9
The Journal of Socio-Economics 39 (2010) 420–428 Contents lists available at ScienceDirect The Journal of Socio-Economics journal homepage: www.elsevier.com/locate/soceco Absenteeism and peer interaction effects: Evidence from an Italian Public Institute Maria De Paola University of Calabria, Economics and Statistics, Bucci, Arcavacata di Rende, 87036 Rende, Cosenza, Italy article info Article history: Received 19 March 2009 Received in revised form 26 October 2009 Accepted 8 February 2010 JEL classification: J41, M41, J45 Keywords: Absenteeism Shirking Peer effects abstract Using microdata for a sample of about 350 workers employed by an Italian public institute, we explain individual absence rates considering both variables that may be related to health conditions and vari- ables that may suggest shirking behaviour. Among these variables we especially focus our attention on the influence produced by the behaviour of randomly assigned peers. In order to handle reflection prob- lems, we combine random assignment with the use of instrumental variables. The proportion of females in the peer group is used as an instrument of peer absence behaviour. From Two-Stage Least Square esti- mates it emerges that social and group interaction plays an important role in shaping individual absence behaviour. © 2010 Elsevier Inc. All rights reserved. 1. Introduction A number of empirical studies argue that absenteeism is a seri- ous problem in many countries. Barmby et al. (2002) show that sickness absence rates range from 1.78 for Switzerland to 6.31 for Sweden. There is also empirical evidence that public sector workers are usually more inclined to taking sick leave than similar employ- ees working in the private sector are (Winkelmann, 1999; Banerjee and Duflo, 2006). In Italy, as is frequently lamented in the press (for example The Economist, 28th August 2008), differences between private and public sector absenteeism are impressive. Data from the Italian Economic Ministry (Ragioneria Generale dello Stato) show that, in 2006, Italian public sector employees took an average of 15.1 days off in sick leave. This was from 30% to 50% more than their private sector counterparts. 1 Public sector absenteeism produces direct and indirect costs: the public administration sustains the costs of wage payment to the absent workers and, in many cases, it has to pay wages to their substitutes (for example, in public schools, temporary contracts are used to hire teachers to cover for absent workers) while indirect I would like to thank Giorgio Brunello, Valeria Pupo, Vincenzo Scoppa and Laura Mazzuca for advise and the INSSI for providing access to the data. The usual dis- claimer applies. E-mail address: [email protected]. 1 Self-reported data from the Bank of Italy’s Survey of Household Income and Wealth (SHIW) point out a difference between absence rates in the private and public sector of 20%. costs, especially in terms of adverse effects on the quality of services offered, also play an important role. In spite of the magnitude of the phenomenon, possibly due to the lack of suitable data, little economic research has been conducted to understand the determinants of absenteeism among public sector employees. As suggested by previous studies, absenteeism can be related either to health factors or to shirking behaviour. Being able to disen- tangle these aspects and to distinguish between voluntary absence and involuntary absence is crucial in order to design adequate pol- icy responses to absenteeism. In fact, the right to absence when sick is a central part of the ‘contract’ between employer and employee, and those who are off work, sick, have the right to expect sensitive treatment and support. On the other hand, since it is often very costly to ascertain an individual’s health conditions, people may be tempted to adopt opportunistic behaviour by taking days off also when they are not genuinely sick, so imposing considerable costs on the organisation employing them. 2 As a consequence, being able to identify drivers for absence frequency and to take the necessary actions would permit public administrations to save large amounts of money. However, it is generally difficult to disentangle absence due to shirking behaviour from genuine sickness leave. Recently, some 2 Psychological theories suggesting that absenteeism may enhance efficiency by providing workers in stressful situations with temporary relief do not seem relevant when absenteeism becomes such a diffuse phenomenon as it is in the Italian public sector. 1053-5357/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.socec.2010.02.004

Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

Embed Size (px)

Citation preview

Page 1: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

AP

MU

a

ARRA

JJ

KASP

1

osSaea

EpE2os

ttsu

Mc

Wp

1d

The Journal of Socio-Economics 39 (2010) 420–428

Contents lists available at ScienceDirect

The Journal of Socio-Economics

journa l homepage: www.e lsev ier .com/ locate /soceco

bsenteeism and peer interaction effects: Evidence from an Italianublic Institute�

aria De Paolaniversity of Calabria, Economics and Statistics, Bucci, Arcavacata di Rende, 87036 Rende, Cosenza, Italy

r t i c l e i n f o

rticle history:eceived 19 March 2009eceived in revised form 26 October 2009ccepted 8 February 2010

a b s t r a c t

Using microdata for a sample of about 350 workers employed by an Italian public institute, we explainindividual absence rates considering both variables that may be related to health conditions and vari-ables that may suggest shirking behaviour. Among these variables we especially focus our attention on

EL classification:41, M41, J45

eywords:bsenteeism

the influence produced by the behaviour of randomly assigned peers. In order to handle reflection prob-lems, we combine random assignment with the use of instrumental variables. The proportion of femalesin the peer group is used as an instrument of peer absence behaviour. From Two-Stage Least Square esti-mates it emerges that social and group interaction plays an important role in shaping individual absencebehaviour.

© 2010 Elsevier Inc. All rights reserved.

hirkingeer effects

. Introduction

A number of empirical studies argue that absenteeism is a seri-us problem in many countries. Barmby et al. (2002) show thatickness absence rates range from 1.78 for Switzerland to 6.31 forweden. There is also empirical evidence that public sector workersre usually more inclined to taking sick leave than similar employ-es working in the private sector are (Winkelmann, 1999; Banerjeend Duflo, 2006).

In Italy, as is frequently lamented in the press (for example Theconomist, 28th August 2008), differences between private andublic sector absenteeism are impressive. Data from the Italianconomic Ministry (Ragioneria Generale dello Stato) show that, in006, Italian public sector employees took an average of 15.1 daysff in sick leave. This was from 30% to 50% more than their privateector counterparts.1

Public sector absenteeism produces direct and indirect costs:

he public administration sustains the costs of wage payment tohe absent workers and, in many cases, it has to pay wages to theirubstitutes (for example, in public schools, temporary contracts aresed to hire teachers to cover for absent workers) while indirect

� I would like to thank Giorgio Brunello, Valeria Pupo, Vincenzo Scoppa and Lauraazzuca for advise and the INSSI for providing access to the data. The usual dis-

laimer applies.E-mail address: [email protected].

1 Self-reported data from the Bank of Italy’s Survey of Household Income andealth (SHIW) point out a difference between absence rates in the private and

ublic sector of 20%.

053-5357/$ – see front matter © 2010 Elsevier Inc. All rights reserved.oi:10.1016/j.socec.2010.02.004

costs, especially in terms of adverse effects on the quality of servicesoffered, also play an important role.

In spite of the magnitude of the phenomenon, possibly due to thelack of suitable data, little economic research has been conducted tounderstand the determinants of absenteeism among public sectoremployees.

As suggested by previous studies, absenteeism can be relatedeither to health factors or to shirking behaviour. Being able to disen-tangle these aspects and to distinguish between voluntary absenceand involuntary absence is crucial in order to design adequate pol-icy responses to absenteeism. In fact, the right to absence when sickis a central part of the ‘contract’ between employer and employee,and those who are off work, sick, have the right to expect sensitivetreatment and support. On the other hand, since it is often verycostly to ascertain an individual’s health conditions, people may betempted to adopt opportunistic behaviour by taking days off alsowhen they are not genuinely sick, so imposing considerable costson the organisation employing them.2 As a consequence, being ableto identify drivers for absence frequency and to take the necessary

actions would permit public administrations to save large amountsof money.

However, it is generally difficult to disentangle absence due toshirking behaviour from genuine sickness leave. Recently, some

2 Psychological theories suggesting that absenteeism may enhance efficiency byproviding workers in stressful situations with temporary relief do not seem relevantwhen absenteeism becomes such a diffuse phenomenon as it is in the Italian publicsector.

Page 2: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

ocio-E

aaeo(2calFpIrbttsi

sasrWrsab

iesrswtcbt

1temagtomcwZage(f

aejtbdtL

ysis of the determinants of absenteeism.We use personnel data from 8 units of the Italian National Social

Security Institute (INSSI), located in a Southern Italian province,which together employ 329 workers. Our data refer to the year

M. De Paola / The Journal of S

uthors have looked at the existence of peer effects on individualbsence as an indicator of shirking behaviour. Group interactionffects have been extensively studied in relation to educationalutcomes and social phenomena such as crime, alcohol, drug useSacerdote, 2001; Gaviria and Raphael, 2001; Kremer and Levy,008). With regards shirking, this kind of effect may operate amongo-workers via different mechanisms based mainly on monitoringnd stigma channels. In fact, both peer monitoring and stigma areower when shirking is a common and widely accepted behaviour.ollowing this type of reasoning, Ichino and Maggi (2000), in aaper studying regional absenteeism differentials within a large

talian bank, show that peer group interaction effects play a relevantole in explaining these differentials. A similar result was obtainedy Bradley et al. (2007) analysing the impact of group interac-ion on the absence behaviour of primary and secondary schooleachers. Finally, Hesselius et al. (2009), on the basis of a large-cale randomised experiment, show that peer effects are relevantn explaining individual absence behaviour.

In this study, we follow a similar approach in trying to under-tand whether shirking behaviour plays a relevant role in shapingbsence rates of public sector employees. We use a dataset on aample of 329 workers employed at the Italian National Social Secu-ity Institute, INSSI (Istituto Nazionale della Previdenza Sociale).

e explain absence rates considering both variables that may beelated to individual health conditions and variables that may hidehirking behaviour. Among these variables we especially focus ourttention on the influence produced by social interaction and groupehaviour.

More precisely, we analyse how the absence rate of employeess affected by the absentee behaviour of co-workers. We considermployees working in the same division as being the relevantphere of social interaction: on the one hand these workers areequired to cooperate and communicate among themselves and,o, they contribute to the shaping of their working environmenthile, on the other hand, they interact on a daily basis and tend

o establish friendly relationships. Therefore, for each worker i, weonsider all the workers employed in the same division as mem-ers of his peer group and peer absence behaviour is calculated ashe average of absence rates of individuals in the peer group.

As is well demonstrated in the economic literature (Manski,993), empirical analyses trying to detect peer group effects facewo main problems: one is related to the fact that individuals gen-rally choose their peers (self-selection problem) and the choiceay be related to unobservable individual factors; the other, known

s reflection problem, emerges because members of the sameroup undertake interdependent behaviour. In our case this meanshat each member’s shirking behaviour influences the shirking ofther members, but, at the same time, is influenced by how otherembers behave. Different strategies have been adopted to over-

ome these problems. Some of these works rely on situations inhich peers are randomly assigned (for example Sacerdote, 2001;

immerman, 2003), other analyses use an instrumental variablepproach in trying to identify exogenous determinants of peerroups (Caze and Katz, 1991; Gaviria and Raphael, 2001; Hanushekt al., 2003) while still other authors add group specific fixed effectswhen more than one observation for group is available) to controlor correlated unobservables (Arcidiacono and Nicholson, 2005).

Self-selection problems are not a major concern in our analysiss our sample employees are not able to choose between differ-nt divisions or even between different units. In fact, to obtain aob at the Italian National Social Security Institute, it is necessary

o pass a national recruitment examination and, once the job haseen obtained, the assignment to different units and to differentivisions within the same unit is related to the particular needs ofhe Institute. To deal with reflection problems, we use a Two-Stageeast Squares estimation and instrument peer absences with the

conomics 39 (2010) 420–428 421

percentage of females in the division. We pointed to this particularpre-determined feature of peer groups since, according to our anal-ysis, the gender variable has a great impact on absence behaviour,while other personal characteristics play a minor role. This vari-able should not directly influence individual i’s absence rate, butmay have a relevant impact on peer absenteeism. To be more con-crete, we do not expect working in a division with a greater femalepresence to directly influence the individual’s decision to be absent,but it may influence one’s propensity to be absent mainly throughthe increased probability that one’s peers will be absent. We alsoexperiment by using the proportion of peer workers on probationas an additional instrument.

From our analysis, it emerges that absenteeism is strongly influ-enced by peer group effects and that workplace absence normsproduce a relevant effect on individual shirking behaviour. Inour instrumental variable preferred specification, we find that anincrease of 1% in the peer group absence rate produces an increasein the individual i absence rate of 0.60 percentage points (the OLSestimates show a smaller coefficient equal to 0.38). This impliesthat when all individual i’s co-workers increase their absence rateby 1% (about 18 h each), individual i will increase his absence by10 h.

As the higher absence rates observed in certain divisions mayalso be due to a contagion effect, we have tested the “social inter-action assumption” by analysing the effect on individual sicknessabsence of peer absence due to family/study leave. From our anal-ysis it emerges that individuals whose peers have higher absencerates due to family/study leave tend to be absent more often forsickness reasons, giving support to the idea that the higher absen-teeism observed in certain divisions is likely to be due to socialinteraction effects rather than to contagion effects.

These results suggest that social interaction is an important fac-tor in the determination of economic outcomes. We find a strongeffect of peer absence behaviour on individual absence decisions,which may be due to the fact that when workers are provided withfull insurance and are subject to low monitoring (as in the Ital-ian public sector), it is easier for each individual to adapt his ownbehaviour to the behaviour of his peers. Therefore, our findingspoint out a moral hazard problem and suggest that there is room forpolicy interventions aimed at reducing absence behaviour. In fact,these policies would be very effective, since by increasing the costof absenteeism, they would influence individual absence decisions,both directly and indirectly, through the behaviour of co-workers.

The paper is organised as follows. In Section 2, data are pre-sented and some descriptive statistics are offered. In Section 3, weestimate a simple OLS model to examine the relationship betweena number of personal and job characteristics and absenteeism. Sec-tion 4 investigates peer group effects and presents both OLS andTSLS estimates. Section 5 concludes.

2. Institutional framework, data and descriptive statistics

While many other works on absenteeism rely on measures ofabsence which are based upon self-reported absence information,we have very accurate administrative data allowing us to avoidproblems relating to misreporting3 and to provide a reliable anal-

3 For example, Johns (1994) reports that employees tend to be absent more thantwice as often as their self-reports indicate and individuals with higher actualabsences tend to be more inaccurate in their self-reports than individuals with loweractual absence.

Page 3: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

422 M. De Paola / The Journal of Socio-E

Table 1Absence rates.

Absence rate Mean Std. Dev. Min Max

Leave due to illness 0.050 0.067 0 0.383Family and study Leave 0.027 0.094 0 0.849Leave due to union activities 0.003 0.014 0 0.175

2aeou(oam

et

wsfitni

wwatfis

gaafrcw

aaaedre

lo

mIiaa

ho

Leave for blood donors 0.0003 0.002 0 0.016Leave related to handicaps 0.020 0.045 0 0.266Total leaves 0.101 0.124 0 0.853

007 and offer detailed information on a number of personal char-cteristics (such as gender, number of children, place of residence,ducation, age, number of hours contracted to work, tenure, wage),n a number of work-place characteristics (such as place of work,nit size, work division) and on different categories of absenceabsence due to sickness, absence due to family/study leave,4 timeff due to participation in union assemblies, leave for blood donorsnd leave for individuals with disabilities or individuals with familyembers who have a handicap).We focus our attention mainly on absence due to sickness. How-

ver, we also provide some descriptive statistics which deal withhe other categories of absenteeism we have at hand.

Until July 2008, Italian public employees were entitled to fullage income for the first nine months of sickness.5 Absence for

ickness had to be justified by a physician’s certificate (within therst two days of sickness) and the employer could ask for an addi-ional “official medical check”, performed by physicians who wereot chosen by workers, but autonomously appointed by subjects

nvolved in monitoring absences due to sickness.Our data do not provide information on length of absence and

e only have information regarding the total number of hours theorker was away from work during the year 2007. As in Bramby et

l. (1995) and Barmby et al. (2002), we define the absence rate Ri ashe ratio of the total number of hours of absence (due to sickness, toamily/study leave, etc.), Ai, recorded by the administrative officen the year 2007, to the total number of contracted hours, Ti, for theame year, then: Ri = Ai/Ti.6

Table 1 presents absence rates by considering different cate-ories of absence based on different types of leave. The averagebsence rate due to sickness is about 5%, corresponding to anverage of about 12 days of sick leave. The absence rates due toamily/study leave and related to handicaps are also quite high,espectively 2.7% and 2%, while the absence rates due to otherauses are negligible. The total absence rate is 10.1% on average,hich corresponds to 25 days off work during the year.

These descriptive statistics show the existence of a substantialmount of heterogeneity in the absentee behaviour of workers. Onverage, 22% of them were never absent for health reasons, while,t the opposite end of the distribution, 10% of our sample employ-

es accumulated more than 22 days absence over the year. Sharperifferences emerge when we consider absence due to family/studyeasons: 32% of workers never took this type of leave, while work-rs in the top 10% of the distribution were absent for about 50 days.

4 This type of absence includes a number of different types of leave, such as paideave in the event of serious illness or death of relatives, leave due to a child’s illnessr other needs, and leave to sit university examinations.5 In June 2008, the Italian government introduced a law (133/2008) imposing aandatory medical check, even in the event there was just a single day of sick leave.

n addition, this law established that, for every day of illness, there would be a cutn “all allowances or emoluments of a fixed, continuative nature” and “all otherncillary economic benefits”. Our data, concerning absences in 2007, do not allownalysis of the effect of this policy intervention.6 However, since 95% of the employees have the same number of contractedours, this measurement is not very different from that based on the total numberf absence hours.

conomics 39 (2010) 420–428

When we look at the total absence rate, it emerges that only about8% of workers were never absent, workers in the lower 50% of thedistribution show, on average, a total absence rate of 2%, whilethose in the upper 50% of the distribution present an average totalabsence rate of 18%.

In Table 2, we present some descriptive statistics regarding theemployees subject to this study. They are, on average, in their fifties,about 50% of them are female and 36% have obtained a universitydegree. 81% of workers have at least one child, while the aver-age number of children is 1.58 (unfortunately we do not have anyinformation with regards their age).

Using data about leave due to union activity and blooddonations, we have defined two dummy variables, Union andBlood donors, which take a value of one when workers haveobtained leave motivated respectively by union activity and byblood donations. Similarly, we infer information on personal orfamily disabilities from the information we have on leave due tothese reasons and define a dummy Handicap with a value of onewhen the worker has taken this kind of leave. It emerges that 19%of workers have obtained leave in relation to personal or family dis-abilities, while 30% of them have participated in some union activityand 4% of them are blood donors.

About 47% of workers live in the town in which the unit employ-ing them is located. 3% of the sample workers were hired in 2007and, in accordance with the National Collective Employment Con-tract for public sector employees, were on probation for the firstsix months after hiring. It is worthwhile noting that the employeris allowed to dismiss the worker without reason or warning duringthis period.

As far as occupational variables are concerned, the workers weconsider are mainly employed in administrative jobs.7 The averagetenure is of 20 years and the average yearly gross wage is about38,400 euros. 73% of the sample workers are employed in a largeunit (employing 243 workers), while the remaining work in smallunits with an average size of 15 workers. Within the same unit,workers are organised into different divisions.

We consider all the workers employed in the same division asindividual i to be peers of employee i.

In Table 3, some descriptive statistics concerning each divisionare reported. Divisions vary in their size, in the average absencerate, in the percentage of female employees and in the percentageof graduate employees.

Neither female employees nor graduate employees are concen-trated in just a few divisions. Workers on probation are employedin seven different divisions, while the remaining division does notemploy workers with this type of contract. Each division generallydeals with a particular type of activity, for example family support,income support, self-employed pensions. In some cases, divisionsperform exactly the same activity, but for subjects who reside in dif-ferent geographical areas. Therefore, tasks managed by employeesin different divisions are quite similar.8

3. The effect of personal and job characteristics on absencerates

In this section we analyse the effect of personal and job charac-teristics on public sector absenteeism, focusing on absence due to

7 Jobs can be classified into three main levels: 13% of the workers are employed atthe lower occupational level, 75% of workers have a job classified at the intermediatelevel, while 12% of them have a top level job.

8 The smaller units are not always organised into divisions and, in these cases, theunit is treated as a division.

Page 4: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

M. De Paola / The Journal of Socio-Economics 39 (2010) 420–428 423

Table 2Summary statistics.

Mean Std. Dev. Min Max Observations

Female 0.523 0.500 0 1 329University Degree 0.362 0.481 0 1 329Number of children 1.583 0.944 0 4 329Age 49.607 7.091 29 66 329Age2 2513.97 696.296 841 4356 329Dummy for workers travelling to work 0.526 0.500 0 1 329Dummy for residence in a different province 0.030 0.171 0 1 329Tenure 20.263 10.622 0 39 329Yearly gross wage/1000 38.448 15.283 7.281 182.081 329Dummy for employees hired in year 2007 0.030 0.171 0 1 329Unit size 184.055 100.948 4 243 329Division size 14.214 4.781 1 22 329Peer absence rate due to sickness 0.053 0.028 0 0.135 329Blood donors 0.042 0.202 0 1 329Union 0.305 0.461 0 1 329Handicap 0.193 0.395 0 1 329

Table 3Descriptive statistics for each division.

Division Employees Sickness absence rate % females % graduate Division Employees Sickness absence rate % females % graduate

Division 1 8 0.051 0.375 0.500 Division 15 14 0.044 0.429 0.286Division 2 20 0.027 0.600 0.350 Division 16 12 0.047 0.583 0. 083Division 3 18 0.053 0.277 0.167 Division 17 12 0.040 0.500 0.167Division 4 15 0.041 0.533 0.400 Division 18 14 0.093 0.571 0.357Division 5 4 0.010 0.500 0 Division 19 13 0.038 0.692 0.308Division 6 8 0.012 0.125 0.500 Division 20 17 0.075 0.705 0.411Division 7 13 0.019 0.461 0.080 Division 21 12 0.067 0.750 0.583Division 8 19 0.007 0.368 0.368 Division 22 14 0.046 0.642 0.571Division 9 22 0.033 0.545 0.727 Division 23 9 0.094 0.666 0.222Division 10 8 0.072 0.250 0.500 Division 24 3 0 0 0

s

R

wiu

etcw

cwpAs2

aUi

5

o

l

r

workplace nor that for being resident outside the province in whichthe units we consider are located show a statistically significanteffect.

Division 11 7 0.012 0.714 0.287Division 12 13 0.084 0.461 0.308Division 13 3 0.104 0.666 0.667Division 14 5 0.048 0.800 0.400

ickness.9 By OLS, we estimate the following simple model:

i = ˛ + ˇ1Pi + ˇ2Ji + ˇ3Dk + ε (1)

here Ri is the absence rate,10 Pi is a vector of individual character-stics, Ji is a vector of job characteristics and Dk is a dummy variablesed to capture unobserved unit effects, with k = 1 . . . 8.

Table 4 reports estimates of Eq. (1). In all specifications, standardrrors (reported in parentheses) are corrected for heteroskedas-icity. We cluster standard errors at division level to correct forommon shocks which may influence the absence behaviour oforkers in the same division.

The main personal characteristics we consider are: sex, age, edu-ation, number of children, a dummy for employees travelling toork11 and a dummy for those whose residence is not in the samerovince as the one in which the Institute providing data is located.mong the job features, we observe tenure, yearly gross wages, unitize and a dummy variable for workers who were on probation in007.

In columns (1) and (2) OLS estimates, including unit fixed effects,re reported. In the first column we do not control for Blood donors,

nion and Handicap, while we do control for these dummy variables

n the second column.The female dummy is positive and statistically significant (at

% and 10% levels according to the specification) implying that

9 See Brown and Sessions (1996), for a survey of theoretical and empirical researchn absenteeism.10 Very similar results are obtained when the dependent variable is defined as aog-odds ratio ln[Ri(1 − Ri)] (estimates are available upon request).11 In an alternative specification, we control for distance from the workplace, butesults remain substantially unchanged.

Division 25 13 0.045 0.384 0.154Division 26 8 0.064 0.750 0.250Division 27 19 0.100 0.578 0.473Division 28 6 0.047 0.500 0.667

female employees are absent from work more often. This resultmay be related either to the gender division of household work, toa weaker attachment to work for females (Leight, 1983) or to bio-logical differences between males and females (Ichino and Moretti,2009).12

Absences increase with age at a decreasing rate and reach amaximum when the employee is about 52 years old.

The number of children has a negative but not statistically signif-icant effect. The negative coefficient shown by this variable mightbe due to the fact that employees can make use of special leave forfamily needs in the Italian public sector.13 Employees with a uni-versity degree have a lower propensity to be absent for sicknessreasons, however, the effect is statistically significant at 10% levelonly in specification (2).14

Neither the dummy variable for individuals travelling to the

12 However, since the female dummy is also positive and statistically significantwhen we consider measures of absences that are not related to health conditions,for example absence due to family or study leave, the higher level of absenteeismon the part of females cannot be exclusively related to biological aspects.

13 In fact, when we consider the total absence rate or the absence rate due to familyand study leave, it emerges that employees with children exhibit a higher rate ofabsence.

14 The negative impact of a higher level of education on absenteeism may berelated to the fact that better educated individuals tend to give more attention tohealth problems and to prevention schemes or it may be due to motivational issues,since more highly educated people generally perform better jobs, which are usuallyassociated with a higher degree of job satisfaction.

Page 5: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

424 M. De Paola / The Journal of Socio-Economics 39 (2010) 420–428

Table 4OLS estimates of the effects of personal and job characteristics on absence rates. Dependent variable: Ri = Ai/Ti .

(1) (2) (3) (4)

Female 0.02001** 0.01720* 0.01840** 0.01587*(0.00842) (0.00846) (0.00817) (0.00813)

University Degree −0.01377 −0.01424* −0.01277 −0.01317(0.00826) (0.00813) (0.00826) (0.00812)

Children 0.00008 −0.00031 −0.00019 −0.00064(0.00188) (0.00189) (0.00188) (0.00189)

Age 0.00621* 0.00595* 0.00676** 0.00617**(0.00315) (0.00296) (0.00306) (0.00293)

Age2 −0.00006 −0.00006* −0.00007* −0.00006*(0.00004) (0.00003) (0.00004) (0.00003)

Yearly gross wage −0.00050* −0.00046* −0.00050* −0.00045*(0.00029) (0.00027) (0.00029) (0.00026)

Tenure 0.00043 0.00056 0.00054 0.00072(0.00062) (0.00062) (0.00057) (0.00058)

Probation −0.04366*** −0.03986*** −0.03742*** −0.03220**(0.01338) (0.01258) (0.01286) (0.01177)

Dummy for employees travelling to work 0.00685 0.00691 0.00416 0.00413(0.00834) (0.00842) (0.00802) (0.00817)

Dummy different province −0.01118 −0.00835 −0.01217 −0.00953(0.01385) (0.01346) (0.01373) (0.01324)

Blood donors −0.01715 −0.01933*(0.01184) (0.01134)

Union −0.00142 0.00051(0.00798) (0.00805)

Handicap 0.01748* 0.01818*(0.00944) (0.00958)

Unit size 0.00008* 0.00008**(0.00004) (0.00004)

Unit dummies Yes Yes No NoConstant −0.12185* −0.07697 −0.12430* −0.11123*

(0.06528) (0.06555) (0.06459) (0.06276)Observations 329 329 329 329

S uped ba

tfia(aa

cslbba(b

tfttisW

d

ai

R-Squared 0.10288

tandard errors (corrected for heteroskedasticity) and incorporating clustering grore statistically significant, respectively, at 1%, 5%, and 10% levels.

As far as job characteristics are concerned, a negative rela-ionship between yearly gross wage and absenteeism emergesrom our estimates which is statistically significant at a 10% leveln both specifications. Similar results have been found by Dragond Wooden (1992) and Chaudhury and Ng (1992), while Leigh1991) finds no impact of wages and paid sick leave on individualbsenteeism.15 On the other hand, once we control for wages andge, tenure does not produce any statistically significant effect.

Employees who were on probation in year 2007 took signifi-antly fewer days of absence than their colleagues (the effect istatistically significant in both specifications at a 1% level). This is inine with findings obtained by works investigating the relationshipetween employment protection legislation and workers’ shirkingehaviour. For example, in analysing the behaviour of employees inlarge Italian bank during and after probation, Ichino and Riphahn

2005) find that absenteeism is significantly lower during the pro-ationary period.

Unit fixed effects are jointly highly significant in both specifica-ions, however, in columns (3) and (4), instead of using dummiesor units, we have experimented by including a variable measuringhe unit size among the regressors. It emerges that an increase inhe unit size produces an increase in the rate of absence. The effects statistically significant at 5% or 10% levels depending upon the

pecification. This corroborates previously reported findings (e.g.

inkelmann, 1999; Hesselius, 2007).Finally, it emerges that absence is higher among individuals with

isabilities or those who have family members with a handicap.

15 We have also experimented with including among the regressors a dummy vari-ble for employees who are division managers. The effect is negative but statisticallynsignificant while no other substantial effects emerge.

0.11482 0.08552 0.09932

y division are reported in brackets. The symbols ***, **, * indicate that coefficients

Being a blood donor produces a negative but statistical insignifi-cant effect on absence rates, while being involved in union activitiesproduces an ambiguous effect since the sign of the dummy variableUnion changes from negative to positive according to the specifica-tion (it is never statistically significant).

The pseudo R2 measures are low in all specifications, but theyare comparable with those found in other studies. Overall, the esti-mates are generally well defined and of the expected sign.

4. Peer effects on absence behaviour

In this section we examine the effects produced by group inter-action on individual absence behaviour. First, we describe oureconometric methodology and, then, we present both OLS and TSLSestimates.

4.1. Empirical methodology

Our analysis now concentrates on establishing whether the indi-vidual’s absence rate is influenced by the absence behaviour of hispeers. The definition of the relevant sphere of interaction poses anumber of problems since it is not clear whether individuals aremainly influence by their friends, by people from their place of res-idence, or by co-workers. Our definition of peer group is based onworkers employed in the same division. We think that subjectswho work together and interact on a daily basis tend to estab-

lish friendly relationships and, as a consequence, the division inwhich an employee works may represent the relevant sphere ofinteraction.

Our empirical specification follows the line taken by Caze andKatz (1991) and Gaviria and Raphael (2001). We extend the simple

Page 6: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

ocio-E

m

R

wicienn

bbta

puesbotic

mdaehaotaoicepf

vaaqbwovaimc

dfaata

d

M. De Paola / The Journal of S

odel represented by Eq. (1) as follows:

i = ˛ + ˇ1Pi + ˇ2Ji + ˇ3DK + ˇ4Ai + ε (2)

here Ai is the average incidence of absenteeism among individual’s co-workers. To be more precise, peer absence behaviour is cal-ulated as the average of the absence rates of workers employedn the same division as worker i (the absenteeism of individual i isxcluded): Ai = 1/Nd

∑Ndn=1Fn where Fn measures absences of peer

= 1 . . . Nd working in division d with subject i and Nd is the totalumber of peers in division d.

According to the specification proposed in Eq. (2), the absenceehaviour of individual i is influenced by the behaviour of his peers,ut is not directly affected by the average pre-determined charac-eristics of his peer group, Pi, which are assumed to influence hisbsences only indirectly through peer interaction.

The estimation of Eq. (2) poses a number of econometricroblems. First, employees may sort according to some personalnobservable characteristics in different divisions generating rel-vant endogeneity problems. Secondly, we have to deal with theo-called reflection problem. In fact, on the one hand, the averageehaviour of peers influences individual behaviour, while, on thether hand, individual absence behaviour also affects group absen-eeism. Thirdly, common shocks or correlated effects may influencendividuals belonging to the same group. For example, in our case,ertain divisions may involve more unpleasant tasks.

We are confident that the first bias is not relevant for our esti-ates since the assignment of workers to one division or another

epends exclusively on the needs of the INSSI. In order to obtainjob with the Italian National Social Security Institute, it is nec-

ssary to pass a national recruitment examination. Once the jobas been obtained the assigning of individuals to different units,nd to different divisions within the same unit, is related to thepening of a vacancy in that particular division, due for exampleo the retirement decisions of older workers. Therefore, employeesre not able to choose as the assignment is random and conditionaln the employee’s meeting some formal requirements, generallyn terms of level and type of education acquired.16 During theirareers, employees can move from one division to another, forxample in relation to career advancement, but, in this case too, therocedure is very formal and it is necessary to pass an examinationor any place that becomes vacant.

Correction for the second source of bias relies on an instrumentalariable approach using some peer average pre-determined char-cteristics as instruments. This estimation strategy is based on thessumption that contextual effects are not relevant and, as a conse-uence, there is no direct relationship between subject i’s absenceehaviour and the average pre-determined characteristics of his co-orkers, Pi. These variables should not produce a direct influence

n the absence behaviour of individual i, but may have a rele-ant impact on peer absenteeism. In fact, individuals working indivision with a greater female presence should not be directly

nfluenced by these aspects in their absence decisions, while theyay be indirectly effected since this pre-determined peer group

haracteristic tends to influence peer group absence.We have experimented with using different sets of average pre-

etermined peer characteristics, however only the proportion ofemales in the division turns out to be relevant in explaining peer

verage absence behaviour. Therefore, we have instrumented peerbsence behaviour with the proportion of females co-workers inhe division.17 As a robustness check we have also used the percent-ge of co-workers on probation in the division employing subject i

16 Italian law offers a high level of protection in terms of equal opportunities amongifferent types of workers. “Regolamento organico del personale”, November 1990.17 Individual i is not included in calculating the instruments.

conomics 39 (2010) 420–428 425

as an instrument. We think that, since contextual effects should notbe relevant in our case, these instruments, f, should comply withthe usual conditions:

Cov(Peer absence rate, f ) /= 0 and Cov(f, ε) = 0.

As the data at hand does not allow us to control for division fixedeffects, we cannot completely overcome problems deriving fromcommon shocks. However, the use of the proportion of females andthe proportion of workers on probation as instruments also helpsavoid problems deriving from common shock bias. In addition, asexplained in Section 2, since divisions perform very similar activi-ties, we do not expect common shocks or correlated effects (such asthe unpleasantness of tasks carried out in specific divisions), whichmay be correlated to the instruments, to bring about the resultsrather than peer interaction.

In what follows we present both OLS and TSLS estimates of Eq.(2).

4.2. OLS and TSLS estimation results

Table 5 presents OLS estimates of alternative specifications ofour model considering the absence rate of subject i, Ri = Ai/Ti, inrelation to the absence behaviour of his peers.

Estimates shown in columns (1) and (2) include unit fixedeffects. When controlling for personal and job characteristics, itemerges that, peer effects appear positive and significant. Anincrease in the absence rate of individual i’s peer group is posi-tively correlated with his absenteeism (statistically significant ata 5% level). To be more precise, an increase of one percentagepoint in peer absence rate is associated with an increase in indi-vidual absence rate of 0.39%.18 Similar results (but larger in size)are obtained when, instead of controlling for unit fixed effect, wecontrol for unit size (columns 3 and 4). In these specifications thepeer effect is statistically significant at a 1% level.

In column (5), using an interaction term Peer absencerate × Tenure (where the mean has been subtracted from tenure),we investigate whether the magnitude of peer effects varies in rela-tion to the worker’s tenure. The coefficient on this interaction termis positive, implying that workers with higher than the averagetenure are more influenced by peer absence behaviour. However,the effect is of weak statistical significance.

As workers cannot self-select in divisions, we can say that ourdata reject the hypothesis of no relationship between the individ-ual’s absence behaviour and peer absenteeism. However, due to thereflection problem deriving from regressing outcomes to outcomes,we cannot give these coefficients a causal interpretation.

Table 6 displays TSLS estimates using the proportion of femalesin the division in which individual i is employed as an instrument.In the estimates presented in this Table, unit fixed effects, ratherthan unit size, are controlled for, but results are not influenced bythis choice.

From First Stage regressions, in all specifications, it emerges thatthe Proportion of female co-workers greatly determines Peer absencerate. F-Statistics for the test of whether the instrument coefficientis equal to zero are always well above the threshold value of 10suggested by Staiger and Stock (1997).

The effects of personal and job characteristics are, for the mostpart, very similar to the OLS estimates. The TSLS estimates of

peer effects are higher than OLS estimates, suggesting that simul-taneity problems produce a downward effect on the estimates ofpeer influence (similar results are obtained by Gaviria and Raphael(2001)). The higher TSLS estimates may also be related to measure-

18 The other explanatory variables have approximately the same level of signifi-cance as in Table 4.

Page 7: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

426 M. De Paola / The Journal of Socio-Economics 39 (2010) 420–428

Table 5OLS regressions relating individual absenteeism to peer group behaviour dependent variable: Ri = Ai/Ti .

(1) (2) (3) (4) (5)

Peer absence rate 0.38906** 0.38061** 0.45272*** 0.43551*** 0.39860***(0.16636) (0.16691) (0.13537) (0.13744) (0.12403)

Female 0.01922** 0.01640* 0.01789** 0.01532* 0.01467*(0.00855) (0.00862) (0.00824) (0.00826) (0.00856)

University Degree −0.01505* −0.01527* −0.01439* −0.01443* −0.01336(0.00811) (0.00793) (0.00800) (0.00784) (0.00822)

Children 0.00069 0.00036 0.00056 0.00015 0.00038(0.00157) (0.00157) (0.00153) (0.00155) (0.00154)

Dummy for employees travelling to work 0.00563 0.00570 0.00376 0.00365 0.00339(0.00868) (0.00871) (0.00832) (0.00839) (0.00844)

Age 0.00497* 0.00484** 0.00501* 0.00470* 0.00491*(0.00252) (0.00232) (0.00249) (0.00239) (0.00255)

Age2 −0.00005 −0.00005 −0.00005 −0.00005 −0.00005(0.00003) (0.00003) (0.00003) (0.00003) (0.00003)

Yearly gross wage −0.00001 −0.00001* −0.00001 −0.00001 −0.00001*(0.00000) (0.00000) (0.00000) (0.00000) (0.00000)

Tenure 0.00048 0.00057 0.00055 0.00068 0.00011(0.00061) (0.00062) (0.00056) (0.00058) (0.00063)

Probation −0.03556*** −0.03332*** −0.02981** −0.02649** −0.03035**(0.01187) (0.01198) (0.01077) (0.01054) (0.01122)

Dummy different province −0.01028 −0.00708 −0.01098 −0.00801 −0.00882(0.01495) (0.01434) (0.01471) (0.01399) (0.01344)

Blood donors −0.01534 −0.01678 −0.01695(0.01192) (0.01143) (0.01176)

Union −0.00433 −0.00311 −0.00277(0.00781) (0.00774) (0.00759)

Handicap 0.01597* 0.01571* 0.01679*(0.00931) (0.00907) (0.00934)

Unit size 0.00003 0.00004 0.00004(0.00003) (0.00003) (0.00003)

Peer absence rate × tenure 0.00001(0.00000)

Unit dummies Yes Yes No No NoConstant −0.09784* −0.06777 −0.09898* −0.08941* −0.08318

(0.05181) (0.05277) (0.05279) (0.05035) (0.05255)Observations 329 329 329 329 329R-Squared 0.12298 0.13361 0.11558 0.12632 0.12892

Standard errors (corrected for heteroskedasticity) and incorporating clustering grouped by division are reported in brackets. The symbols ***, **, * indicate that coefficientsare statistically significant, respectively, at 1%, 5%, and 10% levels.

Table 6TSLS regressions relating individual absenteeism to peer group behaviour dependent variable: Ri = Ai/Ti .

(1) (2) (3) (4) (5)

Panel A: Two-Stage Least SquaresPeer absence rate 0.69009** (0.32774) 0.63738* (0.35623) 0.65136** (0.26623) 0.05543* (0.02993) 0.66761** (0.28697)Female 0.01861** (0.00827) 0.01586* (0.00875) 0.01869** (0.00857) 0. 01727** (0.00839) 0.01580* (0.0087)University Degree −0.01602* (0.00829) −0.01596 (0.00804) −0.01592* (0.00829) 0.01531* (0.00800) −0.01604* (0.00807)Number of children −0.00161 (0.00142) 0.00081 (0.00144) 0.00110 (0.00142) 0.00074 (0.00172) 0.00087 (0.00143)Age 0.00400 (0.00292) 0.00408 (0.00263) 0.00412 (0.00277) 0.00427 (0.00278) 0.00400 (0.00251)Age2 −0.00004 (0.00003) −0.00004 (0.00003) −0.00004 (0.00003) −0.00004 (0.00003) −0.00004 (0.00003)Dummy for employees travelling to work 0.00468 (0.00889) 0.00489 (0.00891) 0.00481 (0.00882) 0.00664 (0.00830) 0.00479 (0.00888)Dummy different province −0.00958 (0.01598) −0.00622 (0.01520) −0.00967 (0.01575) 0.00822 (0.01416) −0.00612 (0.01507)Tenure 0.00052 (0.00605) 0.00058 (0.00063) 0.00051 (0.00061) 0.00056 (0.00060) 0.00061 (0.00063)Yearly gross wage −0.00042 (0.00028) −0.00042 (0.00026) -0.00001* (0.00000) −0.00045* (0.00025) −0.00045* (0.00026)Probation −0.02928** (0.01377) −0.02890 ** (0.01339) −0.03010* (0.01395) −0.02727** (0.01400) −0.02839** (0.01409)Blood donors −0.01412 (0.01207) −0.02479 (0.01450) −0.01854 (0.01164) −0.01398 (0.01209)Union −0.00628 (0.00872) −0.00966 (0.01078) −0.00398 (0.00829) −0.00651 (0.00871)Handicap 0.01494 (0.00965) 0.01124 (0.01169) 0.01590* (0.00912) 0.01484 (0.00966)Unit dummies Yes Yes Yes Yes Yes

Peer absence rate

Panel B: first stage regressionsProportion of females co-worker 0.05356*** (0.00994) 0.05249*** (0.00994) 0.05360*** (0.01162) 0.59956*** (0.05965) 0.05674*** (0.0096)Proportion of co-workers on probation −0.08189*** (0.01593)Observations 329 329 243 329 329R-Squared 0.254 0.262 0.114 0.454 0.318

Notes: Panel A reports the Two-Stage Least Squares estimates, instrumenting. Panel B reports the corresponding first stage. Standard errors, corrected for heteroskedasticity,are reported in brackets. The symbols ***, **, * indicate that coefficients are statistically significant, respectively, at 1%, 5%, and 10% levels.

Page 8: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

ocio-E

meaw

iaiiB

weHalderctph

iaesoppt

tm

5

leatts

pcassb

aeia

idtuwsta

M. De Paola / The Journal of S

ent errors, since our measure of peer absence behaviour is basedxclusively on absences due to sickness, while other types of peerbsence may also influence an individual’s decision not to go toork.

For example, it emerges in column (1) that an increase of 1%n peer group absence rate produces an increase in individual i’sbsence rate of 0.69 percentage points, while the OLS estimatesndicated a smaller effect, about 0.38. A smaller coefficient (0.63)s shown in column (2), where we control for Handicap, Union andlood donors.

In column (3) estimates on a smaller sample, including onlyorkers employed at the larger unit, are reported. As already

xplained, workers are not able to select between different units.owever, if any selection is possible, we think that it would bet unit level, since workers may try to get a job in a unit which isocated near to their place of residence. To avoid problems that mayerive from this type of selection, we concentrate on the main unitmploying the majority of our sample workers. The main resultsemain substantially unchanged with the peer effect in this specifi-ation being just slightly higher compared with estimates referringo the full sample. This larger effect could be due to the fact thateer effects are more likely to apply when monitoring costs areigher due to unit size.

Finally, since, in divisions in which workers are often sick, thenfluence of peer absenteeism on individual absence behaviour maylso be due to a contagion effect, in column (4) estimates of theffect produced by peer absences for family/study reasons on theickness absence rate of individual i are reported. Our measurementf peer absence behaviour is now represented by the proportion ofeers taking family/study leave. It emerges that individuals whoseeers are more likely to be off work due to family/study leave tendo be more often absent for health reasons.

In columns (5) we experiment using the proportion of peers inhe division on probation as an additional instrument. The esti-

ated coefficient of peer group absences is 0.66.

. Concluding remarks

In this paper we have analysed the absence behaviour of pub-ic sector employees using a dataset on a sample of 329 workersmployed at an Italian Public Agency. The availability of accuratedministrative data allows us to provide a very reliable picture ofhe determinants of absenteeism in the public sector, which is par-icularly relevant given the lack of statistical work available on theubject.

From our analysis it emerges that absenteeism is a function ofersonal and job characteristics such as gender, yearly gross wage,ontractual arrangements and unit size. While some of these vari-bles may be related to individual health conditions and, therefore,uggest that absences occur for valid reasons, others, such as unitize and contractual arrangement, may hide employee shirkingehaviour.

To better investigate the relationship between sicknessbsences and shirking we have looked at the existence of peerffects on individual absences. A positive relationship betweenndividual absenteeism and peer group absenteeism suggests thatbsence is more likely to be due to shirking than to sickness.

We have considered division as the relevant sphere of socialnteraction, since individuals who work in the same place and haveaily interaction are more likely to influence each other. Thankso the fact that employees were assigned to different divisions and

nits exclusively on the basis of institute needs and that individualsere not in the condition to choose, we do not have to deal with

elf-selection problems. On the other hand, we deal with reflec-ion problems using the proportion of females in the division asn instrument of peer absence behaviour. This variable should not

conomics 39 (2010) 420–428 427

directly affect the individual i’s absence decisions, but may have aninfluence through group interaction effects.

Our analysis suggest that peer group effects play a crucial rolein determining individual absence rate. From OLS estimates, itemerges that an increase of one percentage point in peer groupabsence rate produces an increase in individual i absence rate rang-ing from 0.38 to 0.45, depending upon the specification adopted.TSLS estimates point to higher effects, suggesting that simultane-ity problems and measurement errors produce a downward biason the estimates of peer influence.

These results are in line with those emerging from the pre-vious literature on the subject, showing that individual absencebehaviour is related to the absenteeism of co-workers (Ichinoand Maggi, 2000; Bradley et al., 2007). In addition, we showthat the influence of peer absenteeism on individual absencebehaviour is not due to contagion effects. In fact, also we findrelevant peer effects on sickness absences when we measurepeer absenteeism behaviour considering absences for family/studyreasons.

Peer effects emerging from our analysis are larger than thosehighlighted by previous works. This may be due to the specificnature of the sample of employees studied. In fact, when data werecollected, these workers were fully insured against income lossesderiving from illness and, therefore, it could have been tempt-ing for them to adapt their behaviour to that of their absenteepeers.

Our findings suggest that policy interventions aimed at reducingabsence in the Italian public sector might be very effective, sincethey would have both a direct and an indirect effect. Some analysistrying to understand the effects of the 2008 Italian reform, whichreduces sick leave insurance for public employees in case of sick-ness, supports this prediction in that they show a very large impact(see De Paola et al., 2009).

References

Arcidiacono, P., Nicholson, S., 2005. Peer effects in medical school. Journal of PublicEconomics 89 (2/3), 327–350.

Barmby, T., Ercolani, M., Treble, J., 2002. Sickness absence: an international compar-ison. Economic Journal 112, F315–F331.

Banerjee, A., Duflo, E., 2006. Addressing Absence. Journal of Economic Perspectives20 (1), 117–132.

Bradley, S., Green, C., Leeves, G., 2007. Worker absence and shirking: evidence frommatched teacher-school data. Labour Economics 14 (3), 319–334.

Bramby, T., Orme, C., Treble, J., 1995. Worker absence histories: a panel data study.Labour Economics 2, 53–65.

Brown, S., Sessions, J., 1996. The economics of absence: theory and evidence. Journalof Economic Surveys 10 (1), 23–53.

Case, A., Katz, L., 1991. The Company You Keep: The Effects of Family and Neighbor-hood on Disadvantaged Youths. NBER Working Paper, no. w3705.

Chaudhury, M., Ng, I., 1992. Absenteeism predictors: least squares, rank regression,and model selection results. Canadian Journal of Economics 25 (3), 615–635.

De Paola, M., Pupo, V., Scoppa, V., 2009. Absenteeism in the Italian Public Sector: TheEffects of Changes in Sick Leave Compensation. Mimeo.

Drago, R., Wooden, M., 1992. The determinants of labor absence: economic factorsand workgroup norms across countries. Industrial and Labor Relations Review45 (4), 764–778.

Gaviria, A., Raphael, S., 2001. School-based peer effects and juvenile behavior.Review of Economics and Statistics 83 (2), 257–268.

Hanushek, E., Kain, F., Markman, J., Rivkin, S., 2003. Does peer ability affect studentachievement? Journal of Applied Econometrics 18 (5), 527–544.

Hesselius, P., 2007. Does sickness absence increase the risk of unemployment? Jour-nal of Socio Economics 36 (2), 288–310.

Hesselius, P., Johansson, P., Nilsson, J., 2009. Sick of your colleagues’ absence? Journalof the European Economic Association 7 (2/3), 583–594.

Ichino, A., Maggi, G., 2000. Worker environment and individual background:explaining regional shirking differentials in a large Italian firm. Quarterly Journalof Economics, 1057–1090.

Ichino, A., Moretti, E., 2009. Biological gender differences, absenteeism and the earn-

ing. American Economic Journal: Applied Economics 1 (1), 183–218.

Ichino, A., Riphahn, R., 2005. The effect of employment protection on worker effort:a comparison of worker absenteeism during and after probation. Journal ofEuropean Economic Association 2, 120–143.

Johns, G., 1994. How often were you absent? A review of the use of self-reportedabsence data. Journal of Applied Psychology 79 (4), 574–591.

Page 9: Absenteeism and peer interaction effects: Evidence from an Italian Public Institute

4 ocio-E

K

L

L

M

Staiger, D., Stock, J., 1997. Instrumental variables regression with weak instruments.

28 M. De Paola / The Journal of S

remer, M., Levy, D., 2008. Peer effects and alcohol use among college students.Journal of Economic Perspectives 22 (3), 189–206.

eigh, J.P., 1991. Employee and job attributes as predictors of absenteeism in a

national sample of workers: The importance of health and dangerous workingconditions. Social Science and Medicine 33, 127–137.

eight, J., 1983. Sex differences in absenteeism. Industrial Relations: A Journal ofEconomy and Society 22 (3), 349–361.

anski, C., 1993. Identification of endogenous social effects: the reflection problem.Review of Economic Studies 60, 531–542.

conomics 39 (2010) 420–428

Sacerdote, B., 2001. Peer effects with random assignment for Dartmouth roommates.Quarterly Journal of Economics 116, 681–704.

Econometrica 65 (3), 557–586.Winkelmann, R., 1999. Wages, firm size and absenteeism. Applied Economics Letters

6 (6), 337–341.Zimmerman, D, 2003. Peer effects in academic outcomes: evidence from a natural

experiment. Review of Economics and Statistics 85 (1), 9–23.