Abstract
We postulate that the production losses from absence depend on firms’ ability to internally substitute for absent workers, incentivizing firms to keep absence low in jobs with few substitutes. Using Swedish employer–employee data we show that absence is substantially lower in such positions conditional on establishment and occupation fixed effects. The result is driven by employee adjustments of absence to substitutability, and sorting of low (high) absence workers into (out of) positions with few substitutes. These findings highlight that internal substitution insures firms against production disruptions and that absence costs are important aspects of firms’ hiring and separation decisions.
I. Introduction
The high sickness absence rates in many countries have sparked a growing interest in the determinants and consequences of sick leave. Sick leave is the quantitatively most important source of work absence, generating large costs for affected individuals, firms, and for the society as a whole.
Economists have mainly focused on how financial incentives provided by the sickness insurance system affect absence. On the employee side, several studies show that sickness absence is positively related to the replacement rate (Johansson and Palme 1996, 2002, 2005; Henrekson and Persson 2004). There is more limited evidence on the employer side, but evidence from Norway and Sweden suggests that employee absence varies negatively with employers’ sick pay responsibility (Fevang, Markussen, and Røed 2014; ISF 2015).1
This paper focuses on how firms’ absence-related costs affect employee sickness absenteeism. We use data from Sweden where it is argued that firms have little incentives to prevent sickness absence since they only pay for absence during the first two weeks (Hägglund and Johansson 2016). This argument, however, ignores that the absence cost of one worker may be larger than the absent worker’s own output. If there is high complementarity between workers in the production process, such disruption costs can be substantial.
Previous research has shown that absence is more common in larger firms (Barmby and Stephen 2000; Dionne and Dostie 2007; Ose 2005; Lindgren 2012). Because small firms cannot insure against the risk of absence it is argued that these have stronger incentives to prevent sick leave. Without variation in firms’ absence-related costs it is, however, difficult to assess the relevance of this argument in relation to other differences related to firm size.
We hypothesize that firms’ costs of absence depend on the number of employees that can do the same job. Thus, firms should try to keep absence low in jobs with few substitutes. Using linked employer–employee data for a large representative sample of Swedish private sector firms for the period 1997–2010, we find a substantial difference in absence rates between more and less substitutable employees within narrowly defined job cells. This pattern holds irrespectively if we look at employees’ own sickness absence or absence among parents caused by child sickness.
We use several analyses to probe the mechanisms behind our results. In short, firms can respond to the extra absence-related cost in key positions by (i) being more selective in the hiring decision, (ii) providing stronger incentives to exert effort, or by (iii) increasing hiring rates. Our results show that workers hired into more unique jobs have lower pre-hire sickness absence and higher wages than other new hires. But workers also change their absenteeism in response to the number of substitutes. Finally, employees with high realized sickness absence have higher separation rates in unique positions, an effect that seems more important when there was less information about the absence type of the worker prior to hiring.
Together these findings support the personnel economics ideas that firms use both hiring and incentives to overcome problems with sickness absence, as highlighted in Oyer and Schaefer (2011).
II. The Institutional Setting
Sweden has an obligatory, general, and uniform sickness insurance system. Benefits are income related, and the replacement rate is almost 80 percent of the labor income up to a cap.2 The sickness absence rate is high in Sweden compared to many other countries (Hägglund and Johansson 2016). The take-up of sick leave benefits from the Swedish Social Insurance Agency (SSIA) varies between 10 and 16 percent over the period we study.3
Employers pay the benefits for the first 14 days of sick leave. Thereafter, the SSIA pay the benefits. Hence, employers’ sick pay responsibility is relatively moderate by international standards (see Engström and Johansson 2012 for a detailed description of the institutions).
In addition, Swedish parents with children aged zero to ten can receive Temporary Parental Benefits to be absent from work to care for children that are too sick to be in school or in daycare.4 Each parent is entitled to 120 days annually per child paid by the SSIA and the replacement rate is 77.6 percent. Following parents in Sweden who had their first child in 1994, Boye (2015) shows that the average woman (man) is absent from work for 5 (2.5) days per year during the first 10 years of the child’s life, with higher absence rates for children in daycare ages.
III. Firms’ Responses to Employee Absence
This section describes the three main strategies a firm can use to handle the risk of absence:
The firm can look for candidates with low absence risk by screening harder or by increasing wages to attract employees with relatively higher absence costs. Theoretical and empirical evidence suggests that firms engage in active search using public or private signals to find high productivity workers (Oyer and Schaefer 2011; Rees 1966) and that this type of active search allows firms to find better candidates with higher match quality (Hensvik and Nordström Skans 2016; Dustmann et al. 2016).
The firm can hire any worker, but adjust the incentives for workers to induce greater effort if they receive a high-absence worker, for example, by increasing the wage to reduce absence. There is plenty of evidence supporting that financial incentives do change behavior within organizations (Bandiera, Barankay, and Rasul 2009; Lazear 2000). Firms can also provide incentives in ways other than adjusting the wage, such as linking absence rates to promotions or separations.5
The firm can hire more workers—that is, adjust the number of substitutes to the absence risk.
These mechanisms all predict a positive relationship between sickness absence and the number of substitutes in the cross-section, and we will not be able to perfectly distinguish among these. However, if firms primarily become more selective in the hiring decision when they fill unique positions, we expect most of the relationship to disappear once we account for selection into firms (that is when we control for worker fixed effects). If, on the other hand, the relationship remains within ongoing employment relationships, this is consistent with a behavioral mechanism induced by the firms.6
The strategies described above also illustrate that firms may respond endogenously to the absence type of workers by hiring more internal substitutes. Given the difficulties of observing absence types ex ante and the relatively strong employment protection laws in Sweden, we believe that this is a minor concern for new hires, since uncertainty about the absence behavior introduces the risk of excess hiring.7 If the absence type is observed by the firm (as in ongoing employment relationships), we expect that firms will increase the number of substitutes when the cost of hiring is low relative to the production losses incurred in the absence of substitutes. Our empirical strategy will to some extent handle firms’ hiring response to absence types by accounting for worker by job fixed effects. As we will see, these results are consistent with the notion that the number of substitutes changes absence, rather than the other way around.
IV. Context and Data
A. Definitions and Measurements
We use Swedish register data covering 1997–2007. The basis of our analysis is the Wage and Salary Structure Data (WSSD), which contain information on wages and occupations as well as worker and establishment identifiers. The data are collected by Statistics Sweden and cover a large representative sample of establishments in the private sector and all public establishments. In the main analysis we restrict the sample to jobs in the private sector because establishments are more properly defined there.8
The sample of private establishments is stratified by firm size and industry, where establishments within large firms are overrepresented. In the Appendix we show the distribution of establishments and employees with respect to establishment size. The fraction of employees is fairly evenly distributed across the different size groups. In total, the data cover almost 50 percent of private sector workers.
1. Measuring internal substitutability
We define employee substitutability as the number of other workers within the same combination of establishment and occupation (ISCO-88, three-digit level) in a given year.9 For example, an administrator at an establishment that in total employs four administrators will have three substitutes. In most specifications, we define low internal substitutability by an indicator for having zero to five substitutes, but we also show results from more flexible models. There are 107 different occupations in our data at the three-digit level.10 In order to focus on regular workers, we drop employees in managerial positions. We also drop employees at very small establishments (less than three employees).
2. Measuring sickness absence
We add demographic information from the Longitudinal Database about Education, Income and Employment (LOUISE). The data cover the full population aged 16–64 and include variables such as age, gender and education. There is also information about sickness absence benefits received from the SSIA. The data include all spells for which the individual was entitled to sickness benefits from the SSIA. Since spells shorter than two weeks are paid by the employers, these short spells are not available in our data. Sickness absence will be defined as an indicator for having at least one spell longer than two weeks in a given year. The fact that we cannot observe shorter spells is a limitation, and we will therefore complement our analysis with absence due to care leave for sick children. Since the Temporary Parental Benefits are paid by the SSIA from day one, these data also pick up short-term absence spells.
3. Defining hires and pre-hire and realized absence
We examine worker sorting in more detail using a dataset consisting of new hires. We identify new hires using an additional data source, Register Based Labor Market Statistics (RAMS). This register contains information that employers need to supply to the tax authorities each year, which implies that all employment relationships in Sweden are included.
We define new hires as employees observed in an establishment in a given year, but not in the same establishment or in the same firm in any of the five preceding years. For each hire, we define their pre-hire sickness absence as the average incidence of having at least one sickness absence spell (longer than two weeks) per year in the three years prior to employment. In order for all new hires to have at least three pre-hire years, we focus on workers with at least four years of labor market experience.11 We will also examine the probability of job separation when the worker absence type is revealed. To this end we define realized absence as the average sickness absence probability in the hiring year and the year after entry.
B. Descriptive Statistics
Table 1 contains sample statistics. There are six million observations in the full sample, and about 20 percent of these are found in jobs with zero to five substitutes. Positions with low employee substitutability thus account for a significant share of the labor market. About 4 percent of the workers have truly unique jobs (0 substitutes), and 11 percent of the employees have at least one sickness absence spell that is longer than two weeks in a given year.12 Appendix Table A1 shows that it is more common to have few substitutes in administrative and data maintenance jobs. This occupational distribution looks relatively similar across establishments of different size, although, not surprisingly, workers in smaller establishments are much more likely to be unique.
Descriptive Statistics for All Employees
Consistent with our hypothesis, sickness absence is lower for workers with relatively few substitutes (0–5), but these workers differ in other aspects as well. For example, they are employed in smaller establishments and in more skilled professions suggesting that they have key positions within the firms.13 In addition, workers in relatively unique positions are older and more often women, although education levels appear similar to other employees. The wage is lower for workers with few substitutes, which is related to the fact that they tend to work in smaller establishments that generally pay lower wages. Conditional on our empirical model, there is a statistically significant wage premium for unique workers. In Appendix Table A2 we show the same descriptive statistics for the new-hire sample. Overall, it is very much in line with the full sample. Importantly, positions without substitutes are present in all occupational skill levels.
V. Empirical Strategy and Findings
A. Empirical Specification
We start by exploring the association between present sickness absence and the number of substitutes by estimating Equation 1 by OLS:
(1)
where Aijpt is the incidence of sickness absence for worker i in establishment j and profession p in year t. Sjpt measures employee substitutability within each establishment and three-digit occupation.14 αj and αp are establishment and occupation fixed effects, respectively. We also include year fixed effects, θt, to account for, for example, business cycle swings potentially correlated with firms’ organization of work and individual sickness absence. The worker characteristics Xit are gender, age, education, country of origin, and an indicator for having children under the age of three.15 Finally we include establishment size Zjt. ϵijpt is the error term.
The parameter of interest is γ, which aims to capture the relationship between the number of internal substitutes and work absence.16 We also wish to disentangle to what extent γ captures behavioral responses and/or employee selection on the entry and exit margin. As a first step we therefore add worker fixed effects to Equation 1, which means that we account for the selection of workers over jobs with few/many substitutes. This analysis requires repeated observations per employee, which is complicated by the fact that our data are based on a (random) sample of the small firms each year. As an alternative strategy, we therefore estimate Equation 1 separately for new hires, replacing the outcome with the pre-hire sickness absence. Since pre-hire sickness absence is potentially correlated with past employment, we also control for the employment probability in the same time period. Finally, we examine the separation response to realized sickness absence among new hires by estimating the following equation:
(2)
where Separationijpt+2 is an indicator for if worker i hired in year t left establishment j between year t + 1 and t + 2, and Āijpt is the realized absence of entrant i measured as the averaged incidence of absence over the entry year (year t) and the first year into the employment spell (t + 1). The aim of φ is to capture the separation response to the realized absence behavior among newly hired workers. To examine whether this response depends on the internal substitutability of employees we also estimate versions where the model in Equation 2 is fully interacted with employee substitutability.
B. Baseline Results: Employee Substitutes and Absence
Figure 1 shows the estimates from Equation 1 when we include dummy variables for having up to five substitutes (employees with more than five substitutes is the omitted category). The estimates are statistically negative at the 1 percent level, ranging between 1 and 2 percentage points. Hence, workers with few close substitutes have lower absence rates. The estimates become smaller as the number of employees performing the same job increases, which is consistent with the idea that the costs of employee absence, in terms of production disruptions, increase as the possibilities of internal substitution decrease.
Sickness Absence and Internal Substitutability
Notes: The figure shows the estimated coefficients on dummies for 0–5 substitutes in Equation 1 (with 95 percent confidence intervals). The reference category is employees with more than five substitutes and the background controls are gender, age, education, birth country, having children aged 0–3, and establishment size. The model also includes year, occupational, and establishment fixed effects. Standard errors are clustered on the establishment level.
Table 2 shows the estimates when we use an indicator for low substitutability, defined as having zero to five substitutes. These results confirm the strong negative correlation between low internal substitutability and work absence.
Sickness Absence and Internal Substitutability
C. Evidence from Child Sick Spells
Our data are limited to absence spells longer than two weeks. To test whether our results also extend to shorter absence spells we include an alternative absence measure: Care leave for sick children, which also includes short-term spells.17
We restrict this analysis to individuals with at least one child aged zero to ten and use an indicator for having received Temporary Parental Benefits in a given year as the outcome. To see if substitutability at work affects the division of care for sick children within the family we replace own absence with the corresponding absence measure for the partner.
Table 3 presents the results using the model described by Equation 1. The estimate in the first column suggests that workers are significantly less likely to be absent to care for sick children in jobs with few substitutes. Thus, the results are in line with our general findings in Table 2, although smaller in magnitude. The weaker relationship may reflect that most parents have at least one child sickness spell in a year, which is likely to make the extensive margin less relevant. This is confirmed by the results in Table 4, where we show that the effect on the intensive margin is relatively more important in this case. Interestingly, the partners of employees with few substitutes are more likely to be home caring for sick children (see the second column), and the magnitude of the estimate is almost equal to the estimate in the first column.18 Thus, children of workers with few substitutes are no less sick than other children; instead these workers seem to avoid work interruptions by shifting absence to their partners.
Evidence from Child Sick Spells
Absence on the Extensive and Intensive Margin
D. Robustness Checks
So far we have looked at absence on the extensive margin. In Table 4 we use the log of annual sickness benefits as the outcome, which picks up the length and number of absence spells. Employees with fewer substitutes have 2 percent fewer absence days (Panel A) and almost 4 percent fewer care leave days due to child sickness (Panel B) than employees with more substitutes.19 The results thus seem robust to variations in the way we measure sickness absence.
Appendix B presents further robustness checks. Here we show that the baseline result is robust to including public sector employees as well and to using an even more detailed level of the occupation code when defining the number of employee substitutes. In addition, we provide a longer discussion of the relationship between the number of substitutes and establishment size, and we show that our results are robust to various definitions of the variable Low substitutability.
E. Behavioral vs. Entry Responses
As discussed in Section III, the relationship between sickness absence and substitutability may both reflect a selection effect (systematic sorting of certain workers into and out of jobs with few substitutes) and a behavioral effect (workers adjusting their absence behavior to their substitutability at work). To examine the relevance of these two explanations we exploit variation in the number of substitutes for the same worker over time by adding worker fixed effects to the baseline specification. The estimate presented in the second column of Table 5 is roughly halved compared to the baseline estimate in the first column but remains significantly negative on the 1 percent level. Hence, workers do seem to adjust their work absence depending on the number of substitutes.
Behavior vs. Sorting into Jobs
About 60 percent of the within worker variation in the variable Low substitutability comes from changes in the number of substitutes due to job changes, and 40 percent comes from changes in the number of substitutes within a given job. In the third column of Table 5 we focus on the variation within jobs by replacing the worker fixed effects with a fixed effect for each combination of a worker and a job (establishment × occupation). The relationship persists, which further strengthens the interpretation that workers in a given job decrease their absence as a response to a reduction in the number of substitutes.20
In the fourth column of Table 5 we replace present sickness absence with the pre-hire sickness absence, described in Section IV, in a sample of new hires. Consistent with our earlier results, this estimate clearly suggests that workers hired into positions with fewer substitutes are more likely to be low-absence types. Reassuringly, this estimate is very similar to the difference between the estimates with and without worker fixed effects in the full sample, which supports the interpretation that workers with few absence spells sort into jobs with low substitutability. Overall, these results suggest that the correlation between substitutability and sickness absence entails both a selection and a behavioral component that appear to be of similar importance.
1. The role of wages
The strong association between sickness absence and substitutability raises the question whether wages differ between more and less substitutable jobs. Table 6 suggests that this is indeed the case. We obtain these estimates by replacing sickness absence as the outcome in Equation 1 with the log of the monthly full-time wage.21 The results in the first column suggest that employees with zero to five substitutes have 1.3 percent higher wages than employees with more than five substitutes, conditional on establishment and occupation fixed effects.
Substitutability and Wages
The wage premium is consistent with the notion that firms can use higher wages to attract workers with lower absence risk or to incentivize workers to increase their effort. Another, competing explanation is that unique jobs are more productive per se. In the third and fourth column we look at the entry wages of new hires, as well as wage responses to changes in the number of substitutes within an ongoing employment relationship (that is, within the worker × job match). These results suggest that entry wages indeed are higher for unique jobs. However, based on the very small and precise estimate in the fourth column, there is no evidence that firms raise wages as a response to a reduction in the number of substitutes or that unique jobs are more productive in general.
2. The role of separations
Table 7 further explores whether the realized sickness absence among new hires (measured as the average sickness absence probability in t and t + 1) affects (i) the probability of exiting the employment relationship and (ii) the probability of receiving more substitutes within three years after entry. We study the first question by estimating Equation 2 using a sample of new hires that stay in the establishment for at least one year. The outcome is an indicator for exiting the establishment between t + 1 and t + 2 (Panel A).22 We study the second question using a sample of new hires that are observed in the establishment at least until t + 3. The outcome is an indicator for having more substitutes in t + 3 than in t (Panel B).
Realized Sickness Absence and Post-Hire Outcomes
The results suggest that higher realized sickness absence generally is associated with significantly higher turnover rates (Panel A, first column) and a higher likelihood of receiving more substitutes (Panel B, first column).23 This relationship is particularly strong for workers employed in jobs with low internal substitutability (Columns 2–4), suggesting that sorting on the basis of sickness absence also occurs via the exit margin.
F. The Role of Information
Finally, we examine the role of information in the hiring stage for the selection into and out of jobs. The data from the SSIA are not publicly available to employers (but are available for research purposes). Hence, it is reasonable to believe that it is more difficult for employers to observe the pre-hire absence behavior of new hires compared to formal credentials. We use three different proxies for the amount of information about the employees in the matching stage: (i) an indicator for being employed in t – 1 to t – 3 (Pre-hire employment), (ii) an indicator for previous employment in another establishment within the same firm (Firm connection), and (iii) an indicator for previous employment in the same establishment as an incumbent employee (Coworker connection).24
Although these measures are far from perfect, several studies support our choice of information proxies. Eriksson and Rooth (2014) show that employers are reluctant to hire people from nonemployment, which indicates that nonemployment is associated with uncertainty about worker type. Thus, we find it reasonable to expect that there is more information about workers with a strong attachment to the labor market.25 Schönberg (2007) further shows that hard-to-observe characteristics of college graduates are more easily assessed by the current firm than by outside firms. We therefore expect that matches involving workers with an earlier connection to the recruiting firm are based on better information about the worker absence type. Finally, there is recent evidence that incumbent employees can provide valuable information about the productivity of prospective hires with whom they have worked in the past (Dustmann et al. 2016; Hensvik and Nordström Skans 2016). Based on these findings, we assume that firms can make better predictions about the absence type of former coworkers to current employees.
This is also what we see in the first row of Panel A of Table 8: hires have between one and two percentage points lower pre-hire absence for two out of the three information measures. Importantly, the negative relationship between information availability and pre-hire sickness appears about twice as strong in jobs with few substitutes as in jobs with many substitutes. Thus, when employers are recruiting to positions with few internal substitutes they react even stronger to information about worker absence type.
The Role of Information
Panel B shows how job separations induced by realized sickness absence are related to the information available in the hiring stage. Intuitively, separations should respond more to realized sickness absence if there was less information about absence type beforehand. For simplicity, we restrict this analysis to jobs with few substitutes (≤ 5) and interact realized absence with our information proxies. Consistent with the results in Panel A of Table 7, there is a strong relationship between realized sickness absence and the probability of job separation. However, this relationship is weaker when the match was based on more precise information (suggested by the interaction terms). Depending on the information proxy, the point estimates are between 4 and 7 percentage points lower when there was more information, although the difference is not statistically significant when we use pre-hire employment as the information proxy (see the first column). In sum, these findings suggest that matches formed with less precise information are more likely to be affected by revelations of worker absence type.
VI. Conclusions
We document that workers matched to jobs with few internal substitutes are significantly less absent from work, compared to other workers in the same narrowly defined occupations. The difference is substantial and holds irrespectively if we look at employees’ own sickness absence or absence among parents caused by child sickness. Parents in jobs with lower substitutability shift part of their child sick absence spells to their partners.
About half of the correlation remains when we account for worker fixed effects, suggesting that both sorting based on pre-hire absence types and on-the-job changes in absence behavior are important mechanisms behind the strong association between sickness absence and substitutability. But, sorting also occurs via the exit margin: recent hires with high realized absence are more likely to part with the job, particularly if they have few substitutes.
Overall, our results highlight the importance of internal labor markets for firms to handle the costs of production disruptions caused by work absence. For jobs with low internal substitutability, sickness absence is a significant determinant in the selection process of new workers, but the difficulties of perfectly predicting the absence type of new employees also seem to call for adjustments via the incentive structure and through job separations. Our findings thus validate previous theoretical and empirical work on the importance of sorting and point at sickness absence as a previously unexplored dimension of worker mobility. It should be noted that we document these effects in Sweden, which has a fairly generous sickness benefit system and high absence rates compared to many other countries. In this respect, we could expect that firms have particularly strong incentives, and perhaps also possibilities, to influence the sick leave behavior.
From the workers’ perspective, our findings suggest that episodes of sickness absence affect the chances of accessing and retaining unique positions, which account for a significant share of the labor market. Hence, workers have strong incentives to keep absence low in jobs with low internal substitutability, which they do, for example, by shifting child care to their partners. In future work it would be valuable to further explore the interplay between job characteristics and the allocation of time within the household. Studying this question could potentially enhance our understanding of the systematic gender pay differences in modern labor markets.
Appendix
A. Additional Descriptive Statistics
Occupations with Few Substitutes
Descriptive Statistics for New Hires
B. Additional Robustness Checks
Tables B1–B3 presents a number of robustness checks. For comparison, the first column of Table B1 repeats the baseline estimate from the second column of Table 2. In the second column, we add the public sector employees to our sample. Reassuringly, the relationship between the number of substitutes and sickness absence holds in the full economy. In the third and fourth column we estimate the model using the two- and four-digit level of occupations instead.26 The results are almost identical to the results using the three-digit level. Hence, the interaction between two-digit occupations and establishments seems to quite distinctly identify jobs, and that further disaggregation provides little additional information about work content.
Robustness Checks
Descriptive Statistics with Respect to Establishment Size
Regressions Using Alternative Definitions of Employee Substitutability
Our baseline measure of substitutability is the number of substitutes in the same occupation, but it is possible that the substitutability of workers could interact with the size and organization of the workplace. We may overstate the degree of substitutability in large establishments if employees are organized in different departments that make substitution difficult. But, more coworkers could also imply that employees are more substitutable as there is a higher likelihood that some workers have overlapping skill sets even though they occupy different jobs. It is therefore not clear how to (and if we should) adjust the number of substitutes to establishment size. Column 4 of Table B2 shows how much of the identifying variation in the variable Low substitutability that comes from establishments of different sizes. The figures are based on the squared residuals from a regression of Sjpt on the full covariate set in Equation 1. It is clear that small to medium establishments account for a large share of the variation: 41 percent comes from establishments with 3–49 employees, and 40 percent comes from establishments with 50–249 employees. To test how relevant our results are for establishments of different sizes we therefore reestimate the baseline model separately for those with 3–49, 50–249, 250–500, and more than 500 employees.27 The estimates from this exercise are plotted in Figure B1. All four estimates are significantly negative on the 1 percent level, and the magnitudes of the estimates are roughly similar to the estimate presented in Column 2 of Table 2. The fact that our measure of substitutability is relevant for both small and large establishments also speaks against supply-side responses as the main mechanism behind our results. If unique employees internalize the firm’s cost of absence, it is reasonable to expect the relationship to be stronger in smaller establishments.
Sickness Absence, Substitutability, and Workplace Size
Notes: The figure shows the results from separate estimations of Equation 1 by establishment size (coefficient on the 0–5 substitutes dummy variable and 95 percent confidence interval). The standard errors are clustered on the establishment level. The reference category is employees with more than five substitutes. The background controls are gender, age, education, birth country, having children aged 0–3, and establishment size. The model also includes year fixed effects, occupational fixed effects, and establishment fixed effects.
For completeness we have also tested two other definitions of low substitutability on the basis of the logic that employees are less substitutable in larger establishments (for a given number of substitutes). First, we define low substitutability when one of the following criteria is fulfilled: (i) no substitutes in establishments with 3–49 employees, (ii) less than four substitutes in establishments with 50–249 employees (iii), less than seven substitutes in establishments with 250–500 employees, or (iv) less than ten substitutes in establishments with more than 500 employees. The second alternative definition is based on the number of substitutes divided by establishment size and defines low substitutability when this share is below 0.03 (which corresponds to the tenth percentile). These definitions are of course arbitrary but offer a way of relating the notion of substitutability to the overall size of the establishment (Columns 5–6 of Table B2 show that more of the identifying variation now comes from larger establishments). However, when we reestimate Equation 1 using these two alternatives we obtain virtually identical estimates (see Appendix Table B3). Column 4 also shows the linear relationship between the share of substitutes and absence, which suggests that a standard deviation increase in the share of substitutes is associated with an 0.8 percentage point higher probability of absence (0.3 × 2.7). Overall, we conclude that the link between employee substitutability and absence is relevant for establishments of all sizes (rather than only relatively small ones) and that our results are robust to different variations in the definition of internal substitutability.
Footnotes
↵1. With the term sick pay responsibility we mean that the firm has to continue paying the worker during sickness absence.
↵2. In 2003, the midst of our period, the cap amounted to 289,500 SEK (31,500 USD), which roughly corresponded to earnings in the 80th percentile of the income distribution.
↵3. Calculated as the fraction of the population aged 20–64 who received sick leave benefits during a year (see Hägglund and Johansson 2016, for a more detailed description of the variation in take-up during our study period).
↵4. 90 percent of all parents in Sweden have their children between 3–6 years of age in subsidized child care (Mörk, Sjögren, and Svaleryd 2013).
↵5. We thank an anonymous referee for pointing out these channels.
↵6. It is of course possible that an observed relationship could also be driven by supply-side responses. If workers internalize the firms’ costs of absence, low absence types might select into firms with high absence-related costs or change their absence behavior if they enter a job with few substitutes. We hypothesize that such concerns are particularly likely to arise in small firms, where the productivity loss from one person’s absence is large relative to total firm productivity, and where individual absence is more visible to the firm. Our results, however, are very similar for small and large firms.
↵7. That is, hiring beyond the point where the marginal cost of hiring an additional worker is greater than the marginal benefit.
↵8. In the public sector, all individuals that are employed by the same municipality are sometimes registered as belonging to the same establishment.
↵9. Jäger (2016) shows that when an employee exits (due to death) firms increase their demand for the remaining workers in the same occupations, but not in other occupations as the deceased. This clearly indicates that firms regard employees within the same narrowly defined occupations as closer substitutes than employees in other occupations.
↵10. Some examples of three-digit occupations are: shop and stall salespersons and demonstrators, client information clerks, computer associate professionals, and office secretaries and data entry operators.
↵11. In other words, labor market entrants with less than four years since they graduated from their highest education are excluded.
↵12. The figure on sickness absence is confirmed by estimates from Statistics Sweden (2007).
↵13. The summary statistics show the distribution of workers/hires across a broader set of occupations (one-digit level). When defining the number of substitutes we use more detailed occupation codes (three-digit level).
↵14. In our main specifications, low substitutability is defined as jobs with zero to five substitutes in the same occupation and establishment. Sometimes we also use models with slightly different definitions, which we then state clearly.
↵15. We group individuals by their country of origin into the following six categories: Sweden, the rest of the Nordic countries, the rest of Europe, North America, South America, and the rest of the world.
↵16. The baseline analysis focuses on sickness absence on the extensive margin. As a robustness check we also consider outcomes that capture the intensive margin.
↵17. The reason is that parents receive benefits from the Social Insurance System from day one.
↵18. The estimate in the first column is almost identical if we use the same sample as in the second column (that is employees with cohabitating partners).
↵19. The received benefits are closely related to the number of leave days.
↵20. Using worker × establishment fixed effects produces a very similar relationship (−0.0063 [0.0010]).
↵21. The wage is the wage the employee had during the sampling week expressed in full-time monthly equivalents. The variable includes all fixed wage components, including piece-rate and performance pay as well as fringe benefits. However, overtime pay and paid leave are not included. The monthly wage is adjusted to fulltime for part-time workers by Statistics Sweden. For the blue-collar workers the wage is typically obtained by the hourly pay rate times the number of hours that correspond to full-time employment. For the white-collar workers it reflects the September wage adjusted by the share of part-time work during the same month.
↵22. As a robustness check we have also used an indicator for not being observed in the establishment in either t + 2 or t + 3. This does not substantially change the results.
↵23. Interestingly, when we condition on being observed in t + 1 and t + 2 and use the average sickness absence probability in t + 1 and t + 2 as an explaining variable for leaving the establishment in t + 3 the estimate in Panel A, Column 1, is substantially lower. This is consistent with the notion that the marginal effect of exhibiting bad properties (in this case high sickness absence), in relation to the job, on job separation probability should decrease with tenure (see Kwon 2005 for an interesting contribution on this topic).
↵24. When we use the previous firm connection as the information proxy we relax the new-hire definition and include new hires on the workplace with a history within the firm.
↵25. Farber and Gibbons (1996), Altonji (2005), and Lange (2007) show that employers overprice formal credentials (and underprice hidden talents) among inexperienced workers, which further supports that there is less information about worker type for employees with weaker labor market experience.
↵26. The four-digit information is only available for the years 2005–2007.
↵27. The division is based on a classification that Statistics Sweden uses when they collect data from firms.
- Received May 2016.
- Accepted June 2017.
