A financial bonus as a solution for absenteeism

Amsterdam School of Economics July 2017

Master Thesis

A financial bonus as a solution for absenteeism

Program: Business Economics - Managerial Economics and Strategy Supervisor: Prof. Dr. Randolph Sloof

ECTS: 15

ABSTRACT

The purpose of this thesis is to examine the effect of an attendance-bonus on workplace absenteeism. Data for this study was collected at organization X. This organization introduced a bonus for employees that are not absent for a certain period of time. The dataset consists of an unbalanced panel of 177 employees. The results show that the observed decrease in absence rate is not due to the incentive effect of the bonus. However, the results suggest that there is a sorting effect that is mainly coming from the fact that employees, who quitted before the introduction of the bonus, have relatively high absence rates compared to employees that have worked in both regimes and new hires. Secondly, the bonus does not significantly influence the frequency of absence. The negative effect disappeared when both control variables and employee-fixed effects are included. The average length of absence is not influenced by the bonus at all.

Statement of Originality

This document is written by Student Anne van der Klugt, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Acknowledgements

After five moths, I finally finished my Master Thesis. There are some people I really would like to thank because without them, this was not possible.

First, I would like to thank Hetty van Griensven-Rothengatter and Marloes van Winkel for the fine cooperation and great enthusiasm. You have helped me a lot by sharing all your data with me. Thank you for making time for me and providing me with all documents and information I needed to make a valuable dataset.

I also would like to thank my supervisor Prof. Dr. Randolph Sloof for helping me with all my questions about my Master Thesis. I really appreciated the quick answers to my e-mails and the feedback you gave me.

TABLE OF CONTENTS

1. Introduction 5

2. Literature review 6

2.1 Demographic determinants of absenteeism 7

2.1.1 Age and tenure 7

2.1.2 Gender 8

2.1.3 Education 9

2.2 Employment protection 10

2.3 Cross-country differences in absenteeism 11

2.4 Instruments to reduce absenteeism 12

2.4.1 Negative incentive programs 12

2.4.2 Positive incentive programs 14

2.4.3 The effect of an attendance-bonus on absenteeism 16

2.4.4 Crowding out effect of monetary incentives 17

3. Hypotheses 20

4. Data & methodology 22

4.1 Organization and context 22

4.2 Descriptive statistics 29

4.3 Empirical specification 36

5. Results 37

5.1 The effect of the bonus on the absence rate 37

5.2 The effect of the bonus on the average length of absence 43

5.3 The effect of the bonus on the absence frequency 46

5.4 The effect of employee protection on the absence rate 50

5.5 The sorting and incentive effect 51

6. Discussion & conclusion 55

REFERENCES 59

1. Introduction

Absenteeism can be defined in different ways. The most common one is the non-attendance of employees for scheduled work (Gibson, 1966). In addition to that, Ng and Feldman (2008) make a distinction between involuntary and voluntary absence. Involuntary absence means sickness absenteeism. By voluntary absence the researchers mean non-sickness related absenteeism, like non-medically absences and withdrawal behavior. Hackett (1990) made a similar distinction but referred to it as avoidable and unavoidable absences. Furthermore, Mercer (2008) defined four types of absence: unplanned incidental absences, planned absences, extended absences and unplanned incidental and extended absences.

Because absenteeism is a serious problem for firms, it has gained increasing attention of researchers. For years, the absence of workers cost organizations significant amounts of money (Mercer, 2008). According to Mercer (2008), absence of workers cost firms on average 35.8% of their workers average annual salary. Robbins and Judge (2012) state that absenteeism is not only disruptive, but it can also lead to a reduction in the quality of the output or even cause bankruptcy. They emphasize that high absence rates can decrease the effectiveness and efficiency of an organization.

Most literature written about absenteeism is about the effects of individual factors. In fact, not many of them suggest solutions for this problem. However, organizational variables, like reward systems, are far better predictors for absence rates than individual factors (Bartol, 1979; Wanous, Stumpf & Bedrosian, 1979). Therefore, an important question is how firms can use financial incentive systems most effectively to motivate workers and thereby boost productivity (Hassink & Koning, 2009). In this thesis, the effect of an attendance bonus on workplace absenteeism will be examined.

Of course, all organizations have some absenteeism. In 2016, the absence rate in the Netherlands was 3,8% (CBS, 2017). These rates considerably differ between sectors. For this study, data of an inter-municipal social service in the Netherlands is used. According to Walstra (2016), absenteeism is a real problem in this sector. He reports that on average the absence-rate of employees at a municipality is 1,4% higher than the average absence rate in the Netherlands. In a report of the foundation

Arbeidsmarkt en Opleidingsfonds (2015) the four most important reasons for the high

absenteeism are the high workload and stress. In addition to that, workers are absent due to physical ailments and personal circumstances. On top of that, they indicate that executives do not pay enough attention to their absence.

For this thesis, the effectiveness of an attendance bonus incentive scheme of an inter-municipal social service in the Netherlands is analyzed. The workers at this organization are responsible for unemployment benefit, social support, debt service, re-integration and integration of people at these particular municipalities. The employees at this organization receive a bonus if they are not absent for a certain period of time. The aim of this bonus was to increase work-floor attendance. The results show that the introduction of the bonus not significantly reduced the absence rate. In addition to that, calculations show that there is a sorting effect, which is mainly coming from the fact that employees, who quitted before the bonus, have a relatively high absence rate compared to employees that have worked in both regimes and compared to the new hires. Furthermore, the bonus does not significantly influence the absence frequency. The negative effect was not robust for including both worker-fixed effects and control variables. On the average duration of absence, the bonus had no significant effect at all.

In the next section, an overview of the existing literature on absenteeism is presented. Section 3 provides the hypotheses that will be tested. After that, a description of the data and methodology is given. Section 5 presents the results. Finally, in section 6 a conclusion and a discussion of this research are given.

2. Literature review

In the first subsection of this section, demographic determinants that have been found to have an important link with absenteeism are discussed. A couple of these factors are included as control variables to obtain a more precise measure of the effect of the bonus. After that, the effect of employment protection is discussed. This factor is also found to be relevant and should be included in the analysis. In the third section, the difference in the degree of employee protection across countries is discussed. Some countries do protect employees more than others and this could influence the absenteeism of employees across the world. If significant differences are found, the results of this analysis are not necessarily generalizable for all countries. Finally, different kinds of incentive programs are discussed to get an idea about the effect of the bonus on the behavior of employees.

2.1 Demographic determinants of absenteeism

At first, the different personal characteristics, which were found to have an impact on employee absenteeism in previous research, will be discussed (Winkler, 1980).

2.1.1 Age and tenure

Several studies have looked at how age and tenure affect absenteeism. Gellatly (1995) examines whether absenteeism was influenced by age and organizational tenure. He surveyed 166 workers employed in a hospital in Canada. The employees had to indicate for how long they were employed and were asked for their age. Participants also had to identify themselves so that their responses could be matched to the absenteeism database of the hospital. Gellatly concluded that age as well as tenure was negatively related to the absenteeism of the employees. According to the researcher, this was due to the fact that older employees had higher absence norms and fewer incidents of absence than younger employees had. In addition to that, they are more loyal to the organization.

Other researchers found that age is positively related to absenteeism. Hackett (1990) made a distinction between avoidable and unavoidable absences. Het tested three hypotheses. (i) Age and tenure are inversely related to avoidable absences, particularly for males. (ii) Age and tenure are positively related to unavoidable absences, particularly for males and (iii) age is more directly related to both absence types than tenure. Hackett showed that age, but not tenure, was negatively related to avoidable absences. The effect was indeed particularly strong for men. For the unavoidable absences, no significant relationship was found between age or tenure.

Later, Ng and Feldman (2008) studied the relationship between age and performance. For their study, they used different measures of job performance, including absenteeism. Absenteeism was divided into three different categories. The first one is general absenteeism, which does not make a distinction between sickness and absenteeism because of random reasons. In addition to this, these two kinds of absenteeism were separated. Sickness absenteeism means involuntary absenteeism. Non-sickness-related absences are non-medically absences or withdrawal behavior. The researchers conclude that age had a small, but significant, positive relationship with sickness absences. For non-sickness-related absenteeism, no significant relationship was found.

2.1.2 Gender

Previous studies have shown that there exists a male-female illness gap (Ichino and Moretti, 2009). The most commonly given reason for this is that women stay at home more often because they take care of the children. This is also what Scott and McClellan (1990) found. The researchers examined if there was a significant difference between the absenteeism of women and men. For their study, they surveyed employees on junior and senior high schools in the United States. The demographics of the participants were matched to the absenteeism database of the schools. The researchers used data on absence frequency as well as absence duration. The results showed that absenteeism of women was not significantly higher than that of men. However, female employees take significantly more days off than men do. In addition to that, women were more likely to stay at home when children are sick. Furthermore, women often worked fewer hours than men did and were more satisfied with their work.

Within the literature, also other explanations for the illness gap are found. Herrmann and Rockoff (2012) used data of teachers on a public school in New York. The researchers collected data on absence timing, absence duration and reasons for absenteeism. The results indicated that younger female teachers are more often absent relative to older female teachers and male employees. In addition to that, they found that absenteeism of the teachers negatively influenced the performance of their students. A possible explanation for this result could be menstrual problems (Herrmann & Rockoff, 2013). The researchers conclude that female employees with menstrual problems are significantly more likely to be absent than women without these problems. According to the researchers, menstrual problems could declare up to 52% of the male-female illness gap.

Similar results were found by Ichino and Moretti (2009), who examine the relationship between gender and absenteeism at an Italian bank. Their results show that illness-related absenteeism is higher for women than for men. The probability of absence for women, relative to men, increases 28 days after their previous absence. This effect was not significant for employees older than 45. The researchers conclude that this result can explain a significant fraction of the male-female illness gap.

Research carried out by Hedges (1973) examines the influence of various factors that could influence unscheduled absenteeism. Unscheduled absenteeism

means absenteeism that results from illness or injury or related causes and arbitrary personal reasons (funeral leave, family responsibilities, attending to personal business,   and   didn’t   feel   like   working).   For   his   research,   Hedges used the Current Population Survey of the United States. The results show that not only gender, but also other factors must be taken into account by explaining the gender illness gap. Women are more likely to be new hires, and more likely to be employed in lower skilled and lower paid jobs. According to Hedges, these two factors are associated with relatively high absence rates. The difference in absence rates between women and men becomes smaller when comparisons are made within a particular occupational group.

2.1.3 Education

Koopmanschap et al. (1995) study the relationship between education and absenteeism. For their study, a sample of 25,000 Dutch workers was used. The researchers divided the labor market into five education levels. They used data on absence rate, absence frequency and length of absence. Their results show that absenteeism is strongly related to education. The absence rates of the lowest educated group are four times as high as the absence rates of the group with the highest education level. In addition to that, the results show that wage is positively related to both age and education level. According to the researchers, a possible explanation for these results is that education leads to good work habits, pleasant working conditions and good health (Grossman, 1975).

A couple of years later, Granlund (2009) studied the same relationship. For this study, a dataset of 23 countries was used. He made a difference between three education levels namely: elementary school, high school and university. The results showed that university graduates had the lowest absence rates. This result suggests that education is negatively related to absenteeism. However, the researchers stress that the results should be interpret with care, since employees with higher education were exposed to fewer health risks at work and had better health.

The above literature suggests that age, tenure, gender and education are relevant determinants of absenteeism. According to these papers, age is positively related to absenteeism. Tenure seems to be negatively related to absenteeism, just like education. However, the conclusions about the effect of education are solely based on

existing survey data. Therefore, caution is needed to draw causal claims. It is also suggested that women are absent more often than men.

2.2 Employment protection

Ichino and Riphahn (2004) test if there is a relationship between employment protection and absenteeism. For their study, they used data from Germany and Italy. In German, employees are generally more protected in the public sector relative to employees in the private sector. Therefore, the researchers compared absenteeism between these two sectors. The results indicate that the absence rate of the public sector is 4.1% higher than that of the private sector. In Italy, the degree of employment protection varies with the size of firms. It is relatively cheap to fire workers in firms that count less than 16 employees. These costs cannot be higher than six months of the  employee’s wage. The researchers compared absenteeism in firms of different sizes. The results show that the absence rate was significantly higher in the larger firms.

One year later, Ichino and Riphahn (2005) test the effect of probation. At the Italian bank that was studied, employees had a probation period of one year. This means that the employees were not protected against firing the first 12 moths of tenure. The researchers used weekly observations of 858 participants. They compared absenteeism of each employee in their probation period with his or her absenteeism after this period. Results show that there was a jump in absenteeism after the 12th month. The number of absences increases significantly, which suggests that employment protection causes an increase in absenteeism. The effect was particular significant for men.

Thalmaier (2001) found similar results. She used an empirical approach to test if absenteeism changes after an   employee’s probation period. For her analysis she used data of the GSOEP. This is a representative sample of native and an oversample of foreign residents in Germany. Both full-time and part-time employees were considered, but not the self-employed. Absenteeism is measured by indicating if an employee missed at least one day of work in a specified year. Employees in the sample had a probation period of six months. The results showed that there was a jump in the absence rate after the probation period. Thalmaier found that there was a much steeper increase in absenteeism between tenure months five/six and seven/eight

than between any of the other periods. She concluded that there is indeed an increase in absence rate after the end of the probation period.

Also Pfeifer (2009) proves that probation has an effect on absenteeism. For his study, he used the absence rates of employees of a German company. The absenteeism of employees was used as the dependent variable. This binary variable could take one if the worker was absent and zero otherwise. The variable of interest is probation, in this case the first three months of tenure. This variable could take one if the employee is in his or her probation period and zero otherwise. For his analysis, Pfeifer made use of the fixed effects model. The results showed that workers in their probation period are on average more than 50% less likely to be absent and have on average more than 60% fewer absent working days during their probation compared to the period of nine months afterwards.

The results of Ichino and Riphahn (2004 & 2005), Pfeifer (2009) and Thalemaier (2001) do not have exogenous variation and therefore do not allow for causal claims. Nonetheless, all papers present the same conclusion: when an employee is not in his or her probation period, the worker’s   absenteeism will increase. Therefore, it is plausible that probation decreases the absence rate.

2.3 Cross-country differences in absenteeism

Research showed that absenteeism was positively associated with the replacement rate, defined as the level of compensation if not going to work (Buzzard & Shaw, 1952). Therefore, it could be that differences across countries in sickness payment will lead to differences in absenteeism between workers only because they are living in different countries (Frick & Malo, 2008).

Frick and Malo (2008) focus on the effect of employment protection and sickness benefits on absenteeism. The researches used survey data of workers employed in different countries in Europe. They include employees as well as self-employed people. Danish and Finnish workers had on average the most days of absence whereas Greek and Spanish workers had the highest attendance at work. Although there are differences between countries, the results suggest that employment protection does not influence the days of absence. However, the marginal effect of sickness payment increases the number of days absent by two. Nevertheless, the researchers indicate that the impact of employment protection and sickness benefits is

smaller than that of some individual characteristics like family status, working hours a week, kind of contract and income.

Similar research was done by Lusinyan and Bonato (2007). They discuss the differences in absenteeism across European countries. In addition to the generosity of sickness benefits and employment protection they also discuss the effect of publicly and privately financed insurance schemes on absenteeism. The results show that the degree of employment protection is positively related to sickness absenteeism. In addition to that, a positive relationship between sickness benefits and absenteeism was found. Furthermore, absenteeism and the generosity of the sickness insurance scheme are also positively related. Lastly, the researchers provide suggestive evidence that absence rates increase when employers have to pay a relatively small part of the sickness insurance costs.

Prins and de Graaf (1986) focused on absenteeism in Belgian, German and Dutch organizations. They analyze the effect of job security regulations and sickness benefit schemes. The researchers found significant differences in regulations between countries. Belgium shows the most varied control procedures and the lowest sickness benefits. In addition to that, it is very hard to be qualified for invalidity benefits in this country. They also emphasize that in the Netherlands, qualifying for these benefits is relatively easy. The results show that Belgian employees had on average the fewest days of sickness absence a year, namely 20.3. German employees had on average 28.5 days and Dutch employees had the most with an average of 39.1 days. The researchers conclude that the job security regulations and sickness benefits strongly affect sickness absence.

Out of these papers, it can be concluded that employment protection, sickness benefits and insurance schemes affect absenteeism. When the financial conditions for the sick or unemployed are well organized, the absence rate is relatively high.

2.4 Instruments to reduce absenteeism

2.4.1 Negative incentive programs

Dalton and Mesch (1991) study the effect of an absence policy including a two-day waiting period and a 90-day accumulation. The incentive scheme is designed as follows. Employees, who are absent for one or two days, are not paid during this period. An employee earns 18 days of sick leave in a year. However, to exercise the

benefits of these days, employees need to be sick. If employees do not use their sick leave days, they can save them up to 90 days of sick leave. When an employee has saved 90 days of sick leave, the worker will be paid from the first day of absence. The data that were used are derived from communication workers of a public utility company in the United States. These workers were paid on an hourly-wage basis. The variable of interest is whether an employee had accumulated 90-days sick leave or not. The researchers used three different categories of absenteeism: total absence, absence due to illness and avoidable absence. The results show that the incentive program was strongly associated with the number of avoidable absences. Therefore, they conclude that this absence policy is a strong predictor of employee absenteeism.

Barmby, Orme and Treble (1995) discuss the effect of another incentive system that works with sick pay points. This policy is designed in the following way. Absence, which is unacceptable to the firm, will give employees sick pay points. Workers are graded A, B, or C according to their average yearly points over the past two years. An employee is graded A if he or she has less than 10.5 points, B if he or she has 10.5 to less than 20.5 points or C, if more than 20.5 points. For one day of unacceptable absence, the employee receives one point. If a worker is graded A, his or her sick pay will be set at normal pay plus a bonus. If a worker is graded B he or she receives a basic pay and if C, the worker will receive no sick pay at all. The researchers used data on absences of two manufacturing plants in the United Kingdom. They control for personal characteristics that could also influence absenteeism. Absenteeism itself is indicated with a dummy variable that could take one if a worker was absent or zero otherwise. The results suggest that this absence policy effectively controls the absenteeism of employees. Furthermore, they conclude that full-time workers have higher absence rates than part-time workers. In addition to that, they found that wage is negatively related to absenteeism. According to the researchers, this could be due to a positive selection process, so workers with good attendance have a higher probability to get higher paid jobs.

Winkler (1980) examines the effectiveness of four different absence policies for teachers. He tests (i) forgo salary, which means the teacher does not receive salary when he or she is absent. Secondly, (ii) income protection, by this the researcher means that the teacher is covered by an income protection plan. The most common income protection plan is insurance against the loss of salary if the length of illness exceeds the accumulated sick-leave days. A third policy is (iii) proof of illness, this

means the teacher has to prove to the administration he or she is sick. The last policy that is discussed is (iv) report to principal, which means the teacher has to report his or her absences to the principal of the school. The results show that the forgo salary policy significantly reduced the absence rate. In addition to that, requiring the teacher to demonstrate proof of illness also leads to lower absenteeism. Requiring the teacher to report every absence directly to the principal results in a large reduction of term absenteeism. However, the income protection plan leads to an increase of short-term absenteeism.

2.4.2 Positive incentive programs

Instead of punishing absenteeism, it is also possible to reward attendance. In many European countries, negative incentives for absenteeism are not easy to implement due to income and job protection. In the Netherlands, firms are obligated to pay 70% of gross earnings during sickness (de Jong & Lindeboom, 2003). Because it is often not possible to apply negative incentives, firms can try to influence the behavior of workers with positive incentives.

Scott, Markham and Robers (1985) discuss four ways how firms could reward good attendance. In order to study the effect of different financial incentive programs, research was done in a company that was willing to try a number of different attendance improvement programs over a period of one year. Although the workers were not paid when absent, the absence rate of the firm was around 6%. The researchers tested the following four programs: (i) the financial incentive program, (ii) the recognition program, (iii) the lottery program and (iv) the information feedback program. In the financial incentive program, workers receive a small amount of money when they are not absent for a year. The recognition program consists of the manager sending a signed card to praise the employee with the fact that he or she was not absent for a year. In the lottery program, employees were able to win a big prize. Employees who had a perfect attendance were entered twice at the list of potential winners. The last program the researchers tested was the information feedback program. In this program, the workers received feedback about their absences every month. However, no further attempt   was   made   to   change   the   worker’s absence behavior. The programs were all introduced in a different plant. Two other plants, where no program was introduced, were used as control group. The absence rate was calculated daily and computed by dividing the number of absences by the total

number of employees working that day. Participants also filled in a questionnaire to share their attitudes toward absenteeism and the different incentive programs. The results show that the recognition program decreased the absences the most. In this plant, absenteeism decreased by 36.9%. Also the questionnaire indicates that workers thought this program would work. In addition to that, the financial incentive program significantly reduced absenteeism, but less than the recognition program. The information feedback method caused a non-significant decrease in absence. Lastly, the lottery method resulted in a significant increase of absenteeism.

Hassink and Koning (2009) discuss the effectiveness of a lottery. In order to do this, the researchers used data of 481 workers of a Dutch manufacturing firm. Every month, the firm held a lottery. To participate in this lottery, employees should not have been absent for three months in a row and should not have won the lottery earlier. The researchers introduced four states to test the incentive effect of the lottery. In state 1, the worker may participate in the upcoming lottery in month m+1. In state 2 and state 3, the first possibilities to participate are in month m+2 and m+3, respectively. The researchers show that the lottery results in a sick leave of 1.6 percentage  point.  The  workers’  monthly  incidence  of  sickness  absence  decreased  by   4.3 percentage points in the first seven months and by 1 percentage point in the seven months after that. Therefore, the researchers conclude that this incentive program is highly beneficial. The benefits of employees not being sick exceed the costs of the lottery. However, when workers had won the lottery once, they continue their previous rate of absence.

2.4.3 The effect of an attendance-bonus on absenteeism

Herrmann et al. (1973) study the effectiveness of a bonus incentive system designed to increase the punctuality of six employees who were late for work often. The researchers used data of a manufacturing company in Mexico. Every day a worker was on time, he or she received two pesos. However, if the worker was late or absent for any reason, he or she lost the bonus for that day. The punctuality of the employees was compared to records before the bonus was introduced and to records of a control group. The researchers measured late coming by the number of workers that were late during two weeks divided by the total number of possibilities to be on time in these two weeks. The bonus increased the punctuality of the workers compared to before. In the control group, punctuality decreased. According to the researchers, a possible reason for the effectiveness of this incentive system could be that employees were rewarded for their performance by the end of each week. In this way, arriving on time was almost immediately rewarded. Before, the workers had to wait until the end of the year to receive their bonus.

As already discussed above, Scott, Markham and Robers (1985) discuss several ways of how firms could reward good attendance. The researchers used a firm with different plants, four of them implemented an incentive program and the other two served as a control group. One of the programs they discuss is the financial incentive program in which workers receive 50 dollars when they were not absent the whole year. When employees are absent only once, they still received 25 dollars. Employees were not reminded to the fact that they were still eligible for the bonus. The absence rate was calculated daily and computed by dividing the number of absentees by the total number of employees working that day. The results show that the bonus significantly decreased the absence rate from 6.35% to 6.04%. The survey showed that about 60% of the employees thought the program worked.

Robins & Lloyd (1984) study the effect of attendance bonus on absenteeism at a preschool. All full-time employees of the preschool participated in this study. Part-time workers were not covered by preschool sick leave policies and therefore were ignored. This resulted in a sample of 13 employees. An employee is considered absent if he or she was not at the preschool for four or more working hours of an eight-hour workday. To reduce absenteeism, several incentives were made available. Employees could join the program if they had accumulated at least five days in their sick leave accounts. The first option was (i) to receive a 20-dollar bonus. In this case,

participants who had no absences for a full calendar month received 20-dollar bonus instead of accumulating 1.25 sick leave days, which costs the school 21 dollars. A participating employee, who was absent during the month, did not receive 20 dollars. Instead his or her sick leave days subtracted by the number of days the employee was absent were added to his or her sick leave account. Secondly, employees could choose (ii) to take one extra day off. Furthermore, after three months without any absence, the workers could also choose (iii) to receive a 75-dollar bonus or (iv) three extra days off. The last option was (v) a small present like a t-shirt, lunch at a restaurant with the director or three free bottles of soda a day. The results show that since the program was made available, absenteeism for all participating employees decreased from 4% to 2.6%. Although the observation period was short, the results suggest that the bonus and choice programs were responsible for the reduction in the absence rate. Absenteeism decreased since the incentive program was available. In addition to that, employees who were in the incentive program showed lower absenteeism than others. The four employees that were in the bonus program already had a low absence rate of 1.5%. The absence rate of these workers dropped to 0%. After a while, their absence rate increased somewhat, to 2.2%. However, the bonus program reduced their absenteeism enough to substantially affect overall absenteeism rates at the preschool. Another finding of this research is that most employees preferred monetary incentives in the choice program. However, the use of monetary incentives did not result in higher costs for the school.

2.4.4. Crowding out effect of monetary incentives

The literature above indicates that the introduction of a bonus could help firms to reduce their absence rates. Other researchers suggest that monetary incentives not always improve performance. They claim that people could display exactly the opposite behavior (Gneezy & Rustichini, 2000).

Gneezy and Rustichini (2000) discuss the effect of monetary incentives on performance. For their research, they made a distinction between intrinsic and extrinsic motivation. By intrinsic motivation they mean that a person has a motivation of his or her own, independent of any reward. When they talk about extrinsic motivation, they mean that there is an external reward associated with the activity. The researchers did two experiments. The first one was called the IQ-experiment. For this experiment they used 160 students of the University of Haifa. They were asked to

answer 50 problem sets and were told this was a sort of IQ test. The researchers made four different treatments, according to the payment for correct answers. The payments in each treatment for each correct answer were (i) 0 NIS, (ii) 0.1 NIS, (iii) 1 NIS, and (iv) 3 NIS. In the other experiment, called the donation-experiment, they observe the behavior of 90 high-school students in Israel. For this experiment the researchers designed three treatments. In the first one, an experimenter told the students how important it was to collect money and that the results will be published. In the second treatment, the experimenter did the same but now 1% of the amount collected was promised to the students. In the third treatment, this was 10% of the amount. The results of the IQ-experiment show that in the IQ test the average number of correct answers decreased from 28.4 in the first treatment to 23.1 in the second treatment. This is a significant difference. The results of the donation experiment show the same pattern. When the compensation was zero, students collected an amount of 238.6 NIS. However, when the compensation was 1% the average collection was 153.7 NIS, which is again significantly lower. In the third treatment, the average collection is 219.3 NIS. The results show that when a small amount of money is offered, the performance is worse then when there was no compensation at all. This indicates that extrinsic motivation could crowd out intrinsic motivation.

Deci and Ryan (1985) also study the effect of monetary incentives on performance. The subjects were working on different tasks during three stages: (i) no reward, (ii) reward and then again (iii) no reward. Effort was measured during each stage. The results show that even in stage one the participants made effort, which indicates their intrinsic motivation. In stage two, when a monetary reward was promised, the performance of the subjects increased. However, in stage three the performance of the participants decreased below the performance level of stage one. The researchers conclude that extrinsic motivation crowds out intrinsic motivation of people and thus that there are hidden costs of rewards.

The literature above suggests that both negative and positive incentive programs could help to reduce absenteeism. In the Netherlands, positive incentive programs are easier to implement because of employee protection. Besides evidence out of surveys, Hassink & Koning (2009) empirically prove that a lottery reduces absenteeism. However, the effect of a bonus is still ambiguous. Hermann et al. (1973) suggest that a bonus could improve punctuality of workers, but this research is based on a sample

of only six workers and therefore has a very low explanatory power. This also applies to the experiment of Robins & Lloyd (1984) who based their conclusions on a sample of 13 employees. Scott, Markham and Robers (1985) prove that a bonus positively influences the attendance of workers. These researchers made use of a larger sample and a control group. Therefore, their results are more reliable. However, Gneezy and Rustichini (2000) and Deci and Ryan (1985) show that it is also possible that the extrinsic motivation of a bonus crowds out intrinsic motivation of employees to reduce their absenteeism.

Up to now, it is assumed that a reduction in absenteeism is desired. However, sickness presenteeism, which means going to work while sick, could also have negative economic effects because it can lower the productivity of employees (Pauly et al., 2002; Pauly et al., 2008). This has to be taken into account when interpreting the results.

Many researchers have focused on the effect of individual factors that can influence absenteeism. However, organizational factors seem to have greater impact on absence rates. To my knowledge, there is not much literature yet on the incentive effect of a bonus to reduce absenteeism. This paper will examine the effect of the bonus by using quasi-experimental data. Most papers that explicitly research the effect of the bonus often base their results on a small sample. Due to accurate maintenance of absenteeism in this organization, data on absences over 14 years could be used. This resulted in a sample of 177 employees, which will give the results more explanatory power relative to former research. In addition to that, existing literature discusses if the bonus reduced absenteeism in general but says nothing about the tenure or frequency of absence. Because this thesis makes use of a very detailed dataset (not based on self reports), the effect of the bonus on absence tenure and frequency could also be tested. Furthermore, this thesis has data on several background characteristics of the employees. This makes it possible to include and control for different determinants that were found to have an impact on absenteeism and obtain a more precise estimate of the effect of the bonus. On top of that, the workers are surveyed so that their attitudes towards the bonus could also be discussed.

3. Hypotheses

This thesis offers the following main hypothesis:

Hypothesis 1: The introduction of the bonus reduces the absence rate of the firm If the hypothesis is confirmed, this is evidence that the introduction of the bonus reduced the rate of absence. The absence rate is defined as the calendar days of absence divided by the total calendar days employed in the corresponding period. Former literature says that non-monetary, but also monetary incentives, will decrease the absences of employees on the work-floor (Hassink and Koning, 2009; Robins & Lloyd, 1984; Scott, Markham & Robers, 1985). The effect of the introduction of the bonus can be divided into an incentive effect and a sorting effect (Lazear, 2000). The incentive effect means that the bonus will encourage the workers to call in sick less often. It could also be the case that the organization attracts other kind of workers than before or that a particular kind of people left after introducing the bonus. This is called the sorting effect. However, the occurrence of the sorting effect is not very plausible. Because the bonus is relatively small, it is not very likely to influence peoples’   job   application   or   choice   to   quit   or   stay.   In   addition   to   that,   when   the   organization tries to attract new workers, they never mention the bonus in ads. The bonus is only mentioned in the job interview of the potential employee.

This thesis also examines a couple of sub-hypotheses.

Hypothesis 2: Conditional on being absent at least once, the bonus increases the average length of absence

The incentive program of the organization is designed in a way that when employees are absent once, they immediately lose their bonus for that particular period. This means once the employee is absent, he or she is not incentivized anymore. The loss of incentives could lead to less motivation to come to work again. The lack of incentives therefore could lead to employees extending their absenteeism by some days because they already lost their bonus anyway. In addition to that, the workers first were incentivized and all of a sudden not anymore. This could decrease their initial

intrinsic motivation to decrease their absenteeism (Deci and Ryan, 1985). There could also be a sunk-cost effect. This is the tendency of employees to continue their activity because of the money they have already spent on it. In this case, employees continue being sick because they already spent the amount of the bonus on it. Therefore, when an employee is ill, the length of absence could be longer than before the introduction of the bonus because before, the employees did not felt like they lost something.

Hypothesis 3: Conditional on being absent at least once, the bonus increases the frequency of absence

As said before, due to the design of the incentive program, employees lose their incentive when they are absent once. Therefore, it does not matter anymore how often they are absent in that particular period. The employee lost his or her bonus anyway. The loss of incentives could decrease the motivation of the employees and could cause employees to increase their frequency of absence during a particular period (Deci and Ryan, 1985). Here, the sunk-cost effect could also play a role. The employees lost the bonus while they were absent and therefore it may be that they continue being absent for multiple times in the same period.

Hypothesis 4: Workers that have a temporary contract have a lower absence rate than workers with an indefinite contract

Literature suggests that when employees are less protected, they have a lower absence rate (Ichino and Riphahn, 2004; Ichino and Riphahn, 2005; Pfeifer, 2009; Thalmaier, 2001). Employees with a temporary contract could be more easily fired than employees with an indefinite contract. In addition to that, an employee has to make a good impression to the firm in this period to obtain an indefinite contract. Therefore, it is expected that employees with a temporary contract have a lower absence rate than employees who have an indefinite contract.

The measures of absence have a relationship with each other that is given in the formula below. When two of the three measures are known, the other one could be easily calculated.

Absence  rate =          ∗        

According to the first hypothesis, it is expected that the introduction of the bonus will reduce the absence rate of the organization. Because the three measures have a relationship with each other and the total calendar days should approximately remain the same for each period, the average absence length or the absence frequency should also decrease to confirm hypothesis 1. Therefore, the expected increase in average absence length and absence frequency is conditional on an employee being absent at least once. It is expected that the average absence length and absence frequency will increase, only when an employee is absent at least once. This is because only then their extrinsic motivation is lost. In this way, the overall average absence length and absence frequency could still decrease due to the introduction of the bonus because employees are absent less often. This means that hypothesis 1 could still be confirmed without contradicting hypothesis 2 and 31.

4. Data & methodology 4.1 Organization and context

The analysis in this thesis is based on data of an inter-municipal social service in the Netherlands called organization X. The employees are responsible for unemployment benefit, social support, debt service, re-integration and integration of people at municipalities. At this moment, organization X takes care over seven municipalities. In 2008, the organization implemented an incentive scheme to reduce absenteeism. Before, there was no incentive to reduce the absence rate of the organization. Since 2008, a bonus is given to employees that are not absent for a certain period of time. To examine the effect of the bonus incentive scheme, a panel of 177 employees in a timeframe of 2003-2016 is used. This is an unbalanced panel; not all employees have worked for the organization over the whole period of time.

The functions of the employees in question are very different. Table 1 gives an overview of the different jobs an employee of this organization could have. Because there are different functions, employees have different salaries. The organization works with so called ‘salary  scales’.  Table  2  gives  an  overview  of  the  different  salary  

1 _{The hypotheses above are formulated one-sided because theory predicts this. However, the tests that are done are two-sided in}

scales in this organization and the corresponding functions. The salaries vary between €1616, - and €6413, -. This means that some absences are more costly than others.

TABLE 1

Distribution of employees over functions

Function N Share

Director 2 1,13%

Controller 1 0,56%

Department head performance 1 0,56%

Policy officer 4 2,26%

Coordinator 5 2,82%

Legal employee objection and appeal 4 2,26% Employee general services 1 0,56%

Quality employee 6 3,39%

Application manager 3 1,69%

Consultants 101 57,06%

Purchase and contract management 1 0,56%

Social detective 2 1,13% Management assistant 1 0,56% Content manager 1 0,56% Regular Employees 34 19,21% Archive employee 6 3,39% Secretary 1 0,56% Concierge 2 1,13% Receptionist 1 0,56% Total 177 100,00%

In the period that is considered, the organization greatly expanded. Figure 1 shows that the number of employees almost tripled over a period of 14 years. One of the reasons for the expansion is that, over the years, other municipalities decided to join the collaboration. In June 2002 the social service was founded with already five municipalities. On January 1st 2006 and January 1st 2007 two other municipalities were added to the collaboration.

TABLE 2 Overview salary scales Salary scale Salary Functions

4 €1616 - €2409 Concierge 5 €1663  - €2535 Archive workers 6 €1773  - €2662 Regular employees

8 €2277  - €3293 Management assistant, Content manager, consultant 9 €2527  - €3712 Consultants, detectives, Application manager,

Purchase and contract management

10 €2726  - €4121 P&O consultant, Legal employee objection and appeal, coordinator

11 €3266  - €4740 Policy worker

12 €3923  - €5384 Department head performance, controller 14 €4653  - €6413 Director

Note: The salaries of the receptionist and the secretary are not determined within this system. It is assumed that these salaries are closest to that of the concierge. Therefore scale 4 is used for calculations.

The reason the organization introduced an incentive scheme for absenteeism is partly due to the mergers. Because the different municipalities decided to work together, some problems arose. The organization had to reorganize all the different information systems to one working system. This was a difficult task and resulted in dissatisfied customers and many angry phone calls. In addition to that, due to reorganization of departments there were lots of social conflicts between employees. These factors combined resulted in a disproportional high absence rate. Figure 2 shows the development of the yearly absence rate of the organization. Two kinds of absence rates are calculated: one with pregnancy leave and one without these absences. To compare the absence rate of the organization to others, the absence rate of similar organizations is also shown in the figure. This is the absence rate of other employees working for municipalities with a comparable number of inhabitants. Note that these employees are not necessarily employees of a social service. Organization X takes care of multiple municipalities and therefore takes care of more than 100.000 inhabitants. The average absence rate of all employees in the Netherlands is also

0 20 40 60 80 100 120 140 2003 – I 2003 – II 2004 – I 2004 – II 2005 – I 2005 – II 2006 – I 2006 – II 2007 – I 2007 – II 2008 – I 2008 – II 2009 – I 2009 – II 2010 – I 2010 – II 2011 – I 2011 – II 2012 – I 2012 – II 2013 – I 2013 – II 2014 – I 2014 – II 2015 – I 2015 – II 2016 – I 2016 – II Num ber of em plo yee s

Year and period FIGURE 1

Distribution of the number of employees over the years

shown. Until 2012, the absence rate of the treatment organization was substantially higher than that of similar organizations and that of the Netherlands. It also indicates that, since 2009, which is one year after the introduction of the bonus, there is a downward trend in the absence rate.

The organization took several actions because they felt like their absence rate was way to high. In order to prevent the high absence rates, the organization switched to a different ARBO-service in 2008 and held a formation research in 2009 and 2010. In addition  to  that,  they  introduced  the  ‘interne  flexibele  schil’. This means they educate one consultant and one regular employee how to perform in every different department. As a result, external hiring was not needed anymore. These employees were not only cheaper, but it also resulted in a better quality of work. On top of that, these people were more familiar with the organization. Another action the organization took, the one that is most important for this thesis, is the introduction of the bonus for employees who were not absent for a certain period of time.

Note: Absence rates of similar organizations were  obtained  from  the  website  of  the  ‘stichting Arbeidsmarkt  en  Opleidingsfonds  Gemeenten’.  Absence  rates  of  the  Netherlands   were  obtained  from  the  website  of  the  ‘Centraal  Bureau  voor  de  Statistiek’.  For  the  absence  rates  of  the  organization  there  is not accounted for partial recovery. Absence rates of the organization were calculated in the following way:

Absence  rate =                        x  100%. 0 2 4 6 8 10 12 14 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 FIGURE 2

Yearly absence rates in the organization, similar organizations and the Netherlands

Absence rate organization with pregnency

Absence rate organization without pregnancy

Absence rate similar organizations

Absence rate Netherlands Absence rate organization with pregnancy leave

Absence rate organization without pregnancy leave Absence rate similar organizations

Since 2008, employees receive a bonus if they are not absent. In 2008 and 2009, the organization held a trial. Employees were promised an amount of €100, - if they were not absent for a whole calendar year. Since 2010, they introduced a more detailed incentive plan. The bonus is distributed each year on the 1st of January and the 1st of July. The first time a worker manages it to be not absent for half a year he or she receives €100,  - at the distribution moment. When the worker still has no absence report at   the   next   distribution   moment,   he   or   she   receives   €150,   -. The worker will receive   €150,   - each upcoming distribution moment when he or she is not absent. When  the  employee  breaks  this  chain,  he  or  she  will  again  receive  €100,   - when not absent for another half year. The reason the employees are absent is not important. Only pregnancy leave does not count as an absence. Absence as a result of pregnancy does. Table 3 shows how many employees actual earned (2008 and beyond) or would have earned (before 2008) the bonus if it were in place in the corresponding period. According to the table, the percentage of people that is entitled to the bonus seems to increase after the actual introduction of the bonus in 2008.

To obtain this dataset, first the approval of the management team (MT) was needed. When the request was accepted, it was possible to start collecting data. The staff and organization department of organization X collected individual personal records   of   all   workers   regarding   the   workers’   first   name,   surname, date of birth, gender, function, full-time equivalent, date of commencement of employment and date of quitting. Furthermore, HR-reports of the database, in which the organization maintains the absences, were obtained with all information about the absences of the employees. In this reports, the name of the employee was indicated, at which date the employee was absent, until when the employee was absent and in how many days of absence that resulted. To make sure all absences were in these reports, the reported absences were compared  to  absences  indicated  by  the  organization’s  check-in-system. All information was collected at the office of the organization for privacy reasons. Before the data could be used for research, the names of the workers were replaced by random numbers, so the subjects in this study are as anonymous as possible.

Collecting data in this way ensures a very accurate dataset. By using data of the information system of the organization instead of survey data there is no problem with low response rates or missing variables. Besides this, the bonus is real and not hypothetical, which makes the results more reliable.

TABLE 3

Number of participants and employees that would have (<2008) or actually (≥2008) received the bonus A. Before the introduction of the bonus

Year and period Participants €100 €150 % of employees that

would have received the bonus 2003 – I 34 12 - 35,29% 2003 – II 38 4 9 34,21% 2004 – I 40 8 10 45,00% 2004 – II 40 4 11 37,50% 2005 – I 38 8 8 42,11% 2005 – II 37 6 8 37,84% 2006 – I 45 6 11 37,78% 2006 – II 43 11 12 53,49% 2007 – I 48 7 12 39,58% 2007 – II 57 14 15 50,88%

B. After the introduction of the bonus

Year and period Participants €100 €150 % of employees that

actually received the bonus 2008 – I 67 13 16 43,28% 2008 – II 71 17 19 50,70% 2009 – I 73 13 24 50,68% 2009 – II 73 15 29 60,27% 2010 – I 79 13 30 54,43% 2010 – II 87 12 37 56,32% 2011 – I 85 17 39 65,88% 2011 – II 85 15 45 70,59% 2012 – I 85 13 56 81,18% 2012 – II 87 9 63 82,76% 2013 – I 87 6 50 64,37% 2013 – II 87 18 48 75,86% 2014 – I 87 11 51 71,26% 2014 – II 90 12 57 76,67% 2015 – I 89 9 53 69,66% 2015 – II 98 26 55 82,65% 2016 – I 106 15 57 67,92% 2016 – II 103 12 58 67,96%

Note: The amounts before 2008 are calculated on basis of the data, the bonus was not really distributed at that time. For 2008 and 2009, the eligible participants could only earn €100 if they were not absent for a whole year. However, for consistency the bonus is calculated over a half-year basis. In 2010, the bonus was first distributed according to the current incentive system.

4.2 Descriptive statistics

Some basic characteristics of the sample are reported in table 4. The sample counts 177 different individuals that were employed between 2003 and 2016. The data thus includes within-employee and between-employee observations, i.e. a panel structure. The data are organized as follows. Each half-year indicates one period. There are 2123 employee-period observations over 14 years.

Three measures of absenteeism are used. The first one is (i) absence rate. This measure indicates the number of calendar days an individual employee was absent divided by the number of calendar days employed in the corresponding period. The second measure is (ii) average length of absence. This measure indicates the average number of days an employee is absent when he or she is ill. When an employee is absent for longer than one period, this still counts as one sick leave. Therefore, the total number of days absent of that particular sick leave is assigned to the period when the employee called in sick. This means that it is possible that there are more days of absence than actual calendar days in a particular period. The last measure of absence is (iii) frequency of absence, this measure indicates the number of times an employee was reported ill in a specific period of time. In the absence measures, partial recovery is not taken into account. The HR-report indicated if the worker was absent because of pregnancy leave. If this is the case, the absence is left out because when an employee is absent due to pregnancy leave, she can still earn the bonus.

The average absence rate in a period is 5.63% with a frequency of 0.49 and an average duration of 20.18 days per period. According to the formula represented in section 3, the absence rate should be equal to the average calendar days of absence, times the frequency of absence, divided by the average days a worker was employed in the corresponding period. Indeed, 0.49 times 20.18 divided by 175.19 is 5,64% (there is a small rounding difference). This shows that, if two of the three variables are known, the other one could be calculated. The average absence length could be calculated by multiplying the absence rate by the total number of calendar days employed, divided by the absence frequency. In addition to that, the absence frequency could be calculated by multiplying the absence rate by the total number of calendar days employed, divided by the average absence length. The reason that the average calendar days employed in each period is less than a half-year is that not all employees are employed for a whole period. Sometimes an employee is hired or quitting in the middle of a period.

The introduction of the bonus is indicated by a dummy variable. The bonus was introduced in 2008. Therefore, the dummy indicates zero before 2008 and one after. Bonus is the variable of interest. Table 4 shows that the average of the variable bonus is 0.78. This means that there are more observations in the dataset were the bonus-system was used than observations where this system was not yet introduced.

For this analysis, also influential background characteristics of the participants were collected. Female is a dummy variable and indicates one if the employee is a woman and zero otherwise. As shown in table 4, the share of female in this organization is very large and around 78%. Also information on the age of the employees was collected. The age that is included is the age of the employees at the beginning of the corresponding period. The average age is around 43 years old. The variable tenure shows the tenure at the current organization. Some employees came from different municipalities. But, it was not possible to obtain information on their tenure there. Therefore, the tenure at the current organization is included. The variable tenure that is included is the number of years an employee is employed at the organization at the beginning of the period. The average tenure is 3.84 years. This seems short but the organization was only founded in 2002. In addition to that, it has expanded significantly. This means a lot of workers are hired not that long ago. This also indicates that there is only a limited number of workers (n=62) employed both before and after the introduction of the bonus. The full-time equivalent is an indication of the number of hours an employee works in a week. A full-time employee at this organization works 36 hours a week and has a full-time equivalent of one. These hours are flexible to a certain extent. The most recent calculation of the full-time equivalent of a worker is used in every period. The average full-time equivalent of an employee is 0.83, which can be recalculated to an average of 30 hours a week. For salary scale, also the most recent scale of the employee is used. It could be that employees have switched from function over the years. In that case, the salary scale of the last function is included in every period. The actual salary could differ between workers within the same scale. This is because there is a different salary for part-time and full-time workers. The salary scale could be a very rough estimate for education because often employees in a higher salary scale have a higher level of education. The variable education itself is an important determinant of absenteeism, as shown in section 2, but unfortunately not known. The mean salary scale is around 9. This is not surprising because the majority of the employees is

consultant and is paid according to scale 9. The last variable that is included is temporary contract. This control variable is a dummy and indicates one if the worker has a temporary contract that period and zero if not. All workers get a temporary contract the first year they are employed. When everything goes well in the first year, the worker will be offered an indefinite contract. When a worker is employed at the 1st of January 2006 or the 1st of January 2007, it is assumed that he or she worked at the social service of another municipality before. These employees do not have a temporary contract because they already had an indefinite contract before the merger. The overall means do not reveal anything about the potential effect of the bonus. Therefore, in table 5 the summary statistics were separated into the characteristics of employees before and after the introduction of the bonus. Before 2008, there were somewhat less women employed than after 2008. Still, female employees are in the majority in both cases. Age and salary scale are almost equal but tenure differs. This is because the organization was founded in 2002. The lower absence rate is also quite significant. Even so the decline in frequency. The average length of absence is almost the same.

TABLE 4

Summary statistics of employees

Variable Observations Mean Standard

deviation Min Max A. Background characteristics (N=177) Bonus 2123 0.78 0.41 0 1 Female 2123 0.78 0.41 0 1 Age 2123 42.88 9.67 22 65 Tenure 2123 3.84 3.44 0 14 Full-time equivalent 2123 0.83 0.17 0.12 1 Salary scale 2123 8.92 1.47 4 14 Temporary contract 2123 0.20 0.40 0 1 B. Absenteeism (N=177) Absence rate 2123 5.63 17.42 0 100 Average length of absence 710 22.18 64.99 1 733 Frequency of absence 2123 0.49 0.83 0 7 Average number of days

employed in period T

2123 175.20 28.82 6 184

Note: ‘N’   denotes   number   of   employees.   Period: January 2003 - January 2017. Absence rates of employees are calculated without pregnancy leave. For the average length of absence, 1414 observations were dropped because employees were not absent in these periods.

TABLE 5

Summary statistics of employees employed before (<2008) and after (≥2008) the introduction of the bonus

Variable Observations Mean Standard deviation Min Max

I. Before introduction of the bonus (N=74) A. Background characteristics Bonus 462 0.00 0.00 0 0 Female 462 0.69 0.46 0 1 Age 462 41.59 9.35 22 61 Tenure 462 1.75 1.57 0 5 Full-time equivalent 462 0.83 0.19 0.12 1 Salary scale 462 9.08 1.71 4 14 Temporary contract 462 0.27 0.44 0 1 B. Absenteeism Absence rate 462 7.96 19.94 0 100 Average length of absence 241 22.39 81.48 1 733 Frequency of absence 462 0.90 1.15 0 7

II. After introduction of the bonus (N=165) A. Background characteristics Bonus 1661 1.00 1.00 1 1 Female 1661 0.80 0.40 0 1 Age 1661 43.24 9.73 22 65 Tenure 1661 4.42 3.59 0 14 Full-time equivalent 1661 0.84 0.16 0.12 1 Salary scale 1661 8.88 1.40 4 14 Temporary contract 1661 0.18 0.38 0 1 B. Absenteeism Absence rate 1661 4.99 16.60 0 100 Average length of absence 469 19.05 54.69 1 557 Frequency of absence 1661 0.38 0.68 0 4

Note: ‘N’   denotes   number   of   employees.   Period 1: January 2003 - January 2008. Period 2: January 2008 – January 2017. Absence rates of employees are calculated without pregnancy leave. For the average length of absence, 222 and 1192 observations were dropped respectively because employees were not absent in these periods.

Besides summary statistics, correlations between all variables were calculated. These are shown in table 6. The bonus is correlated to all variables despite the full-time equivalent and the average length of absence. This suggests that the introduction of the bonus will not have a significant effect on the average absence length. The bonus has a negative correlation with the absence rate and absence frequency, which suggest that the introduction of the bonus decreases both. Female is negatively correlated to age, full-time equivalent and salary, which means that female employees are younger, work less and earn less than male employees do. Age is positively correlated to tenure, which means that older employees worked at the organization for a longer period of time. It is also less likely that an older employee has a temporary contract. Age is negatively correlated to the frequency of absence. However, it is positively correlated to absence rate and the average length of absence. This means older employees are less absent than younger employees, but when they are absent, they are longer absent than younger workers. Tenure is negatively correlated to full-time equivalent and frequency of absence. So, employees who work for a longer full-time at the organization are the ones that work less often and are less absent. Salary is negatively correlated to the absence rate and the frequency of absence. This means employees who earn more seem to be less absent. Lastly, the absence rate is positively correlated to absence frequency and average length of absence.

These simple statistics do not take other factors into account. However, period effects can matter. The financial crisis for example could affect the absence rate of the employees. The other actions that the organization took over the years to reduce their absenteeism could also have influenced absenteeism. To deal with these factors, a time trend variable is included in the model. The model will be explained in the next section.