1 Is work absence contagious? Patrik Hesselius, Per Johansson and Johan Vikström Department of Economics, Uppsala University and IFAU.

1

Is work absence contagious?

Patrik Hesselius, Per Johansson and Johan VikströmDepartment of Economics, Uppsala University and IFAU

2

Motivation for investigating social interactions

• Large differences, in for example work absence, between countries, regions and over time

• The differences could hardly be explained only by institutions!?

• Social interactions could be one explanation

3

Sickness absence in six countries

1

1,5

2

2,5

3

3,5

4

4,5

5

5,5

6

1983

1985

1987

1989

1991

1993

1995

1997

1999

2001

%

FR

DE

NL

NO

SE

UK

4

Sickness absence in Sweden 1970-2005

0%

1%

2%

3%

4%

5%

6%

7%

8%

1970 1975 1980 1985 1990 1995 2000 2005

Men

Women

5

Identification problem

Four explanations to segregation in work absence:

• Endogenous interactions

Absence varies with the overall behavior

• Exogenous interactions

Varies with exogenous characteristics in the network

• Correlated effects

A) Institutions differs between networks

B) Similar individual characteristics

6

Earlier studies on int. in social insurance

Unemployment:

Clark (2003), Topel (2001), Topel & Conley (2001)

Åberg, Hedström & Kolm (2003)

Work absence

Ichino & Maggi (2000), Lindbeck, Palme & Persson (2004)

Conclusion:

• Social interactions seem to be important

• Identification need that certain members for some exogenous reason starts to be more absent /or change network

7

This paper

Adds to the literature on empirical evidence of social interactions by utilize a randomized experiment in Gothenburg 1988, that previously have shown to increase sickness absence of those randomized into treatment.

• It introduces exogenous variation

• Enables identification of endogenous social interactions

• We find evidence of endogenous effects

8

Outline

• Swedish sickness insurance (in 1988)• The randomized experiment• Identification strategy• Data and first evidence• Theory• Results• Summing up

9

Swedish sickness insurance (in 1988)

• Benefits if absent from work due to sickness

• Covers all employed workers, and nationally financed by proportional pay roll taxes

• 90 percent replacement rate (up to a ceiling)

• No qualification day and no experience rating (all benefits paid from the government)

• Benefits paid for a week before checking the claimants’ eligibility through doctor’s certificate (though possible for the public insurance offices to monitor more strictly)

10

The randomized experiment• In the second half of 1988 in Gothenburg municipality • Randomized (born on even date) • The treated was allowed to receive benefits for two

weeks instead of one week without showing a doctor’s certificate

• Non-blind experiment (large information campaigns)

Effects of less strict control of eligibility;i) Increases the duration of absence (short-term absence,

until day 14 of a spell)ii) No significant effect on the incidence into sickness

absence

11

Fraction still absent before the experiment

0,0

0,2

0,4

0,6

0,8

1,0

7 14 21 28

Day of sickness absence

Fra

ctio

n st

ill ab

sent

due

to s

ickn

ess Control

Treatment

12

Fraction still absent during the experiment

0.0

0.2

0.4

0.6

0.8

1.0

7 14 21 28

Day of sickness absence

Fra

ctio

n st

ill a

bsen

t due

to s

ickn

ess

Control

Treatment

13

Identification

Four requirements need to be fulfilled in order to use a intervention for identification of social interactions.

(i) Cannot change the group composition

(ii) Can only affect a proportion of the individuals in each network and the proportion have to differ between at least some of the groups

(iii) Exogenous with respect to unobserved variables

(iv) The treated must be representative for the network

14

Our setup• Ethnicity (country of origin) as network definer• Use immigrants in the whole Gothenburg MA

=> Different proportion of treated, depending on were their network members live

• Data for all individuals also before the experiment

=> Control for individual (and network) heterogeneity in sickness absence behavior

• Randomly selected treated

=> Treated and non-treated representative for the whole network

15

Data and sample selection

• Population database; from IFAU database (individual and employment characteristics) and National Security Board (sickness absence)

• Use working immigrants in age 20-60

• Immigrants from 84 countries with more than 10 members in Gothenburg MA

• Large immigrant groups: Nordic countries, Hungary, Former Yugoslavia, Poland, Germany, Iran, Turkey and Chile

• Large variation in proportion treated (23 – 52 percent) and in mean short-term absence (2.7 – 9.1percent)

16

A first look at the data

Estimate Standard error t-ratio Non-treated1 Intercept -.540 .048 -11.30 1988 1.011 .067 15.07 1989 .412 .066 6.34 Treated2 Intercept -.431 .080 -5.41 1988 2.445 .110 21.85 1989 0.300 0.11 2.72

Notes: 1n = 73 343, 2n = 32 144

17

A first look at the data (cont)

Non-treated Treated Estimate Standard error Estimate Standard error Intercept 0.333 ***0.093 0.91 ***0.20 2nd quartile 0.183 0.121 1.63 ***0.25 3rd quartile 0.072 0.189 0.81 ***0.30 4th quartile 0.232 *0.138 1.22 ***0.27 R2 0.016 % 0.48%

Notes: *** significant at the 1% level, * significant at the 10 % level

18

Theory

• Include social utility into a regular labor supply model

• Follows Brock and Durlauf (2000) • Assumes rational expectations and that all

individuals in the network is given equal weight• Individual absence in presence of endogenous

interactions depend on the mean absence in the network

19

Estimation

• Control for network heterogeneity (network fixed effects)

• Control for seasonal effects (seasonal dummies)

• Control for network specific trends (allow for correlation between treated and non treated absence also before the reform

• Sensitivity analysis (artificial treatments for 1987, 1989 and Stockholm)

1 1 2 1 , 1,...12.ijm j ijm im ijm ijmy D month m

1( / ) ( , 1)jm j j jn n m R

20

Results - OLS

Network Size jn > 10 jn >30 10< jn <1000

Time period and Area

Parameter Estimate Std Error Estimate Std Error Estimate Std Error

1988 1 0.161 **0.079 0.114 0.085 0.110 0.091

2 -0.053 0.085 0.003 0.099 -0.030 0.090

1987 1 0.013 0.092 0.008 0.109 0.034 0.110

2 -0.048 0.085 -0.008 0.108 -0.070 0.090

1989 1 -0.025 0.078 -0.024 0.082 -0.068 0.095

2 0.044 0.079 0.128 0.088 -0.050 0.086

Stockholm MSA 1 -0.021 0.072 -0.031 0.075 0.015 0.127

2 0.551 ***0.090 0.716 ***0.105 0.292 ***0.117

Notes: Robust standard errors. ** and *** denotes significantly different from zero at the 5 and 1 percent level.

21

Results – Hazard regressions




1988 1 -0.186 **0.072 -0.192 **0.084 -0.174 **0.084

2 0.102 0.078 0.054 0.096 0.108 0.084

1987 1 -0.042 0.108 -0.022 0.120 -0.035 0.126

2 -0.114 0.084 -0.204 0.102 -0.015 0.090

1989 1 0.060 0.072 0.051 0.078 0.021 0.090

2 0.036 0.078 -0.001 0.084 -0.006 0.084

Stockholm MSA 1 -0.042 *0.024 -0.042 * 0.024 -0.066 * 0.042

2 0.042 0.030 0.030 0.036 0.030 0.036 Notes: *, ** and *** denotes significantly different from zero at the 10, 5 and 1 percent level.

22

Results – Hazard regressions (cont)

Sensitivity analysis not controlling for general trends




1988 1 -0.120 **0.053 -0.156 ***0.058 -0.102 *0,060

Note: *** significant at the 1% level and * significant at the 10% level.

Sensitivity analysis control for; gender, age, age square, government employed, income and parish

Network Size jn > 10


Parameter Estimate Std Error

1988 1 -0.258 ***0.072

2 0.156 **0.078

Note: *** significant at the 1% level and * significant at the 10% level.

23

Results – Incidence into work absence




1988 1 0.022 0.016 0.023 0.018 0.017 0.018

2 0.012 0.017 0.014 0.020 0.010 0.019

Notes: *, ** and *** denotes significantly different from zero at the 10, 5 and 1 percent level.

24

Summing up and conclusions

• A 10 percent increase in the means absence

=> decrease hazard from work absence by about 1.5 percent because of endogenous interactions.

• No social interaction effect on the incidence into work absence

• Sensitivity analysis shows that the results are robust

1 Is work absence contagious? Patrik Hesselius, Per Johansson and Johan Vikström Department of Economics, Uppsala University and IFAU.

Documents