickness absence, preliminary analyses indicate that different dynamics might be at work for longer-term absence. When including also individuals who have been absent due to sickness the whole reference week in the dependent variable, there is a significant positive association (p < 0.05) between union density and sickness absence (when controlling for the same variables as in models 2 and 3, Table 2). There is, however, no significant association using variables in first-difference form.

A number of sensitivity tests support the results from the models in Table 1. To check for possible outliers, models 3 and 5 were re-estimated using a jack-knife procedure omitting first one country, then every possible combination of two and three countries. When one country at the time is excluded from model 3, all coefficients associated with union density are negative and significant at the 1 percent level, and when any combination of two countries are excluded, over 88 percent of the coefficients are significant at the 1 percent level, only one coefficient is significant at the 10 percent level, and remaining coefficients (31 out of 276) are significant at the 5 percent level. When any combination of three countries is excluded, all coefficients associated with union density are still negative (average β = −0.556), 78 percent of the coefficients are significant at the 1 percent level, almost 20 percent are significant at the 5 percent level, and 37 coefficients (out of 2024) are significant at the 10 percent level. There are, however, 12 coefficients that fail to reach conventional significance level. In no case, the p-value associated with these coefficients is above 0.25, and the average p-value when these 12 tree–country combinations are excluded is 0.15. Further analyses indicate that when sickness benefit generosity is included in model 3 (β = 0.110, p = 0.134) and sensitivity tests as described above are performed, all coefficients are significant at least on the 10 percent level with close to 96 percent of the coefficients being significant at the 1 percent level (average β = −0.696).³

Similar results are obtained for the coefficient associated with employee representation from model 3. When two countries are excluded, 82 percent of the coefficients are significant at the 1 percent level, only 1 percent of the coefficients are significant at the 10 percent level (average β = −0.629), and when any combination of three countries are excluded, a majority of the coefficients (73%) are significant at the 1 percent level and 4 percent are significant at the 10 percent level (average β = −0.629). There are, however, 25 coefficients that fail to reach conventional significance level, but all of these coefficients are negative (average β = −0.308, average p = 0.167).

The same form of sensitivity tests indicate that the coefficients associated with union density for models where all variables are in first-difference form (model 5) are somewhat more sensitive to the exclusion of certain countries. When one country is excluded, all coefficients are negative with 22 being significant on the 1 percent level and 2 at the 5 percent level, and when any combination of two countries are excluded, 82 percent of the coefficients are significant on the 1 percent level (and only three coefficients are significant at the 10 percent level). When any combination of three countries are excluded, 72 percent of the coefficients are significant at the 1 percent level, and 4 and 23 percent are significant at the 10 and 5 percent level, respectively. There are, however, 21 coefficients that fail to reach conventional significance levels. Ten of these combinations are when two of the Nordic countries are excluded together with another country. The coefficients associated with union density when these combination of countries are excluded are, however, still negative (average β = −0.502, average p = 0.150).

Model 6 is estimated on a panel containing only countries with non-missing values for at least 14 years, including time dummies to account for period-specific effects. Effectively, in this model most central and eastern European countries (CZ, EE, HU, LV, LT, PL, SK and SI, together with NO and DE) are excluded from the analyses. Also, in this model the coefficient associated with union density is negative and significant (β = −0.919, p < 0.05), as is the coefficient associated with employee representation β = −1.340, p < 0.05). Excluding employee representation from model 6 has very little effect on the coefficient associated with union density.

Finally, model 7 tests whether the effect of union density is curvilinear by fitting a polynomial regression model of order 2 (i.e. by including a linear $β_{1}$ X term and a quadratic $β_{2} X^{2}$ term, where X represents union density). The coefficient $β_{2}$ will be positive for U-shaped relationships and negative for inverted U-shaped relationships. In order to reduce the collinearity between the product terms, union density (and employee representation) was centred around their means (Dalal and Zickar, 2012).⁴ This model indicates that there is an inverted U-shaped relationship between union density and sickness absence, with β₁ = −0.548 (p < 0.01) and β₂ = −0.046 (p < 0.05). This result is insensitive to whether the model also controls for employee representation or not. The same form of relationship is also present in a polynomial regression model, including a quadratic term of employee representation (not shown), where the linear term $β_{1}$ is negative (−0.624, p < 0.01) and the quadratic term implies an inverted U-shaped relationship with sickness absence (β₂ = −0.062, p = 0.189).⁵ These results are congruent with the hypothesis that the negative relationship between employee collective voice and sickness absence is only present at relatively high levels of employee collective voice, and that high absence may lead to increased unionization (and hence a positive relationship between absence and unionization) at low levels of unionization, or when existing channels for employee collective voice are relatively weak or underdeveloped.

How should we interpret these regression coefficients? We will here restrict the discussion to union density, since the effect of this variable is most stable in view of the extensive sensitivity tests performed, and the fact that this variable is measured on an interval scale makes the interpretation easier. Let us begin with the linear effect from model 3, in which the coefficient of union density is −0.0056 (remember that that the coefficients in Table 1 are multiplied by 100). In 2010, the average predicted reduction in working time due to sickness was 0.234, and actual average hours worked per week were 34 hours and 32 minutes. Without sickness absence (but ignoring other reasons for absence), an average respondent would have worked 34 hours, 46 minutes and 2 seconds. Thus, on average, about 14 minutes per week is lost because of sickness absence. A 1 percent increase in union density implies that sickness absence would decrease to 13 minutes and 42 seconds per week, a reduction in sickness absence by around 20 seconds. A miniscule effect at first sight, but over a year (assuming that annual hours actually worked per worker are around 1600 hours), this amounts to over 15½ minutes. Furthermore, the average decrease in union density over the period 1996–2010 has been around 10 percent, which implies that if union density in 2010 had been held at the 1996 level, sickness absence per worker would have been about 2½ hours lower per year.

Taking into account the curvilinear effect of union density (model 7) means that the effect of union density on sickness absence depends on the level of union density. At low levels of union density (here corresponding to the lowest 25th percentile), a 1 percent increase in union density means that sickness absence increases by about 32 seconds per week, or close to 25 minutes per year. At high levels of union density (here corresponding to the highest 75th percentile), an increase in union density by 1 percent would decrease sickness absence by 36 seconds per week, or around 28 minutes per year.

Concluding discussion

This study has offered a new perspective on why short-term sickness absence differs across countries and over time. In stark contrast to approaches which have dominated scholarly debate and emphasize the role of sickness benefits and monitoring systems, the perspective outlined here emphasizes the role of institutional channels through which employees may voice their opinion and have some influence over both their immediate work environment and more strategic issues relating to organizational goals and production processes.

Although the institutional measures of worker collective voice used in this study – union density and employee representation at the enterprise level – do not necessarily say anything about the effectiveness of such institutions in representing workers’ interests, the results indicate that this perspective may offer new and important insights into why countries experience different trends in sickness absence over time. Although the statistical effects found might seem small, they are significant and, we would argue, non-trivial. A 10 percent reduction in union density (roughly equivalent to the average decrease between 1996 and 2010) in the countries studied here translates into an average increase in short-term sickness absence of about 2½ hours per year. However, there are also strong indications that the effect on short-term sickness absence of changes in union density (and employee representation) depends on the initial level: If union density is high, an increase is associated with lower short-term sickness absence; but the opposite is the case at low levels of union density.

This study has also highlighted the importance of distinguishing between short- and long-term sickness absence. It could be argued that short-term sickness absence is influenced more by calculative behavioural decisions (Blank and Diderichsen, 1995). Since employee collective voice can function as channel through which workers may voice their opinions, this may result in reduced withdrawal behaviour in the form of short-term sickness absence. In contrast, in the case of longer-term sickness absence, unions might protect workers against disciplinary action for ‘excessive’ (from employers’ perspective) sickness absence, thereby increasing (long-term) sickness absence (Veliziotis, 2010). Moreover, in most countries there are important differences in the rules governing short- and long-term absence in sickness benefit and related schemes, where longer absence often requires medical certification and more employer responsibility for rehabilitation measures. The protection unions provide workers might be even stronger when these and related factors are also taken into account. It is, in this context, also important to emphasize that although the aim of this article has been to analyse short-term sickness absence, preliminary empirical analyses indicate that different dynamics might work for longer-term absence and that employee collective voice might be associated with an increase in long-term absence also at the aggregate level.

Commentators have noted that sickness absence is a complex and multi-causal phenomenon, and that it is hardly feasible to synthesize knowledge from different theoretical perspectives into a unified theory (Allebeck and Mastekaasa, 2004). I believe this to be true. A multitude of factors, ranging from the health status of individuals (or populations) to the state of the economy, are probably important for understanding why sickness absence differs across countries and over time. I hope that this study has contributed to a fuller understanding of this phenomenon by underlining the importance of factors relating to democracy at the workplace and channels through which workers may influence the conditions under which they work. Future research in this area should, I believe, develop better measures and theoretical understandings of these channels and how they may interact with other factors of importance for sickness absence.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship and/or publication of this article.

Notes

Author biography

Ola Sjöberg is a professor of Sociology at Stockholm University.

References

Allebeck

Mastekaasa

(2004) Causes of sickness absence: Research approaches and explanatory models. Scandinavian Journal of Public Health 32(Suppl. 63): 36–43.

Barmby

Ercolani

Treble

(2002) Sickness absence: An international comparison. Economic Journal 112(480): F315–F331.

Beemsterboer

Stewart

Groothoff

Nijhuis

(2009) A literature review on sick leave determinants (1984–2004). International Journal of Occupational Medicine and Environmental Health 22(2): 169–179.

Belin

Zamparutti

Tull

Hernandez

Graveling

(2011) Occupational Health and Safety Risks for the Most Vulnerable Workers. Brussels: European Parliament.

Bennett

Kaufman

(2004) What do unions do? A twenty-year perspective. Journal of Labor Research 25(3): 339–349.

Blank

Diderichsen

(1995) Short-term and long-term sick-leave in Sweden: Relationships with social circumstances, working conditions and gender. Scandinavian Journal of Public Health 23(4): 265–272.

Chadwick-Jones

Nicholson

Brown

(1982) Social Psychology of Absenteeism. New York: Praeger.

Choi

(2001) Unit root tests for panel data. Journal of International Money and Finance 20(2): 249–272.

Dalal

Zickar

(2012) Some common myths about centering predictor variables in moderated multiple regression and polynomial regression. Organizational Research Methods 15(3): 339–362.

10.

Ebbinghaus

Visser

(1999) When institutions matter: Union growth and decline in Western Europe, 1950–1995. European Sociological Review 15(2): 135–158.

11.

Eurostat (2015) Eurostat database. Available at: http://ec.europa.eu/eurostat/data/database

12.

Freeman

(1976) Individual mobility and union voice in the labor market. American Economic Review 66(2): 361–368.

13.

Freeman

Medoff

(1984) What Do Unions Do? New York: Basic Books.

14.

Furåker

Bengtsson

(2013) Collective and individual benefits of trade unions: A multi-level analysis of 21 European countries. Industrial Relations Journal 44(5–6): 548–565.

15.

Hirschman

(1970) Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States. Cambridge, MA: Harvard University Press.

16.

Karasek

Theorell

(1990) Healthy Work: Stress, Productivity, and the Reconstruction of Working Life. New York: Basic Books.

17.

Kivimäki

Head

Ferrie

Shipley

Vahtera

Marmot

(2003) Sickness absence as a global measure of health: Evidence from mortality in the Whitehall II prospective cohort study. British Medical Journal 327(7411): 364–370.

18.

Kohl

Platzer

H-W

(2007) The role of the state in central and eastern European industrial relations: The case of minimum wages. Industrial Relations Journal 38(6): 614–635.

19.

Leigh

(1984) Unionization and absenteeism. Applied Economics 16(1): 147–157.

20.

Leigh

(1986) Correlates of absence from work due to illness. Human Relations 39(1): 81–100.

21.

Lewis

Linzer

(2005) Estimating regression models in which the dependent variable is based on estimates. Political Analysis 13(4): 345–364.

22.

Luchak

Gellatly

(1996) Exit-voice and employee absenteeism: A critique of the industrial relations literature. Employee Responsibilities and Rights Journal 9(2): 91–102.

23.

Mastekaasa

(2013) Unionization and certified sickness absence: Norwegian evidence. Industrial & Labor Relations Review 66(1): 117–141.

24.

Mazzuco

Suhrcke

(2010) What Does Eurostat’s Labour Force Survey Say about Health and Health Inequalities in the European Union? Copenhagen: WHO Regional Office for Europe.

25.

OECD (2014) OECD indicators of employment protection: Annual time series data 1985–2013 Available at: www.oecd.org/employment/emp/

26.

Osterkamp

Röhn

(2007) Being on sick leave: Possible explanations for differences of sick-leave days across countries. CESifo Economic Studies 53(1): 97–114.

27.

Scruggs

Jahn

Kuitto

(2014) Comparative welfare entitlements (Data Set 2, Version 2014-03). Available at: http://cwed2.org/

28.

Sparks

Cooper

Fried

Shirom

(2013) The effects of working hours on health: A meta-analytic review. In: Cooper

(ed.) From Stress to Wellbeing Volume 1: The Theory and Research on Occupational Stress and Wellbeing. Basingstoke: Palgrave Macmillan, pp. 292–314.

29.

Spector

(2009) The role of job control in employee health and well-being. In: Cooper

Quick

Schabracq

(eds) International Handbook of Work and Health Psychology, 3rd edn. Chichester: John Wiley & Sons, pp. 173–195.

30.

Steers

Rhodes

(1978) Major influences on employee attendance: A process model. Journal of Applied Psychology 63(4): 391–407.

31.

Theorell

(2004) Democracy at work and its relationship to health. In: Perrewe

Ganster

(eds) Emotional and Physiological Processes and Positive Intervention Strategies. Amsterdam: Elsevier, pp. 323–358.

32.

Tompa

Marshall

Miao

(2010) Social protection and the employment contract: The impact on work absence. Work: A Journal of Prevention, Assessment & Rehabilitation 37(3): 251–260.

33.

Traxler

(2000) Employers and employer organisations in Europe: Membership strength, density and representativeness. Industrial Relations Journal 31(4): 308–316.

34.

Veliziotis

(2010) Unionization and sickness absence from work in the UK. ISER Working Paper Series no. 2010–2015, May. University of Essex, Institute for Social and Economic Research. Available at: https://www.iser.essex.ac.uk/files/iser_working_papers/2010-15.pdf

35.

Visser

(2009) The quality of industrial relations and the Lisbon strategy. In: Visser

(ed.) European Commission Industrial Relations in Europe 2008. Luxembourg: Office for Official Publications, pp. 45–72.

36.

Visser

(2011) Variations and trends in European industrial relations in the 21st first decade. In: Visser

(ed.) European Commission Industrial Relations in Europe in 2010. Luxembourg: Office for Official Publications, pp. 17–54.

37.

Visser

(2013) ICTWSS: Database on institutional characteristics of trade unions, wage setting, state intervention and social pacts in 51 countries between 1960 and 2014. Amsterdam Institute for Advanced Labour Studies (AIAS). Available at: http://archive.uva-aias.net/207White