Empirical strategy
While it is commonly observed that peers such as friends, classmates, or work colleagues behave similarly, it is empirically challenging to identify the different channels through which the effects operate. A subject’s adaptation of behavior can depend on the peers’ behavior, attitudes, and characteristics; common opportunities; or available information.
1 Given that many people spend the greater part of their time with colleagues at work, social and occupational surroundings can be expected to influence the health behavior of employees. We focus on workers (hereafter movers) who join a new firm and are potentially influenced in health screening behavior by their new work colleagues (hereafter, stayers).
For individual
i who moves to firm
j in period
t, we estimate the following equation:
$$\begin{aligned} s_{ijt} = \alpha + \beta p_{jt-1} + {\varvec{X}}_{it-1} \gamma + {\varvec{F}}_{jt} \delta + \zeta s_{ijt-1} + \mu _{ijt}. \end{aligned}$$
(1)
The dependent variable
\(s_{ijt}\) is a binary variable indicating whether a mover participated in a medical screening examination in period
t, that is, after he joined the new firm. In the empirical analysis, we use participation in general health checks, mammography screening, and prostate-specific antigen tests as outcome variables. To avoid potential influence of movers on stayers’ behavior [
20], we measure peer behavior at a point when the subject has not yet become part of the group. In particular,
\(p_{jt-1}\) captures the past behavior of the new work colleagues in firm
j. This is defined as the average past screening behavior of the stayers, that is, the number of screenings of stayers in firm
j in period
\(t-1\) divided by the number of stayers
\(K_j\):
$$\begin{aligned} p_{jt-1}=\frac{\sum _{k=1}^{K_j} s_{kjt-1}}{K_j}. \end{aligned}$$
Given that we use the lagged behavior of stayers, the peer effect variable is exogenous with respect to the movers’ behavior and the coefficient of interest
\(\beta\) does not include within-firm feedback effects.
We further control for the lagged dependent variable, \(s_{ijt-1}\), a set of pre-treatment individual characteristics, \(X_{it-1}\), and firm-level covariates, \(F_{jt}\). In the empirical analysis, period t covers a mover’s first two calendar years in the new firm. Period \(t-1\) covers the two preceding calendar years. Stayers are defined as those who are employed in the same company for at least 4 years (i.e., in periods \(t-1\) and t). They were already employed in firm j in period \(t-1\) and worked for another 2 years with the movers who joined firm j at the beginning of period t.
Identification of social interaction For peer effects to occur, employees need to communicate about their health screening participation. These medical check-ups are preventative in nature and not associated with diseases. Taboos and social stigma of diseases may prevent employees from talking about their personal health problems.
One major concern with the approach, however, is the selection of individuals into firms due to choices of the movers themselves or firm decisions. Movers may generally collect information about the importance of primary and especially secondary health prevention in a potential new firm, including the behavior of firm employees and the existence of firm-level (health) policy measures. Firm movers might, in accordance with their own attitude toward medical prevention or health consciousness, self-select into an appropriate firm.
Conversely, the hiring firm may strongly tend to employ new staff in accordance with its corporate health and health consciousness policy for its staff. A firm attaching great importance to the healthy lifestyle of its staff to preserve their good health will tend to hire healthier employees. The companies’ procedures for recruiting this type of employees will probably put more emphasis on the applicants’ tendency toward overweight, smoking behavior, and alcohol consumption. Both employee- and firm-driven sorting entail that individuals with similar or same characteristics and attitudes would move into similar jobs.
In our empirical analysis, we control for important characteristics such as economic sector of the firm and the mover’s age, sex, past health care expenditure, and participation in previous screening exams that may affect both the choice of a firm and screening participation. The identifying assumption is that, conditional on these covariates, sorting into firms is not related to new work colleagues’ screening behavior. Job movers start working in a firm that (randomly) exposes them to peers with a higher or lower screening affinity. While in principle, we cannot rule out selection based on unobservables, we would like to argue that the most important drivers of screening participation, for example, health consciousness, are not key factors in job matching. The major determinants of an individual’s move from one firm to another are probably not those impacting the decision of whether to undergo a screening exam. A survey of employees in Austria suggests that 20% of the employees consider leaving their jobs. As the most important motives, people state their low wage (49%) and lack of career opportunities (29%) [
22].
Another critical issue is that this approach cannot convincingly disentangle the peer effects from firm policy effects. A significant coefficient of our variable of interest [
\(p_{jt-1}\) in Eq. (
1)] may reflect either the direct influence of peers on individual health behavior or shared influences at the firm level, such as health promotion measures. Some firms may undertake a persistently increased commitment to health promotion activities. Such firm policies could influence movers’ health behavior in the absence of any true peer effect.
Peers or firm policy? To explore this issue further, we divide the employees of a firm into groups based on their characteristics and construct different peer behavior measures. Selection mechanisms and firm policies related to health behavior should largely affect the entire workforce. In contrast, we expect that work peers with similar characteristics would exert a greater influence on an individual having the same characteristics. As regards gender, for example, we define
\(p_{jt-1}^f\) and
\(p_{jt-1}^m\) to measure the average behavior of female and male work colleagues, respectively, and estimate
$$\begin{aligned} s_{ijt}^{g} = \alpha + \beta _1 p_{jt-1}^f + \beta _2 p_{jt-1}^m + {\varvec{X}}_{it-1} \gamma + {\varvec{F}}_{jt} \delta + \mu _{ijt}, \end{aligned}$$
(2)
for female (
\(g=f\)) and male (
\(g=m\)) job movers. Similarly, we split the workforce according to job type (blue-collar vs. white-collar jobs) and age (above and below 40 years of age). We use general health check participation as our outcome variable
\(s_{ijt}^{g}\) in Eq. (
2), given that this program targets all age groups and both sexes.
Institutional background
Austria represents a comprehensive Bismarck-type social welfare system that includes mandatory health insurance for almost the entire population. Membership of private employees in one of the nine regional health insurance funds cannot be freely chosen, but is determined by the location of their workplace.
2 All insured patients have access to a wide range of health care services in the inpatient and outpatient sectors. With a few exceptions, such as a small co-payment for hospitalization and prescription fees for medical drugs, health insurance covers all medical care expenses.
Insured persons above 18 years of age are entitled to the general health check program (in German, Allgemeine Vorsorgeuntersuchung). The scope and procedures of this program are regulated legally. Since its introduction in 1974, the program underwent several revisions based on developments in medical knowledge. The program offers free voluntary participation in yearly general health checks. The medical examination includes an anamnesis and a series of age- and sex-specific diagnostic and laboratory tests focusing on the identification of health risks and early detection of diseases. Following a major revision in 2005, health promotion has become an additional goal and medical doctors are asked to provide information and counseling on lifestyle choices. The questions and procedures for screening physicians were expanded and stated more precisely. Furthermore, regular invitations were sent out to increase participation of the insured.
3
Apart from the general health check, women over 40 years of age are entitled to a mammography screening every 2 years. This screening is aimed at early detection of breast cancer using X-ray imaging. While the general health check is usually performed by a general practitioner (GP), mammography screening must be done by a radiologist.
The general health check does not by default include a prostate-specific antigen (PSA) blood test for prostate cancer. Instead, the GP provides information about the pros and cons of this test and may refer male patients to an urologist for the PSA test and further examination. In addition, men can always undergo a PSA test independent of the general health check program.
Data and descriptives
The Austrian Social Security Database (ASSD) is a linked employer–employee dataset containing the labor market history of all private sector workers in Austria, along with individual- and firm-level characteristics [
26]. We match this information with data of the regional health insurance fund for Upper Austria (in German, Oberösterreichische Gebietskrankenkasse), which include detailed information about health care utilization in the inpatient and outpatient sectors. Individual-level medical attendance data cover each single visit at the GP or medical specialist and information about participation in the general screening exam, mammography screening, and the PSA blood test, with the date of service utilization.
ASSD and health insurance data are available for the period from 1998 to 2012. First, we construct an annual panel data set of all private sector workers and their associated firms. If individuals have two jobs or move from one firm to another during a calendar year, we select the job with the higher annual earnings as their major occupation. We use this data set to identify the job movers who comprise the unit of observation in the empirical analysis. As described in “Empirical strategy”, our baseline specification allows for 2-year windows in the outcome variable, that is, we estimate an individual’s screening participation during the 2 years following the move, given that medical check-ups are typically not done annually. We additionally require that movers stay in the new firm for at least 2 years; that is, we disregard a small number of workers who switched jobs twice within two years. Additional results with 3-year time windows are presented in “Robustness checks” section.
4
In total, we observe 181,496 persons moving to 4222 firms. Table
1 provides the descriptive statistics of the movers and stayers based on our main sample. As the table shows, 18.7% of movers and 20.7% of stayers participate in a general health screening exam in a 2-year period.
5 The 2-year participation rates for female movers and stayers in mammography screening are 17.2% and 28.5%, respectively. The male employees’ participation rates for PSA tests are lower, at 7.9% and 14.6% for movers and stayers, respectively. The most obvious reason for the participation rate of movers being significantly lower than that of stayers is the lower age of the former group. Movers are on average 6 years younger than stayers. The lower daily wage of movers (70€ versus 80€) may also be related to age. The 2-year outpatient expenditure (medical attendance and medication) of movers and number of days they spent in hospital are significantly lower than those of the stayers, obviously because the movers are on average significantly younger. A higher percentage of movers live in urban areas (the cities of Linz, Wels, and Steyr, with a population of over 30,000 each), and, as compared to the stayers, previously worked in smaller firms.
Table 1Descriptive statistics
Outcome variables | | |
General health check | 0.187 | 0.207 |
Mammography\(^\mathrm{a}\) | 0.172 | 0.285 |
PSA test\(^\mathrm{b}\) | 0.079 | 0.146 |
Average characteristics | | |
Age (years) | 33.8 | 39.7 |
Female | 0.419 | 0.401 |
Daily wage (€) | 70 | 80 |
Outpatient expenditures (€) | 694 | 751 |
Days in hospital | 2.304 | 2.333 |
Urban area (Linz, Wels, Steyr) | 0.178 | 0.119 |
Firm size (# employees) | 549 | 1135 |
Job type | | |
Blue collar | 0.487 | 0.467 |
White collar | 0.452 | 0.424 |
N | 181,496 | 602,855 |