Just-in-time Production, Work Organisation and Absence Control by Joseph Lanfranchi (LEM, Université Panthéon-Assas and Centre d’Etudes de l’Emploi) and John Treble (Swansea University)∗

Abstract Studies of sick pay and absenteeism have traditionally treated absence as a worker-related phenomenon. There are good reasons to suppose, though, that firms’ incentives to control absenteeism are not uniform. Using an employee/employer matched dataset, we investigate the relationship between the firm’s production methods and the generosity of its sick pay. The results suggest that firms who might be expected to value reliability highly, which we characterise as those which use Just-In-Time (JIT), are more likely to provide less generous sick pay. Those findings survive when we control for the use of complementary policies that buffer production from absence shocks.

∗

ACKNOWLEDGEMENTS: We thank the participants at seminars at University of Newcastle upon Tyne and Paris II,

the EALE Conference in Blankenberge, AFSE Conference in Lyon and Journées de Microéconomie Appliquée in Rennes. We also thank three referees and Martyn Andrews, the editor of this journal, who have helped to improve the paper. We are also grateful to the British Council and the French Ministry for Foreign Affairs for financial support under the Alliance programme.

1. Introduction Studies of sick pay and absenteeism have traditionally treated absence as a worker-related phenomenon. There are good reasons to suppose, though, that firms’ incentives to control absenteeism are not uniform. In particular, it has been shown1 in the context of a formal model that the presence of complementarities in production increases the costs of absence to a firm. Thus absence by a worker who works as part of an interdependent team may be more expensive than for one whose work is more individualistic because, for a team worker, absence is detrimental to the productive effort of other team members. Another source of complementarity has to do with the maintenance of stocks of semi-finished goods. If stocks are held, absence will be less expensive than if not, since without stocks, production downstream of a missing worker will be unable to continue, while if stocks can be held, production downstream of a missing worker can continue until the stocks are exhausted. There are, of course, many ways in which costs of absence can be managed. One is by using buffer stocks of workers, either employing more than are strictly necessary for the operation of the production process, using supply arrangements (through agency suppliers, or a register of supply workers), or by ensuring that workers are sufficiently flexibly-skilled to enable the missing worker’s place to be temporarily filled. Firms can also manage the incidence of absenteeism itself, and there are many ways of doing this, including hiring workers with known reliability characteristics; monitoring attendance; and providing financial (or other) incentives either as part of a sick pay (or sick leave) scheme, or as part of the firm’s disciplinary arrangements. The point made and documented empirically in the present paper, is that just as firms are likely to differ in their attitudes to absenteeism, so are they likely to differ in their modes of

1

By Coles and Treble (1996). In addition, Coles et al. (2007) demonstrate empirical support for this idea , while Heywood et al. (2006) present an analysis of the relationship between production technologies and monitoring, which is in a similar spirit.

1

absenteeism control. Furthermore, these differences have a clear connection with the production methods used by firms. Our work belongs to the literature that focuses on the relationships between technological and organizational changes and new practices in human resources management. Among the initially influential work, Womack and Ali (1990) emphasised the trend towards the reduction of buffers in production techniques summarised in the term ‘lean production’. Following this line of research, MacDuffie (1995) focused on the increased value of problem-solving capabilities in the workforce when inventories of semi-finished goods are not available to mitigate production problems. Using his words, “dealing with the problems raised by the new flexible production systems requires motivated, skilled and adaptable workers”. We propose to add ‘reliable’ to this list of required qualities, reliability being indeed a component of workers’ effort 2,3 After providing a theoretical discussion of why firms’ choices of personnel policy may depend on reliability considerations, we investigate the relationship between a firm’s production methods and the generosity of its sick pay using an employee/employer matched dataset constructed from two French data sources. It provides information about firms’ production methods, their labour force management (including absence-related incentive schemes), and the workers’ characteristics including their recorded absence rate. The conjecture that there are patterns in the human resources policies of firms related to specific characteristics of their production methods is also broadly confirmed by the empirical evidence presented here. For reasons that are made clear in the next section of the paper, the analysis focuses on the usage of just-in-time (JIT) production methods. We show that firms who use JIT, also display employment patterns that are concentrated on those demographic groups (the male, the young) that typically have lower absence rates than others. Furthermore, the generosity of 2

Note that we do not deal with the patterns of productive and organizational changes on firms’ productivity. For an empirical evaluation of the efficient association of practices, see for example MacDuffie (1995), Ichniowski et al. (1997), Cappelli and Neumark (2001). 3 The case for reliability, and specifically, attendance being a component of worker effort is made by Flabbi and Ichino(2001), and by Audas et al.(2004)

2

the sick pay regimes offered by JIT firms is less than for their non-JIT counterparts, even when buffering measures are in place. Using a policy that buffers production from absence shocks tends to raise the generosity of sick pay: for example, firms that have flexible teams, enabling substitution of absent workers, can be more generous with sick pay, ceteris paribus, than those without. We do not suggest that the main driver of these relationships is an overwhelming concern with absenteeism on the part of management. The theory proposes an equilibrium generated by workers making choices in a manner that they see as benefiting themselves, or their households, and firms making choices that will generate profit. So, for instance, when we speak of ‘buffering measures’, we do not imply that these measures have been introduced specifically with the intention of buffering absence. It is rather that whatever reasons may have caused the firm to introduce the measures, they have an absence buffering effect. The paper opens with a theoretical discussion about the hypothesized link between absenteeism control and methods of production. It then continues with a detailed description of our data source and definition of both dependent and independent variables. Finally, we present our statistical analysis and discuss our main results before making some brief concluding remarks.

2. Theory The main idea of the theory of absence described by Coles and Treble (1996) is that the observed rate of absence is not a simple consequence of a labour supply function, but a more complex equilibrium outcome generated by market interaction between the interests of workers and their households on the one hand, and the interests of firms on the other. Coles and Treble show that variation in firms’ cost of absence arises through the extent of complementarity between inputs in the production function. In particular, if workers contribute jointly to production, then the absence of one of them will involve the loss, not only of that worker’s output, but also of 3

productivity of his/her partners. The same is true of complementarity between labour and capital inputs. If stocks of semi-finished goods are not available, then they have to be produced continuously in order to enable downstream workers to continue production. Of course, there are techniques that can be used to buffer the impact of complementaries in the production process. These include the use of some kinds of flexible working, in particular, the ability of workers to take over another worker’s role during an absence, and the maintenance of stocks of semi-finished goods4. The contracts envisaged in the model specify, among other things, pay and an acceptable level of absence. Again, among other things, the equilibrium features an efficient acceptable level of absence that the firm attempts to enforce as part of its workers’ conditions of work. What techniques can firms use to manage absence towards the efficient level? They fall into three broad categories: hiring, monitoring and incentive systems. That these are not independent is highlighted by the fact that firms offer employment packages that give higher wage rates to more reliable workers. To the extent that sorting, through the usual mechanisms of hiring, probation and firing, fails to deliver the efficient level, the firm will benefit by devoting resources to enforcement. The advice given to managers regarding enforcement usually involves some kind of monitoring and incentive schemes5. These schemes provide rich evidence of managerial ingenuity. Compare for example, the schemes described by Barmby et al (1991), Brown (1999) and Hassink and Koning

4

Although the rigorous derivation provided by Coles and Treble (1996) is novel and useful as a basis for empirical work, many of the implications of the theory have been noted before. For instance, a converse version is given in Aoki’s (1988) well-known book on lean production methods in Japan: “Savings on inventory cost: If the process used to produce a variety of final outputs using a large number of components is organised in a (quasi-)tree structure, it can respond to continual market fluctuations with reduced inprocess inventories through horizontal co-ordination. But it may be vulnerable to drastic shocks because of the reduction in buffer inventory. The “zero inventory” requirement to dispense with buffer inventory necessitates the effective control of local shocks, such as the malfunctioning of machines, absenteeism of workers, and quality defects, in order to minimize their effect on the smooth operation of horizontal coordination.” (p.36) 5

See for example the British government’s advice at http://www.hse.gov.uk/sicknessabsence/

4

(2009), which involve, respectively, changing sick pay entitlements, changing overtime opportunities, and free lottery tickets. The starting point for this paper is that just as a firm’s efficient level of absenteeism is determined in part by its choice of production method, so are its incentives to control workers’ behaviour. Firms for which absenteeism is cheap will neither be prepared to pay a wage premium for very reliable workers, nor will they be prepared to devote substantial resources to absence control. On the other hand, firms for which absence is expensive will find it worthwhile to attract reliable workers by paying a wage premium and also to devote resources to enforcing desired standards of attendance. Therefore, we will argue that those firms will enforce the efficient level of reliability through an increase in the cost of absenteeism for the worker. That is, by offering a less generous replacement rate in case of sickness. Whether absenteeism is cheap or not depends on the kind of production methods used by the firm. Coles et al. (2007) use the phrase robust to absenteeism (or robust for short) to describe production methods where the loss of production due to a single worker’s absence is close to that worker’s own marginal product. The distinction between robust and non-robust production methods depends crucially on the extent to which productivity of the production process hinges on the attendance of any particular worker. In a craft workshop where workers are assigned individual projects to work on, the cost of an absence is limited to the consequences of the absent worker not working, because his or her absence does not affect the productivity of any other worker. If work is organised in teams, the cost of an absence is higher, since the absence of an individual not only means the loss of that person’s product, but that the productivity of others will be affected, too6. Of course, most workplaces will include a mix of methods, with maintenance functions (for example) perhaps being more robust than a production line.

6

Heywood and Jirjahn (2004) offer evidence that firms using teams have lower absence rates.

5

Aspects of a worker’s relationship with capital equipment are also important. In particular, note that the storage of semi-finished product is important in determining the costs of an absence. A half-finished chair will still be a half-finished chair, even if it is not worked on for six months. A half processed chicken is waste if it is not frozen within a matter of hours. Finally, instead of trying to manage absence costs, firms can employ buffering techniques, in an attempt to mitigate the impact of absence on the rest of its operations. These include the use of overmanning, the use of supply workers, and the use of temporary work agencies. Another buffering technique is the use of flexible working methods, such as ensuring that all members of a team are able to carry out any of the tasks for which the team is responsible, allowing work to be rotated between members of the team. If a team member is not present, his or her role can be undertaken by other team members, thus reducing the impact of the absence on team productivity. In the empirical work that follows, our criteria for choice of measures of production method and work organisation follows these arguments. We concentrate on measures that capture aspects of inventory-holding and team work. Our claim is that firms who do not have a robust production method (specifically, those that use JIT production methods) should have an employment structure that is biased towards those demographic groups that display low levels of absence, that they should be more likely to monitor absence and that they should be less generous in sick pay provision. However, the generosity of sick pay will also be influenced by the extent to which firms use practices that buffer the impact of absence on production.

3. Measuring Generosity of Sick Pay This section describes the dataset used in our empirical analysis, and how we construct measures of the generosity of sick pay. Two matched data sources enable us to draw on a wider range of information about the respondent firms than has previously been used.

6

The Enquête sur le coût de la main d’oeuvre et la structure des salaires (or ECMOSS) is a large scale survey of French industrial establishments and their workers carried out by INSEE in 1992. At establishment level, the survey provides the most extensive set of information about labour costs (including sickness absence) and compensation policies. The 1992 wave also contained questions about work organization. Each surveyed establishment was instructed to draw a representative sample of employees for which detailed information about individual characteristics, compensation and behaviour was provided. The Enquête therefore consists of two parts: one concerning the productive unit as a whole; the other concerning individual employees. In the same year, the French Ministry of Labour carried out a second survey known as Réponse. It was designed to study industrial relations within workplaces in a similar manner to surveys in UK (WIRS), Australia (AWIRS) and Ireland., Réponse provides information about unions and bargaining activity, industrial relations and changes in technology and work organisation. Since it was administered to a subsample of the establishments who responded to ECMOSS, this information is easily merged into the Enquête database. The resulting matched sample is a representative cross-section of 1,983 establishments belonging to firms with at least 50 employees from the non-agricultural private sector with representative data on 19,699 workers. The uniqueness of the dataset comes from the fact that it embodies the structure of the workforce, the provision of sick pay benefits, the monitoring of attendance, and the nature of the production methods and work organization. We now describe the legal minimum and supplementary sick pay schemes offered by French employers and how we constructed our measure of sick pay generosity. Generosity in the French sick pay system Sick pay in France is regulated by the Social Security law. This lays down minimum levels of provision, and a division of responsibility between the state and employing firms. The Régime Général specifies state payments at a replacement rate of 50% for 60 days, following a three day

7

waiting period (délai de carence). Qualifications for the state benefits imply that virtually all workers will be eligible.7 In addition, employing firms must make complementary payments, of 40% for the first 30 days of a spell of sickness and 16.66% for the next 30 days. These payments are subject to a ten-day délai de carence and are payable only to workers with at least 3 years tenure in their job. The system is illustrated in Figure I. For workers with tenure in excess of 8 years, the period over which sick pay is payable is extended according to the schedule shown in the horizontal scale. The Régime Général provides a minimum level of replacement, but employing firms are free to exceed these provisions if they wish. They can make more generous provision in a number of ways: by increasing the replacement rates, extending the period covered, reducing the délai de carence, or reducing the 3-year tenure qualification. The actual provision for the employees of a particular firm may therefore be quite different from the provision specified by the Régime Général. These variations are made at two levels: the level of the industry, or locally at each establishment. Industry level negotiations take place for each of about 350 branches8, and typically final agreements (“conventions collectives de branches”) determine whether or not the establishments within each branche will use a supplementary scheme or not. A level of supplementation of the Régime Général’s provisions is also agreed, so that all establishments in a branche that uses a supplementary scheme will be subject to higher minimum provision than those in branches without. Local negotiations may also enhance the rates of sick pay provided, but where the Régime Général is used the structure of these enhancements is constrained by its provisions. The dependent variables used in this paper are various measures of the generosity of sick pay. Firms are asked whether the majority of their employees are affiliated to the Régime Général

7

Workers must have had at least 200 hours of work in the 3 months prior to the spell of sickness, or have been paid at least 1.015 times the minimum wage in the previous 6 months. 8 The classification of establishments into branches is different from the industry classification used in France for statistical purposes.

8

or not. This is necessary, because a few industries have special arrangements that replace the Régime Général. We have had to drop establishments in these industries from our analysis, since the survey did not record details of such schemes. After further deletion of establishments that did not even answer the affiliation question, we were left with a sample of 1,690 productive units. Respondent establishments are asked if they have a supplementary system of sick pay. Those firms that do are asked to describe its provisions for up to four groups of workers, distinguished by their tenure in the firm. If the scheme used is too complex to be described by four sets of parameters, the four most frequent sets are supposed to be reported, so that coverage of the workforce is maximised. In the present paper we have used information only from the first scheme reported.9 The replacement rate is a useful measure of the generosity of a benefit system. It measures the proportion of normal income that is paid to a person eligible for sick pay. In the case of the Régime Général, which is illustrated in Figure I, it is clear that replacement rates vary with the duration of a spell. Because of this, we use the mean replacement rates during spells of different lengths as dependent variables in our analysis. Specifically, because of the délai de carence, the replacement rate changes as a spell of sickness absence proceeds. Thus for the first three days the rate is zero, for the next seven it is 50%, it then rises to 90% for eligible workers, and falls again after 60 days. Eligibility depends on seniority. There are three degrees of freedom that firms have in providing more generous coverage than the state scheme allows. They can vary the rates, the minimum seniority requirement, or the délai de carence. Our aim is to develop a parsimonious

9

There are two reasons for doing this. First, establishments reporting the use of more than one supplementary system were also more likely to fail to fully describe all characteristics of the sick pay schemes, implying a trade-off between sample size and measurement error. Second, most of the establishments offering multiple supplementary schemes provide a high replacement ratio for very long spells of absence: 22% of the establishments declared a second supplementary sick pay scheme covering the first month of absence but only 10% offered an arrangement distinct from the first scheme in terms of generosity over this month; similarly, 14% declared a third supplementary sick pay scheme and 8% distinct from the first over the first month; finally only 8% declared up to four schemes with only 4% distinct from the first over the first month. Furthermore, among the 19699 sampled workers, only 1730 employees were protected by a system different from the first scheme reported (8.7%). Consequently, we do not think it will make a great deal of difference to the outcome.

9

representation of the generosity of a scheme . The measure we use is the mean replacement rate throughout a spell. This measure is not unique, since it varies with spell length. Begin by defining, for each worker (indexed by i ), a sequence of variables: {τ i (d ), d = 1,..., T }.

τ i (d ) is the replacement rate for the i ’th worker on day d of a spell of absence. Our data includes all the information necessary to compute these for each sampled worker. For each firm j , n j workers are sampled and we calculate the sequence of mean replacement rates. That is:

 τ j (d ) | τ j (d ) = 

∑

i∈ j

τ i (d )

nj

 , d = 1,..., T  . 

The τ j (d ) measure the mean replacement rate on each day of a spell (that is, the mean marginal replacement rate), but to measure the generosity of a scheme it seems more natural to think of mean replacement rates throughout a spell. These can be computed as:

 τ j (d ) | τ j (d ) = 

∑δ

=1,..., d

τ j (δ )

d

 , d = 1,..., T  

We have computed these measures for T = 29 , which covers absences up to one month in duration. The three measures for d = 4,11, 20 are the focus of the analysis carried out in the next

section of the paper. Their means and standard deviations can be found in Table I, which lists all the variables used in this paper together with two sets of statistics, one for each of the subsets of the data that we use in our analyses.

4 Analysing Generosity of Sick Pay Table II contains a crude preliminary analysis of the relationship between generosity on the one hand, and various measures of work organisation, union influence and production method on

10

the other. We have calculated the means of all our variables, conditional on whether the firm’s generosity is above or below the median generosity. The results suggest that there are indeed significant relationships between the various measures, with significant differences appearing for nine of the variables. In particular, firms with less generous sick pay tend to be small, and more of them use JIT. Fewer firms with low generosity use collective agreements; use multidisciplinary groups; rotate tasks within or between workgroups; encourage co-operation; have union representation; or monitor absenteeism. These bivariate relationships are sufficiently strong (and consistent with our theory) to encourage us to investigate with a multivariate analysis. The three generosity measures, relating to absences with durations of 4, 11 and 20 days, are continuous with a minimum specified by the Régime Général. It is therefore appropriate to use a Tobit specification to analyse them. 10 On the right-hand side, we need to include a variable (or variables) to capture the idea that production methods differ according to their robustness to absenteeism. We follow the approach adopted by Coles et al.(2007), in which firms that use JIT methods are shown to have higher costs of absence than those which do not. The use of JIT as our main characterisation of non-robust ‘production method’ is not without problems. Wood(1999) and MacDuffie(1995) show that production and work organization methods are often introduced in tandem with a number of other measures. For example, MacDuffie highlights that diminished buffers of semi-finished goods and inventory stocks together with an increased percentage of workers involved in multi-tasking and job rotation have a positive effect on performance in the automobile industry. The bundles of measures are referred to by a variety of names in the literature11, including ‘high performance work organisation measures’. The fact of this bundling creates conceptual difficulties for our analysis.

10

In a previous version of the paper, we also reported analyses of the binary variable indicating whether or not a supplementary sick pay scheme is adopted. The results of this were largely negative. For this reason, and to conserve journal space, we have dropped this discussion from the present version. 11 Kalleberg(2001) lists 8 different collective names for these practices.

11

If, as our theory suggests, the introduction of JIT (by itself) increases the potential costs to firms of absenteeism, this is probably not unnoticed by firms. The main rationale for JIT is that it reduces inventory holding costs, and enables production to be more responsive to changes in demand. A firm may therefore find it worthwhile to introduce JIT, but nonetheless anticipate that some of the gains made are dissipated in additional costs of absenteeism. The ways in which these extra costs can be diminished fall into two groups: techniques for reducing absenteeism itself, and techniques for reducing the impact of absenteeism on costs. Viewed in this way, increased control of workers’ behaviour by monitoring absence and providing incentives to attend, either through less generous sick pay provision or, for instance, tying promotion prospects to an absence record are likely to be complementary practices to JIT. Similarly, methods of reducing the impact of absence on costs are also likely to be complementary. These include the use of buffer stocks of workers, either as part of a core labour force (overmanning), or as part of a peripheral work force (supply arrangements, and agency work); increasing the ability (and willingness) of workers to fill in for absent workers, by various measures of workforce flexibility. For these reasons, we adopt an empirical specification that includes as many variables of this sort as we have access to. The rest of this section gives details of these variables which are summarized in Table I.

Production methods and Work organisation variables The ECMOSS survey includes the following questions: For each method of organisation that I cite, please tell me whether it has already been introduced, whether it is about to be introduced, being considered or not under consideration at all in your establishment: •

Just-in-time? production on demand?

12

•

….

•

Multidisciplinary work groups .

The variable that we call ‘Just-in-time’ is a dummy taking the value 1 if the establishment reported that it had already introduced Just-in-time (or Production on demand). The variable called ‘Multidisciplinary work groups’ is defined in a similar fashion. As argued before, JIT method is more sensitive to interruptions in the production process such as those created by absenteeism in the workshops. Consequently, we expect this variable to reduce the generosity of the sick pay arrangement chosen by the establishment. We expect multidisciplinarity of the workers to increase generosity since it mitigates the loss incurred by an absence. The second group of variables concerns work organization. Two questions are asked about cooperation between workers: Some establishments practice work rotations. Type A. Within work groups, workers rotate between tasks during the course of their usual work: Is this the case in your establishment? YES or NO.

Type B. Some multi-skilled workers rotate between certain tasks (independently of team organisation): Is this the case in your establishment? YES or NO and: Is direct cooperation between workers in different sections encouraged. (For establishments comprising a single branch, answer ``Not applicable’’) Three indicator variables are derived from this question: i) ‘Rotation within workgroups’, which equals 1 if the establishment answered YES to Type A; ii) ‘Rotation between workgroups’, which equals 1 if the establishment answered YES to Type B; and iii) ‘Interdepartmental cooperation’, which equals 1 if the establishment answered YES to the second question.

13

We interpret flexible working as a method by which firms can reduce the cost of absenteeism, versatile workers being more able to replace absent colleagues between and within work groups. This should therefore increase the generosity of sick pay. Thirdly, a detailed set of questions about shift working have been collapsed into a single variable as to whether shift working is part of the contract for 25% or more of the workers.12 The variable is important because shift work is a more rigid form of work organization than non-shift work. According to Cette (1995) for France, shift working is primarily introduced to increase the capital operating time in order to maximize the returns of capital asset before it becomes obsolete. Such work organization reflects a higher complementarity between labour and capital and should therefore decrease the generosity of sick pay schemes.

Measures of establishments’ absence policies and attitudes The next group of variables attempts to capture the importance of absenteeism to the firm. There are two policy variables and three attitudinal variables. The policy variables are an indicator of whether absence records are kept by the firm (‘Absenteeism regularly monitored’) and an indicator of whether those records are used by the firm in making decisions about pay (‘Absence is used in setting pay rises’). The attitudinal variables are: i) whether the firm considers that absenteeism affects its social climate (‘Indicator of social climate’); ii) and iii) whether the firm regards absence among blue-collar workers in, respectively, manufacturing and services, as a problem (‘Absence of production workers a problem’ and ‘Absence of service workers a problem’).

Control variables

12

Our definition of shift work is frequently used in the literature on compensating differentials: anyone who has scheduled working time outside normal working hours during the week is considered as a shift worker. Consequently, shift workers include those working rotating teams, but also night workers and those who work uncommon working hours, and an extended work day or work week.

14

First, we use a group of questions about union activity. These are important since supplementary sick pay arrangements are usually the outcome of bargaining between firms and unions, either at branch or at firm level. Respondents are asked if the majority of the firm’s workers are covered by a collective agreement at branch level (‘Collective agreement’). In addition, we use a question about whether there exists union representation at establishment level (‘Union representation’). We take this as a (rather weak) indicator of the probability of an establishment agreement. In addition, we use the logarithm of size of the workforce and industry of the establishment. The first has been shown by many investigators13 to be associated with absence. Dummy variables are included measuring the manufacturing or service industries in the private sector in which the establishment operates.14 Data management and restriction of the original sample are described in detail in the statistical appendix which is available from the authors. Because of missing observations, especially in the detailed description of the supplementary sick pay scheme, we finally use a sample of 842 establishments that have fully completed the set of questions used in the two surveys.15 To summarise: ideally, we would like to estimate our basic specification which is a Tobit with one of the three generosity measures on the left-hand side, and the various production method, work organisation, absence policies and attitudes and the control variables on the right-hand side. It turns out that limitations of our data set prevent us carrying out this plan without making a number

13 See, for example, Barmby and Stephan (2000). Theoretical arguments about this phenomenon are not easy to construct. Coles and Treble (1996) argue that it can be understood in the context of their theory, since larger firms have greater flexibility in rescheduling production than small ones. 14 Only manufacturing and services industries in the private sector are included. There are twelve classes of industry within the classification: 2. Foodstuffs and other agriculture-based products; 3. Production and distribution of energy; 4. Intermediate goods industries; 5. Equipment; 6. Current consumption goods; 7. Construction; 8.Retailing; 9. Transport and Telecommunications; 10. Commercial services; 11. Housing rental; 12. Insurance; 13. Other financial services. Category 3 is heavily nationalised and is excluded from our analysis because it contributes only 2 establishments to our sample. The omitted category is Category 2. 15 The statistical appendix also provides the reader with robustness checks of our empirical results using samples where missing values have been imputed in two different ways.

15

of compromises. These are necessary to cope with two problems: the possible endogeneity of the JIT variable, and the complementarity between the various high-performance work practices. i) The standard method for dealing with endogeneity problems is to extend the model to include equation(s) describing the determination of the endogenous variable(s). Taking this approach literally would require to estimate a system with one equation for each production and work organisation method that we included on the right-hand side of the Tobit. It turns out that trying to endogenise just one right-hand side variable (JIT) leads to convergence problems with the Tobit analysis of the 4-day generosity measure, which imply that we cannot rely on these estimates. However, the Tobit estimates for the 11 and 20 day analyses do not suffer these problems. In these analyses, there are not many observations to the left of the censoring value, and OLS, corrected for endogeneity of JIT, is close to Tobit. In the case of the 4-day measure, where the number of censored observations is great, the OLS alternative is unsatisfactory. Therefore, we have also estimated a linear probability model16, where the left-hand side variable is binary (‘more generous than required by law’ or ‘no more generous than required by law’). ii) In addition, we take account of the complementarity between high performance work practices by adding pairwise interactions of each of the likely complementary practices to the specification. This enables us to evaluate the effect on generosity of sick pay of the joint use of production methods, and also, to correct for a possible problem of collinearity that would imply an upward bias in the estimated standard errors of the parameters17. For comparability purpose, we report estimates of these interaction effects with the linear probability model for the 4 day analysis and with OLS model for the 11 and 20 day analyses, all corrected for endogeneity bias.

16

Wooldridge(2002) observes that the linear probability model “often seems to give good estimates of the partial effects on the response probability near the centre of the distribution of (the independent variable).” (p455) 17 A correlation matrix of production methods and organizational variables is given in Table VII.

16

5. Results Our discussion of the results begins with our estimation of a single-equation Tobit model. These results are presented in Table III, where all displayed estimates are marginal effects of the independent variables on the conditional expectation of the truncated measure of the generosity of sick pay, computed using the method described by Greene(1997, p963). The reported results indicate a robust correlation between the use of JIT and the generosity of sick pay, as generosity of sick pay is lower for JIT firms by about 1 point for 4-day durations, 2 points for 11-day durations, and 2.5 percentage points for 20-day durations. For most of the other variables, the estimates for the three durations tell similar stories. Generosity is no different in firms that use shift working. It is higher when a union delegate is present. Of the four work organisation variables (rotation within teams, rotation between teams, co-operation between teams and the use of multidisciplinary work groups), the first and the last are significant positively associated with generosity. It is only for the 4-day durations that the use of pay incentives appears to be a substitute for incentives based on sick pay generosity. For the longer durations, recording of absence is positively associated with generosity. These results are subject to endogeneity bias, so in Tables IV.1 and IV.2 we present the results of the estimation of a Tobit specification appropriately corrected for endogeneity of the JIT variable. The just-in-time indicator is a binary variable, so that if we wish to adjust the Tobit estimator to correct for endogeneity bias, we must use a binary dependent variable method in the auxiliary equation. We used a probit, and coefficients of the resulting system of equations were estimated by maximum likelihood, using a number of different specifications of the probit equation. As we noted above, the algorithm did not converge for the 4-day measure, so the Tables report results for the other two dependent variables alone. The right-hand side variables include some that are also included in the main equation, and others that are excluded. The exact

17

specifications are apparent from the Tables. The excluded variables are a selection from: the mean age of the workforce, the proportion of female workers, and five dummy variables: i)

Increasing activity and decreasing activity. These represent the establishment’s response to a question about the trend of output during the 5 years prior to 1992. Responses were reported as either ‘increasing’, ‘decreasing’ or ‘stable’, which we represent with two dummies, using the last as the omitted category.

ii)

Irregular activity. Establishments were asked if their activity is ‘rather irregular’ or ‘rather regular’ besides any seasonal variations. Our dummy variable is 1, if they reported ‘rather irregular’.

iii)

Measures to reduce inventories. Establishments were asked if they had introduced any measures intended to reduce inventory levels. The dummy equals 1 if they responded positively.

iv)

Measures to reduce production delays. Establishments were asked if they had introduced any measures intended to reduce production delays. The dummy equals 1 if they responded positively.

Looking first at Table IV.1, note that we have reported the probit coefficients only for the analysis of the 11-day measure of generosity. This is because while these estimates vary across the different analyses, the variations are very small. The column headed ‘JIT Choice’ reports these probit coefficients, which indicate significant effects of mean age, inventory reduction, establishment size and four of the industry dummies. We suggest in the light of these estimates that the inventory reduction dummy is an effective instrument. These results also show that the characteristics of the workforce of firms using JIT differ from those without. Since there is

18

considerable evidence pointing to higher rates of absenteeism among women and older people,18 this finding supports the presence of the selection effects suggested by the theory. The estimated Tobit equations show two main results: Firstly, both σ (the estimated standard deviation of the latent distribution of generosity) and ρ (the estimated coefficient of correlation between the error terms of the two equations) are both highly significant in both sets of estimates. This confirms the endogeneity of JIT. Second, the estimated coefficients of the JIT variable are significantly negative in both estimates. Table IV.2 gives an abbreviated view of the effect of including the dummy for measures to reduce production delays, instead of and as well as, the dummy for measures to reduce inventories. In both cases, the dummy for measures to reduce production delays has a positive and significant impact on the use of JIT. In this Table, we report only the estimates of σ , ρ and the coefficients of JIT in the Tobit equation. These two tables provide considerable support for our main result. It is also apparent that the uncorrected estimates measure smaller effects than the corrected ones. That is, the uncorrected Tobit estimates are biased upwards19. There are two remaining problems with this study. The first is whether we can extract any information about the short duration generosity measure. The second is the issue of complementarity between production methods and organizational practices. We attempt to solve the first problem by using alternative analytical methods to the Tobit. Since for the 11 and 20 day analyses, the number of censored observations is small, there would appear to be little loss in employing OLS in place of the Tobit. For 11 days the OLS point estimate is -1.794 and the Tobit estimate is -2.02. The corresponding figures for the 20 day analysis are -2.300 and -2.57. As it happens, the point estimates generated by Tobit and OLS for the 4-day analysis are close together too. The OLS estimate is -0.954 and the Tobit is -0.93. 18

See, for example, Barmby et al.(2002), this finding is consistent with the idea that firms using JIT have lower exposure to absenteeism induced by the structure of their workforces. The first stage probit equation χ statistic, which tests the joint significance of the instruments, is equal to 55.70. It can be seen as a good indicator that our set of instruments are not weak. 19

2

19

This approach is less convincing for the 4-day measure, which has a large number of censored observations. Here we recode the dependent variable as a binary variable indicating whether the sick pay offered is above the minimum provided by the Régime Général or not, and report the results produced by estimating a linear probability model with this as the dependent variable. Table V collects together results from the Linear Probability and OLS models corrected for endogeneity bias using a treatment effect procedure suggested by Maddala (1983), while in Table VI, we address the problems of complementarity between production methods and organizational practices by including in supplement a comprehensive set of two-way interactions. The results in Table V confirm the strength of the relationship between generosity of sick pay and the use of JIT methods for the longer duration measures., The linear probability model for the 4-day measure still shows quite a strong association with JIT use. However, this relationship is less strong when the complementarity relationships are accounted for in Table VI. As an example of how the estimations in Table VI.1 work, consider the first column. This regression includes the variable of interest, JIT, plus four supposedly complementary measures: rotation within and between workgroups, co-operation between workgroups, and the use of multidisciplinary workgroups. There are then 10 possible two-way interactions. When interactions are included in this way, the impact of any single measure can be assessed (at the mean) by computing the weighted sum of the coefficients in which it is included, where the weights are the means of the variables with which it is interacted. Thus if variable X, is interacted with variables Y and Z, and the estimated coefficients of X, XY and XZ respectively are β X , β XY and β XZ , the gross impact of X on the dependent variable is β X + β XY Y + β XZ Z . In contrast, β X only measures the impact of X in isolation, that is, when Y and Z take the value zero. Tests of linear hypotheses can be carried out in the usual way for the three measures of generosity and the gross impacts of the five measures are reported in table VI.2.

20

It is worthwhile dwelling a little on the interpretation of the results in Tables VI.1 and VI.2. Table IV.2 reveals significant gross associations of JIT, rotation within teams, and the use of multidisciplinary workgroups with the three generosity measures for OLS models. When we consider the linear probability model results, it appears that the impact of the use of JIT is only significant at the weak threshold of 16.2%. The OLS estimates indicate that the use of JIT lowers the generosity of sickpay over 4 days absence by over 5.85 percentage points, that use of rotation within teams tends to raise it by 2.03 percentage points, and the use of multidisciplinary workgroups tends also to raise it by a bit less than 1.10 percentage points. The magnitude of these effects varies across the different durations of spells that we consider here, while the qualitative pattern remains stable. Another way to assess this complementarity effect is to notice that when these various production methods and organizational measures are used in isolation, their level of association with generosity is strictly higher. Thus while firms using measures like rotation within teams and multidisciplinary workgroups may pay more generous sick pay, that generosity is less than it would have been had JIT not been used as part of the package. We interpret this to be an illustration of the buffering effect described above. Multidisciplinarity and rotation tend to make absence cheaper, JIT tends to make it more expensive.

6. Conclusion This paper presents what we believe to be the first evidence for relationships between the nature of a firm’s production methods and its personnel policies, particularly as they are directed towards absence control. Such evidence is hard to assemble because data sources containing information about both technology and monitoring and incentive schemes are rare. Indeed, as far as

21

we know the French data we use here are unique in this respect. Certainly, no similar data exist for Britain. Even given the limitations of the data, the results suggest strongly that firms’ choices of personnel policy are associated in a significant way with reliability considerations. Firms who might be expected to value reliability particularly highly (those using JIT production) are seen to have workforces with a demographic profile different from other firms. They also tend to be more concerned to provide appropriate incentives, in the form of additional and/or more generous sick pay. The effects that we measure are robust in sign, but vary quite substantially in magnitude. We are strongly convinced that the qualitative relationship exists, but we would not be prepared to commit ourselves to any particular size for these relationships. We believe that the importance of this work is rather wider than simple illumination of what determines firms’ policies. Absence research has been dominated by the idea that absenteeism is a supply-side phenomenon: workers decide when and how often they are going to be absent, and thus determine the observed absence rate. The present research suggests that this view is excessively restrictive. Although workers themselves make the decision to be absent, the decision is moderated by the policies of the firm. In seeking explanations of the observed absence rate, we must look beyond models of worker behaviour, to consider the interests and actions of the firms in which they are employed. The work also has relevance for two related debates in the industrial relations literature. There is a considerable body of work discussing the impact on productivity and worker well-being of high performance work organisation measures20. These were originally portrayed as performance enhancing, in the sense not only of increased productivity, but also in the sense that they improved worker welfare through increased involvement in production decisions, and empowerment in the

20

See for example Ichniowski et al.(1997). Kalleberg (2001) provides a useful survey of this literature.

22

workplace. A more recent strand of literature21 has criticised the claim that such measures are uniformly beneficial to workers. Our work can be interpreted as shedding some light on the way in which one vector of work intensification may operate, through increased absence costs to firms.

21

See for example Green and McIntosh (2001), and Brenner et al.(2004)

23

Statistical Appendix This statistical appendix describes our two data sources, how the matching was done, and gives details of the implications of the matching for the final sample size. Our first source, The 1992 Enquête sur le coût de la main d’oeuvre et la structure des salaries (ECMOSS), collects establishment compensation information as well as individual wages for a sample of employees in a sample of establishments with more than ten employees in manufacturing, construction and service industries. The sampling rate has been stratified according to the sector, the region and the size of the productive unit, and varies from 1 for the establishments of more than 500 employees to 1/48 for establishments between 10 and 20 employees. The sampling rate for the employees is based on their month and year of birth. The average number of sampled workers per establishment is around 10, but varies between establishments, partly because the sampling rate falls as establishment size decreases. ECMOSS is then a nationally representative cross-section of establishments with representative data on workers. It contains information on 15,858 establishments as well as an employer-reported description of individual characteristics of 148,976 of their employees. Our second data set is the Enquête sur les Relations Professionnelles et les Négociations d’Entreprise (REPONSE) and collected by the French Ministry of Labour for the same year, 1992. The REPONSE sample was drawn from a subset of the ECMOSS population consisting of 12,293 establishments from firms with at least 50 employees. Among these, 3,091 managers and union representatives of production units were interviewed resulting in 2,998 usable questionnaires. The sampling rate is stratified according to 6 classes of size, 5 large industries and 9 regions. Only two thirds of the REPONSE respondents also responded to ECMOSS, resulting in an original merged sample of 1,983 establishments. Accordingly, our final matched file can be seen as a representative sample of the French establishments belonging to firms with at least 50 employees from the non agricultural private sector.

24

The information about sick pay comes from the establishment questionnaire in ECMOSS. Establishments are asked firstly if they follow the Régime Général of the French Social Security System or not. They are then asked if they have a complementary system of sick pay. Among the 1,983 establishments, 60 did not answer those questions while 233 reported that they did not follow the Régime Général because they have special regulations. Amongst these are firms that are at least partly state-owned. Since we do not have any information on those special rules, we restricted the sample to the remaining 1690 productive units. This sample remains representative of the establishments where the majority of the workforce is regulated by the Régime Général. Respondents who report they operate a complementary system are finally asked to specify its provisions. Up to four systems can be described by each firm. Respondents who have more than four are asked to report details of those that cover the four largest groups of workers. The systems are described by the minimum tenure qualifying a worker for sick pay, the number of days of carence, and the replacement rates month by month for up to six months of absence. Before the data could be used, some cleaning was necessary to correct some obvious mistakes in the figures reported by the establishments for the characteristics of their complementary system. Some of these problems were easily tackled: i) some respondents did not enter their responses in the correct lines of the form. Such responses are easily identified by the pattern of missing values, and also easily corrected. ii) some respondents entered replacement rates as proportions rather than as percentages. Again, these are easily identified and corrected by multiplying by 100. iii) those respondents who reported the firm’s contribution to the Régime Général as a système complémentaire are also easily identified, since their reported replacement ratios are recorded as 4000 for the first month and 1600 (or 1666) for the second month. Once again these cases are easily recognised and amended. If such a response appeared as a single système complémentaire we also altered the response to the question asking if such a system was in use.

25

iv) a number of other changes (and a few deletions) were made in cases where the coded responses were either obvious nonsense or inconsistent with the law.22 Among the 1690 sampled establishments, 205 offer the minimum replacement rates specified in the Régime Général. Of the remaining 1485 productive units 631 provide insufficient information to enable calculation of the generosity measure. Of these, 36 establishments fail to report fully the tenure of the sampled workforce, 631 omit the number of days of carence and 255 omit the minimum tenure qualifying workers for sick pay. On the other hand, it is worth noting that all respondents reported the replacement ratio during the first month of absence spell. Consequently, no establishment failed to describe its complementary sick pay scheme completely. Missing values for these determinants cannot be imputed since they are already the main components of one of our dependent variables, the generosity of the complementary sick pay scheme. Instead, we have restricted our main analysis of the generosity of the establishments to those who had reported all the characteristics of their complementary sick pay scheme, that is, 854 establishments. In the table AI below, we present the average values for the explanatory variables in respectively the initial sample of 1690 establishments and the final sample used in the main body of the paper. Including a higher percentage of establishments not offering any complementary sick pay, the group of 854 establishments contains productive units on average smaller, less unionized, using multidisciplinary workgroups and encouraging cooperation between workers less frequently. However, this sample contains the same proportion of establishments that have implemented Justin-Time and rotation between and within workgroups. The average values of the variables attempting to capture the importance of absenteeism to the firm are also not significantly different between the two samples. Despite choosing this route for our preferred dataset, we also assessed the robustness of our result by replicating the estimations presented in the fourth section of the paper in two different

22

The STATA program used to clean the data is available from the authors on request.

26

samples. In the first one (sample 2), we imputed the value in the Régime Général when the information was missing, that is 3 days of délai de carence and 36 months of minimum seniority requirement. In the second one (sample 3), we have chosen to replace the missing value by the mode of the variable in the largest available sample, that is 3 days of délai de carence and 12 months of minimum seniority requirement. The qualitative nature of the results discussed in the text remains unchanged. In particular, note that the negative effect of the use of Just-in-Time method of production on sickpay generosity is not qualitatively changed.23

23

All results referred to, but not included in the paper, are available from the authors on request.

27

References Aoki, M. (1988): Information, incentives and bargaining in the Japanese economy, CUP, Cambridge. Audas, R, T.A. Barmby and J.G. Treble (2004): “Luck, Effort and Reward in an Organisational Hierarchy”, Journal of Labor Economics v22(2), pp 379-396. Barmby, T.A., M.G. Ercolani and J.G.Treble (2002): “Sickness Absence: An International Comparison”, Economic Journal v112(480) ppF315-331. Barmby, T.A., Orme C. and J.G. Treble (1991): “Worker absenteeism: An analysis using microdata.”, Economic Journal, v101(405), pp 214-229. Barmy, T.A. and G. Stephan (2000): “Worker Absenteeism: Why Firm Size May Matter”. The Manchester School, v68(5) pp568-77. Brenner, M.D., Fairris, D. and Ruser, J. (2004): “‘Flexible’ work practices and occupational safety and health: Exploring the relationship between cumulative trauma disorders and workplace transformation”, Industrial Relations, v43(1), pp242-266. Brown S (1999): “Absenteeism and overtime bans”, Applied Economics, 1999, v31(2), 165-74 Cappelli, P. and Neumark, D. (2001): “Do ‘high performance’ work practices improve establishment-level outcomes?”, Industrial and Labor Relations Review v54(4), pp737-775. Cette, G. (1995): “Capital Operating Time and Shiftworking in France.’, in D.Anxo, G.Bosch, D.Bosworth, G.Cette, T.Sterner and D.Taddei, ed., Work Patterns and Capital Utilization : An International Comparative Study, pp149-175, Kluwer Academic Publishers, Dordrecht Coles M.G. and J.G. Treble (1996): “Calculating the cost of absenteeism” Labour Economics, v3, pp.169-188. Coles M.G., Lanfranchi J., Skalli A. and J.G. Treble (2007): “Pay, Technology and the Cost of Worker Absence”, Economic Inquiry, v45(2), pp268-285. Flabbi, L. and Ichino, A (2001): “Productivity, Seniority and Wages: New Evidence from Personnel Data” Labour Economics, v8, pp.359-387. Green, F. and McIntosh, S. (2001): “The intensification of work in Europe”, Labour Economics, v8, pp.291-308. Greene, W. (1997): Econometric Analysis, 3rd edition. New Jersey, Prentice-Hall. Hassink W. and Koning P. (2009), “Do Financial Bonuses to Employees Reduce Their Absenteeism? Evidence from a Lottery”, Industrial and Labor Relations Review (forthcoming). Heywood J. and Jirjahn U. (2004), “Teams, teamwork and absence.”, The Scandinavian Journal of Economics., v106(4), pp.765-782

28

Heywood J., Jirjahn U and Wei, X. (2006): “Teams, Monitoring, Absence and Productivity.”, German Economic Association of Business Administration, Discussion Paper n°06-01. Ichniowski, C., Shaw, K. and Prennushi, G. (1997): “The effects of human resource management practices on productivity: A study of steel finishing lines”, American Economic Review, v87(3), pp.291-313. Kalleberg, A.L. (2001): “Organizing Flexibility: The Flexible Firm in a New Century”, British Journal of Industrial Relations, v39(4), pp. 479-504. MacDuffie, J.P. (1995): “Human Resource Bundles and manufacturing Performance: Organizational Logic and Flexible Production Systems in the World Auto Industry” Industrial and Labor Relations Review, v48(2), pp. 197-221. Womack, J., Jones, D., and Roos, D. (1990): The machine that changed the world. New-York: Rawson Associates. Wood, S. (1999): “Getting the measure of the Transformed High-Performance Organisation”, British Journal of Industrial Relations, v37(3), pp. 391-417. Wooldridge, J.M. (2002): Econometric Analysis of Cross Section and Panel Data, Cambridge, Mass., MIT Press.

29

Figure I Marginal replacement rates in the Régime Général

In the Régime Général the replacement rate of sick pay depends on the length of the spell of absence. Up to 3 days, no sick pay is payable. From 3-11 days, the replacement rate is 50%. For longer spells, the replacement rate rises to 90%, and entitlement depends on tenure in the job as indicated in the Figure.

Replacement Rate 90%

Extra provision for workers caring for three or more children.

50%

3

11

Years of seniority

<8 8-13 13-18 18-23 23-28 28-33 33+

30 40 50 60 70 80 90

60 80 100 120 140 160 180

30

Days of Absence

Table I: Variable Definitions and Summary Statistics Type25 Definition Mean Variable (std dev.) (842 obs.) C Mean replacement rate for a spell of 4 17.15 τ j (4) days (%) (0.275) C Mean replacement rate for a spell of 11 50.76 τ j (11) days (%) (0.513) C Mean replacement rate for a spell of 20 65.35 τ j (20) days (%) (0.466) Collective Agreement D if collective agreement at branch level 0.917 exists Establishment Size Ct Log size of establishment 4.300 (0.041) Just-in-Time D if just-in-time methods used 0.281 Multidisciplinary D if multidisciplinary work groups are used 0.350 Workgroups Rotation within D if rotation of tasks within workgroups is 0.221 Workgroups practised Rotation between D if rotation of tasks between workgroups 0.387 Workgroups is practised Interdepartmental Co- D if interdepartmental cooperation is 0.631 operation encouraged Shift Work D if shift working used 0.394 Absence of production D if absenteeism of blue-collar production 0.266 workers a problem workers was a problem in 1992 Absence of service D if absenteeism of blue-collar service 0.222 workers a problem workers was a problem in 1992 Absence used in D if absenteeism was a criterion in setting 0.268 setting pay rises individual pay rises Union Representation D if union representation in establishment 0.371 Indicator of social D if absenteeism regarded as indicator of 0.698 climate social climate Absenteeism regularly D if absenteeism regularly monitored 0.770 monitored Mean Age of the C Mean Age of the workforce workforce Proportion of female C Proportion of female workers workers Demand shock D If the demand for the establishment’s product suffered from exogenous shock. Increasing activity D If activity increased in the last 5 years Decreasing activity D If activity decreased in the last 5 years Irregular activity D If activity is irregular besides seasonality Measures to reduce D if measures intended to reduce inventories inventories are used 24

24 25

Variables taken from Réponse in italic C=Continuous ; D=Dummy ; Ct=Count

31

Mean (std dev.) (816 obs) 17.26 (0.283) 51.01 (0.523) 65.68 (0.471) 0.920 4.303 (0.042) 0.279 0.354 0.218 0.387 0.633 0.392 0.262 0.218 0.271 0.375 0.695 0.766 36.11 45.63 0.35 0.51 0.22 0.18 0.52

Table II : Descriptive statistics for establishments above and below median generosity during absence spells of 4, 11 and 20 days duration Variable

Collective Agreement

4 Days Below Above Median Median 0.899** 0.945

11 Days Below Above Median Median 0.883*** 0.950

20 Days Below Above Median Median 0.893*** 0.948

Establishment Size

4.166***

4.483

4.157**

4.121

4.168***

4.449

0.311

0.260

0.306*

0.253

0.311*

0.257

Multidisciplinary Workgroups Rotation within workgroups Rotation between Workgroups Interdepartmental Cooperation Shift Work

0.298***

0. 427

0.328

0.368

0.316**

0.391

0.163***

0.308

0.153***

0.285

0.178***

0.273

0.344***

0. 445

0.337***

0.429

0.352**

0.423

0.599**

0. 677

0.598*

0.660

0.602***

0.664

0.421

0.374

0.408

0.377

0.404

0.377

Absence of production workers a problem Absence of service workers a problem Absence used in setting pay rises Union Representation

0.254

0.283

0.245

0.286

0.241

0.298

0.231

0.204

0.226

0.216

0.235

0.203

0.283

0.241

0.284

0.250

0.281

0.249

0.302***

0.466

0.302***

0.429

0.299***

0.453

0.682

0.722

0.658**

0.738

0.682

0.718

0.746*

0.802

0.724***

0.811

0.740**

0.803

Just-in-Time

Indicator of social climate Absenteeism regularly monitored

Note : difference of means significantly different from zero: * p < 0.1, ** p<0.05, *** p<0,01

32

Table III: Tobit analysis of mean replacement ratios during absence spells of 4, 11 and 20 days duration Exogenous Variables Collective Agreement Log of Establishment Size/1000 Union representation Just-in-Time Rotation within workgroups Rotation between Workgroups Interdepartmental Co-operation Multidisciplinary Workgroups Shift Working Used Absence used in setting pay rises Indicator of social climate Absenteeism regularly monitored Absence of production workers a problem Absence of service workers a problem Industry Dummies: Energy Intermediate goods Equipment Current consumption goods Construction Retailing Transport and Telecommunication Commercial services Housing rental Insurance Other Financial Services Log Likelihood Left-Censored Observations (censored at) Total Observations

4 days 0.60 (0.45) -0.033 (0.877) 1.53 (0.005) -0.93 (0.058) 1.59 (0.015) -0.25 (0.609) -0.16 (0.739) 1.46 (0.005) 0.75 (0.134) -1.03 (0.028) 0.52 (0.323) 0.73 (0.188) -0.61 (0.216) 0.22 (0.700)

11 days 1.43 (0.332) 0.191 (0.630) 3.46 (0.000) -2.02 (0.033) 3.20 (0.004) -0.52 (0.572) -0.37 (0.672) 2.27 (0.012) 0.49 (0.579) -1.20 (0.192) 0.37 (0.709) 1.90 (0.072) -0.36 (0.710) 0.57 (0.576)

20 days 1.90 (0.215) 0.042 (0.918) 4.10 (0.000) -2.57 (0.009) 2.41 (0.027) -0.30 (0.753) 0.18 (0.842) 2.20 (0.016) -0.25 (0.786) -0.56 (0.560) -0.43 (0.672) 1.96 (0.075) -0.59 (0.551) 0.40 (0.701)

-3.63 (0.000) 21.0 (0.000) 11.4 (0.000) 14.6 (0.000) 18.9 (0.000) 7.3 (0.001) 7.1 (0.031) 3.6 (0.079) -3.7 (0.005) 3.4(0.386) 12.3 (0.005) -1583.1 517 12.5 842

-3.64 (0.634) 17.9 (0.000) 9.3 (0.004) 13.9 (0.000) 19.0 (0.000) 5.5 (0.031) 4.5 (0.109) 0.20 (0.931) -2.4 (0.672) 0.49 (0.900) 12.4 (0.000) -3281.2 31 36.36 842

-2.03 (0.818) 15.9 (0.000) 8.5 (0.005) 12.9 (0.000) 17.7 (0.000) 5.4 (0.036) 4.7 (0.082) -0.89 (0.713) -1.11 (0.861) 0.87 (0.829) 12.1 (0.000) -3213.1 29 42.5 842

Note: : significance level for the Student t statistic in parentheses.

33

Table IV.1: Maximum likelihood estimations of Just in Time Choice and of the mean replacement ratio throughout 11 and 20 day spells of absence 11 days* Tobit

20 days Tobit

-3.62 (1.000) 15.09 (0.000) 7.53 (0.052) 11.60 (0.002) 15.44 (0.000) 4.94 (0.186) 3.64 (0.341)

1.28 (1.000) 15.82 (0.000) 7.97 (0.028) 12.35 (0.000) 16.92 (0.000) 5.48 (0.112) 4.30 (0.230)

0.94 (0.801) -1.72 (0.802) 11.94 (0.003) 0.19 (0.579) 1.09 (0.487) 3.06 (0.000) -4.20 (0.000) 2.16 (0.031) -0.22 (0.795) -0.63 (0.434) 1.96 (0.014) 0.58 (0.491) -1.05 (0.214) 0.354 (0.691) 1.67 (0.098) -0.09 (0.917)

-0.14 (0.966) -0.81 (0.765) 12.80 (0.001) 0.10 (0.797) 1.57 (0.322) 4.04 (0.000) -3.79 (0.003) 1.59 (0.148) -0.01 (0.998) -0.11 (0.895) 2.13 (0.019) -0.18 (0.852) -0.57 (0.550) -0.46 (0.657) 1.90 (0.089) -0.21 (0.842)

0.81 (0.413)

0.81 (0.451)

-3581.46 σ 13.94 (0.000) ρ 0.452 (0.000) 816 Total Observations Note : The probit selection equation for the 11-day measure alone is reported.

-3507.813 12.17 (0.000) 0.27 (0.008) 816

Exogenous Variables Mean Age of the workforce Proportion of female workers Demand shock Increasing activity Decreasing activity Irregular activity Measures to reduce inventories Energy Intermediate goods Equipment Current consumption goods Construction Retailing Transports and Telecommunication Commercial services Housing rental Other Financial Services Log of Establishment Size /1000 Collective Agreement Union representation Just-in-Time Rotation within workgroups Rotation between Workgroups Interdepartmental Co-operation Multidisciplinary Workgroups Shift Working Used Absence used in setting pay rises Indicator of social climate Absenteeism regularly monitored Absence of production workers a problem Absence of service workers a problem Log Likelihood

JIT Choice -0.035 (0.000) -0.28 (0.102) 0.017 (0.881) 0.14 (0.253) 0.74 (0.633) -0.13 (0.355) 0.67 (0.000) 0.78 (0.352) 0.59 (0.046) 0.33 (0.265) 0.80 (0.002) -0.38 (0.912) -0.47 (0.062) 0.32 (0.247) -0.19 (0.433) 0.33 (0.965) -1.04 (0.032) 0.085 (0.061)

34

Table IV.2: Maximum likelihood estimations of the effect of the use of Just in Time on the mean replacement ratio throughout a x days spell of absence Change of instruments Measures to reduce Production delays

Measures to reduce Production delays + Measures to reduce inventories

Just-in-Time

σ ρ

Just-in-Time

σ ρ

11 days -3.76 (0.002) 13.75 (0.000) 0.38 (0.001) -4.62 (0.001) 13.72 (0.000) 0.37 (0.000)

20 days -3.36 (0.001) 12.09 (0.000) 0.20 (0.057) -3.55 (0.007) 12.10 (0.000) 0.22 (0.029)

Table V LPM (4-day duration) and OLS (11 and 20 days duration) treatment-effects analysis of mean replacement ratios (excludes interaction effects) Exogenous Variables Collective Agreement Establishment Size/1000 Union representation Just-in-Time Rotation within workgroups Rotation between Workgroups Interdepartmental Co-operation Multidisciplinary Workgroups Shift Working Used Absence used in setting pay rises Indicator of social climate Absenteeism regularly monitored Absence of production workers a problem Absence of service workers a problem

4 days 0.04 (0.411) -0.003 (0.833) 0.11 (0.002) -0.20 (0.087) 0.18 (0.000) -0.04 (0.226) -0.01 (0.723) 0.11 (0.001) 0.04 (0.224) -0.06 (0.049) 0.02 (0.618) 0.06 (0.130) -0.03 (0.388) 0.03 (0.424)

11 days 1.50 (0.401) 0.44 (0.360) 3.97 (0.000) -11.59 (0.000) 3.08 (0.012) -0.01 (0.991) -0.89 (0.374) 2.23 (0.030) 0.87 (0.395) -1.58 (0.140) 0.53 (0.643) 2.28 (0.066) -0.24 (0.830) 1.20 (0.314)

20 days 1.90 (0.240) 0.39 (0.398) 4.21 (0.000) -16.66 (0.000) 1.89 (0.079) 0.38 (0.690) -0.27 (0.759) 1.91 (0.036) -0.06 (0.950) -0.74 (0.434) -0.44 (0.668) 2.07 (0.061) -0.20 (0.841) 1.11 (0.298)

λ ρ σ

0.08 (0.255) 0.21 0.43

6.27 (0.009) 0.45 13.72

9.28 (0.000) 0.70 13.21

Wald test χ 2 (36)

289.93 (0.000)

313.64 (0.000)

309.59 (0.000)

Total Observations

816

816

816

Note: Significance level for the Student t-statistic in parentheses. Dummies for eleven industries are included in both regressions but not reported for space reasons.

35

Table VI.1 LPM (4-day duration) and OLS (11 and 20 days duration) treatment-effects analysis of mean replacement ratios (includes interaction effects) Exogenous Variables Collective Agreement Establishment Size/1000 Union representation Just-in-Time Rotation within workgroups Rotation between Workgroups Interdepartmental Co-operation Multidisciplinary Workgroups Shift Working Used Absence used in setting pay rises Indicator of social climate Absenteeism regularly monitored Absence of production workers a problem Absence of service workers a problem

4 days 0.04 (0.431) 0.00 (0.965) 0.11 (0.001) -0.22 (0.116) 0.25 (0.011) -0.06 (0.339) -0.02 (0.592) 0.14 (0.035) 0.04 (0.279) -0.08 (0.015) 0.04 (0.326) 0.05 (0.166) -0.03 (0.402) 0.03 (0.428)

11 days 1.50 (0.402) 0.48 (0.311) 3.91 (0.000) -12.25 (0.006) 5.72 (0.047) 0.57 (0.776) -1.12 (0.445) 4.56 (0.019) 0.90 (0.381) -1.52 (0.159) 0.69 (0.554) 2.33 (0.061) -0.34 (0.763) 1.10 (0.354)

20 days 1.97 (0.219) 0.41 (0.360) 4.17 (0.000) -17.16 (0.000) 3.88 (0.126) 0.67 (0.710) -0.91 (0.488) 4.01 (0.021) -0.02 (0.985) -0.68 (0.480) -0.34 (0.739) 2.17 (0.049) -0.28 (0.774) 0.98 (0.354)

Just-in-Time* Rotation within workgroups Just-in-Time* Rotation between workgroups Just-in-Time* Interdepartmental Co-operation Just-in-Time* Multidisciplinary Workgroups Rotation within workgroups* Rotation between workgroups Rotation within workgroups* Interdepartmental Co-operation Rotation within workgroups* Multidisciplinary Workgroups Rotation between workgroups* Interdepartmental Co-operation Rotation between workgroups* Multidisciplinary Workgroups Interdepartmental Co-operation* Multidisciplinary Workgroups

-0.08 (0.330) 0.04 (0.565) 0.02 (0.752) 0.09 (0.208) -0.14 (0.094)

-1.50 (0.548) 1.88 (0.403) 0.65 (0.771) 1.01 (0.637) -5.25 (0.038)

-1.41 (0.516) 2.18 (0.264) 0.88 (0.651) -0.20 (0.916) -6.00 (0.007)

0.09 (0.318)

3.44 (0.206)

4.44 (0.061)

-0.07 (0.437)

-2.61 (0.299)

-1.78 (0.421)

0.06 (0.412)

0.48 (0.826)

0.71 (0.711)

-0.00 (0.954)

-1.16 (0.610)

-0.60 (0.764)

-0.05 (0.439)

-2.38 (0.267)

-2.08 (0.275)

λ ρ σ

0.08 (0.281) 0.19 0.44

5.89 (0.015) 0.43 13.56

8.92 (0.000) 0.68 13.01

Wald test χ 2 (46)

299.19 (0.000)

328.63 (0.000)

329.72 (0.000)

Total Observations

816

816

816

Note: Significance level for the Student t-statistic in parentheses. Dummies for eleven industries are included in this regression but not reported for space reasons.

36

Table VI.2 Total effects of independent variables in interaction on the mean replacement ratio Exogenous Variables

4 days (OLS)

4 days (LPM)

11 days

20 days

Just-in-Time

-5.85 (0.009)

-0.17 (0.162)

-11.10 (0.006)

-16.15 (0.000)

Rotation within workgroups

2.03 (0.012)

0.21 (0.000)

4.63 (0.001)

3.45 (0.007)

Rotation between Workgroups

0.03 (0.958)

-0.04 (0.185)

-0.16 (0.885)

0.20 (0.834)

Interdepartmental Co-operation

-0.56 (0.317)

0.01 (0.870)

-0.83 (0.406)

-0.14 (0.872)

Multidisciplinary Workgroups

1.10 (0.056)

0.11 (0.001)

2.33 (0.024)

2.03 (0.028)

Note: significance level for the χ 2 statistic in parentheses.

Table VII: Correlation Matrix between production methods and organizational variables

Just-in-Time

JustinTime 1

Rotation Rotation Interdepartmental Multidisciplinary Shift within between Cooperation Workgroups Work Workgroups Workgroups

Rotation within workgroups

0.1303

1

Rotation between Workgroups

0.1099

0.3518

1

Interdepartmental -0.007 Cooperation

0.1196

0.1814

1

Multidisciplinary Workgroups

0.1417

0.1068

-0.0138

0.1132

1

Shift Work

0.0868

0.1661

0.1664

0.0812

0.0126

37

1