BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Beskæftigelsesudvalget 2020-21
BEU Alm.del Bilag 328
Offentligt

International Journal of

Environmental Research

and Public Health

Article

Expected Labor Market Afﬁliation: A New Method Illustrated

by Estimating the Impact of Perceived Stress on Time in Work,

Sickness Absence, and Unemployment of

37,605 Danish Employees

Jacob Pedersen

*, Svetlana Solovieva

, Sannie Vester Thorsen

, Malene Friis Andersen

and Ute Bültmann

National Research Centre for the Working Environment, DK-2100 Copenhagen, Denmark

Finnish Institute of Occupational Health, 00032 Helsinki, Finland

Department of Health Sciences, Community and Occupational Medicine, University Medical Center

Groningen, University of Groningen, 9713 GZ Groningen, The Netherlands

Correspondence: [email protected]

Citation:

Pedersen, J.; Solovieva, S.;

Thorsen, S.V.; Andersen, M.F.;

Bültmann, U. Expected Labor Market

Afﬁliation: A New Method Illustrated

by Estimating the Impact of Perceived

Stress on Time in Work, Sickness

Absence, and Unemployment of

37,605 Danish Employees.

Int. J.

Environ. Res. Public Health

2021,

18,

4980. https://doi.org/10.3390/

ijerph18094980

Academic Editor: Paul B. Tchounwou

Received: 29 March 2021

Accepted: 5 May 2021

Published: 7 May 2021

Abstract:

As detailed data on labor market afﬁliation become more accessible, new approaches are

needed to address the complex patterns of labor market afﬁliation. We introduce the expected labor

market afﬁliation (ELMA) method by estimating the time-restricted impact of perceived stress on

labor market afﬁliation in a large sample of Danish employees. Data from two national surveys

were linked with a national register. A multi-state proportional hazards model was used to calculate

ELMA estimates, i.e., the number of days in work, sickness absence, and unemployment during

a 4-year follow-up period, stratiﬁed by gender and age. Among employees reporting frequent

work-related stress, the expected number of working days decreased with age, ranging from 103 days

lost among older women to 37 days lost among younger and middle-aged men. Young and middle-

aged women reporting frequent work- and personal life-related stress lost 62 and 81 working days,

respectively, and had more days of sickness absence (34 days and 42 days). In conclusion, we showed

that perceived stress affects the labor market afﬁliation. The ELMA estimates provide a detailed

understanding of the impact of perceived stress on labor market afﬁliation over time, and may inform

policy and practice towards a more healthy and sustainable working life.

Keywords:

longitudinal; registers; multi-state; labor market; behavioral analysis; prediction; per-

ceived stress

1. Introduction

The ageing workforce poses many challenges for modern societies, in terms of facili-

tating healthy ageing and work sustainability [1]. The increase in life expectancy, improved

health of older workers, and national actions to prolong working life may lead to new,

complex patterns of labor market afﬁliation, characterized by multiple and competing

states, such as recurrent sickness absence and unemployment in between periods of work.

A multi-state approach, used in the estimation of working life expectancy, has many

advantages in terms of showing the impact on lifelong labor market afﬁliation [2,3], when

compared to the traditional Sullivan method [4]. Recent studies have used such methods

to show the impact of factors like depressive symptoms, educational level, physical work

demands, and occupational class on lifelong labor market afﬁliation [5–8].

Working life expectancy methods are based on longitudinal data, with age as the

underlying time axis. In addition, these methods are used for investigating factors such

as education level that will remain the same throughout the working life. No similar

conceptual method has yet been developed to examine only a restricted follow-up period

of labor market afﬁliation, using, e.g., dates or time since a major event as the underlying

Publisher’s Note:

MDPI stays neutral

with regard to jurisdictional claims in

published maps and institutional afﬁl-

iations.

Licensee MDPI, Basel, Switzerland.

This article is an open access article

distributed under the terms and

conditions of the Creative Commons

Attribution (CC BY) license (https://

creativecommons.org/licenses/by/

4.0/).

Int. J. Environ. Res. Public Health

2021,

18,

4980. https://doi.org/10.3390/ijerph18094980

https://www.mdpi.com/journal/ijerph

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

2 of 13

time axis, and for investigating factors of a temporary nature; though the idea of summing

up the results of a multi-state model for restricted periods is not new [9–11].

This paper introduces the expected labor market afﬁliation (ELMA) method. The

ELMA method uses a multi-state model for analyzing patterns of labor market afﬁliation

in relation to temporary and permanent risk factors, and restricted follow-up periods,

and the method additionally contains multiple possible selections for the underlying time

axis. The ELMA method is useful for studying complex work participation patterns over

time, and in a multi-state model containing both transitions between recurrent states (e.g.,

work, sickness absence, and unemployment) and transitions to absorbing states, like early

retirement schemes. In the present study, we illustrate the application of the ELMA method

by estimating the impact of perceived stress on labor market afﬁliation.

Perceived stress, deﬁned as the degree to which situations in one’s life are appraised

as stressful [12], is associated with an increased risk of sickness absence and early retire-

ment [13–17]. A recent large national Danish study showed that the prevalence of perceived

stress is particularly high among working adults aged 18–44 years, with a prevalence of

35% and 24% for women and men, respectively [18]. From 2010 to 2017, the prevalence of

perceived stress in Denmark increased from 21% to 25%, primarily among occupationally

active individuals aged 16 to 65 years [18].

Previous studies on perceived stress and labor market outcomes have investigated

the association with single labor market outcomes, such as sickness absence, disability

pension, or return to work [16,17]. However, in a highly ﬂexible and dynamic labor market

such as the Danish labor market, individuals are likely to experience recurrent periods of

sickness absence and unemployment in between periods of work, making analysis of single

outcomes an oversimpliﬁcation of the complex work participation patterns [3,11,19,20].

In the present study, we aimed to estimate the impact of perceived stress on the

labor market afﬁliation of 37,605 Danish employees applying the expected labor market

afﬁliation method. We used a large sample of Danish employees reporting work- and

personal life-related stress, linked with a register of data on labor market afﬁliation states.

Furthermore, we investigated the possible modifying effects of gender and age on the

estimation of perceived stress and on labor market afﬁliation.

2. Material and Methods

2.1. Study Design

The study was based on the linking of data from the Work Environment and Health in

Denmark (WEHD) surveys (2012 and 2014) with longitudinal four-year follow-up data from

the Danish Labor Market Accountant (LMA) register. The WEHD survey was conducted

every second year from 2012 until 2018 on a large sample of Danish employees aged 18

to 64 years. In this study the WEHD 2012 wave with a response rate of 51% (n = 25,804)

and the WEHD 2014 wave with a response rate of 57% (n = 29,192) were used. The total

WEHD sample contained 43,209 respondents, of which 11,787 individuals responded to

both waves, in 2012 and 2014.

The WEHD sample was linked to a combination of six registers from Statistics Den-

mark: (1) the Labor Market Accountant (LMA) register containing daily records of salary

and major social beneﬁt payments (unemployment, pensions, and sickness absence beneﬁts,

etc.) of all Danes from 2008 until 2018; (2) the education register, containing dates of the

highest education level completion for all Danes; (3) the death register, with death dates

for all deceased Danes; (4) the emigration and immigration register, containing dates on

all emigrations and immigrations in Denmark; (5) the Register of Work Absences (RoWA),

containing registrations of employment-related sickness absences down to one day of

duration. The RoWA cover all public employees and a yearly weighted sample from the

private sector with a coverage of approximately 37% of all private employees [17]; and (6)

the employment register, containing dates on employment periods for the individuals in

the RoWA register. It is important to note that in the multi-state model an individual is cate-

gorized as sick-listed, only when the sickness absence refers to the health of the individual.

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

3 of 13

The RoWA register differentiates between sickness absences due to an individual’s own

sickness and due to child sickness (the latter categorized as ‘temporary out’). Moreover,

in the LMA register, sickness absence beneﬁts are registered due to sickness absence of

the individual only, not due to child sickness. The linked data contains individual and

date-based records from 1 January 2010 to 1 December 2018 (both days included). The

linkage was conducted by an encrypted version of the central person register number,

given to all Danes at birth or when fulﬁlling the criteria for immigration during a stay of

more than 3 months [21].

Of the 43,209 individuals who participated in the WEHD 2012 or/and the 2014 wave,

we excluded: 377 individuals who had emigrated and did not have any registered records

during the follow-up period, 4111 individuals because they did not fulﬁll the age criterion

of 18 to 59 years at the start of the follow-up period, 903 individuals who retired prior to

the start of follow-up, and an additional 213 individuals who had inconsistent or missing

answers to the stress questions (n = 5) or reported personal life stress only (n = 208). The

ﬁnal study population consisted of 37,605 individuals (55% women). All individuals were

followed for four years, starting from the individual survey answering date.

The ﬁnal study population was stratiﬁed into six subsamples by gender and three

age-groups: 18–39 years old (young employees), 40–49 years old (middle-aged employees),

and 50–59 years old (older employees).

2.2. Labor Market Afﬁliation States

Seven states were used to model labor market afﬁliation during the four-year follow-

up period (Figure

1).

Each box in Figure

represents a speciﬁc, mutually exclusive labor

market state and arrows represent the possible transitions. Four states are recurrent, i.e.,

multiple individual transitions to and from the states were possible: (1) work, periods

when receiving salary; (2) sickness absence, short periods of sickness absence from one to

thirty days and prolonged sickness absence periods when the employer is compensated

by receiving sickness absence beneﬁt. Sickness absence beneﬁt compensates the employer

for paying salary to sick-listed employees after the 30th day of sickness absence or is

paid directly to the sick-listed when the individual is unemployed. (3) Unemployment,

periods when receiving unemployment beneﬁt as only income but available for the labor

market. The unemployment beneﬁt may rely on an insurance with a restricted duration

or on the public system with no restricted duration; and (4) temporarily out of the labor

market—containing all other recurrent periods of e.g., maternity leave, education, and

emigration. The remaining three states are absorbing, i.e., no other transitions are possible

after entering the state: (5) disability pension, time receiving a disability pension beneﬁt,

including ﬂex-job; (6) retirement pension, time receiving an early retirement pension; and

(7) death. The pension state is mostly relevant for individuals of at least 55 years at the

start of the follow-up period, as the special Danish voluntary retirement scheme makes it

possible to retire at 60 years. All labor market states were recorded by date.

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

Int. J. Environ Res. Public Health

, x FOR PEER REVIE

4 of 13

4 of

Figure 1.

Labor m rket affiliation multi-state model including the descriptive illustration of the

Figure 1.

Labor market afﬁliation multi-state model including the descriptive illustration of the

number of individuals at each state at the start of follow-up and transitions between states during

follow-up. The numbers represent the number of transitions for women/men; the parentheses rep-

follow-up. The numbers represent the number of transitions for women/men; the parentheses

resent the percentages of recurrent transitions.

represent the percentages of recurrent transitions.

2.3. Perceived Stress

Perceived stress was measured with two questions in the WEHD 2

sur-

Perceived stress was measured with two questions in the WEHD 2012 and 2014 sur-

veys. First, individuals were asked “How often have you felt stressed in the last two

weeks?” with answers on a Likert scale: alw ys, often, sometimes, seldom, or never. Sec-

weeks?” with answers on a Likert scale: always, often, sometimes, seldom, or never. Sec-

ond, individuals were asked “What was the most important source of your stress? with

ond, individuals were asked “What was the most important source of your stress?”, with

three response categories: (1) work (2) personal life or (3) work and personal life. Based

three response categories: (1) work, (2) personal life, or (3) work and personal life. Based on

on the responses to both questions, particip nts were classified as: (1) no stress ( some-

the responses to both questions, participants were classiﬁed as: (1) no stress (“sometimes,

times, seldom, or never responses to the first

(2) work-related stress (“always or often”

seldom, or never” responses to the ﬁrst question),

question), (2) work-related stress (“alw ys

or often”

the ﬁrst question, and “work” response to the second

to the second question),

responses to

responses to the first question, and “work response

question), and (3) work-

and ( ) work- and personal life-related stress (“alw ys or often

question and

the first

and personal life-related stress (“always or often” responses to the ﬁrst

responses to

“work

question and “work and personal life” response to

Individuals who reported personal

and personal life” response to the second question).

the second question) Individuals who

reported personal

deleted due to the small number, and

small number, and

who had

life stress only were

life stress only were deleted due to the

so were individuals

so were in-

dividuals

or missing answers.

inconsistent

who had inconsistent or missing answers.

2.4.

4. Covariates

Covariates

We included six cov riates

the analysis, of which three variables were taken from

We included six covariates in

in the analysis, of which three variables were t ken from

the survey data ) working time rrangement (part-time

7 h per week; full

≥

the survey data: (1)

(

working time arrangement (part-time: <37 h per week; full time:

time

week); (2)

(2) body mass index (BMI, kg/m ) (underweight BMI < ; normal weight:

per

per week);

body mass index (BMI, kg/m

) (underweight: BMI < 18; normal weight:

BMI < 25; overweight

; and obese BMI >

); and (3) smoking (yes

BMI

18.5

≤

BMI < 25; overweight: 25

≤

BMI

< 29.9; and obese: BMI

29.9); and (3) smoking

(yes: “daily” and “sometimes”;

“prior smokers and “never”) Two v

Two variables were

“daily” and “sometimes”; no:

no: “prior smokers” and “never”).

riables were obt ined

obtained from the

and education registers: (4) employment sector

sector (private/public),

from the LMA

LMA and education registers: (4) employment

(private/public) nd (5)

and (5) highest accomplished education (low/middle/high).

last v

last variable (6) “survey

highest accomplished education (low/middle/high) The

The

riable ( ) “survey year

year” was constructed to account

the

the WEHD survey: “2012”, “2014”, and “2012 + 2014”.

was constructed to account for

for

WEHD survey:

and “2

. The

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

5 of 13

The employment sector and highest accomplished education variables were allowed to

change during the follow-up period.

2.5. Statistical Analyses

A separate analysis was conducted for each of the six subsamples. We followed the

approach illustrated in Pedersen and Bjorner 2017 and estimated baseline instant transitions

matrices for each time point from day one to day 1461, i.e., four years of follow-up with

day one as the questionnaire answering date (2012 for those answering both the 2012 and

2014 wave). As the ELMA method relies on individual calculations for every combination

of all covariates, the consequent grouping became too small for reliable estimations. To

accommodate BMI and smoking were included in the analysis as normalized inverse

probability weights, using employment sector as an equalizing variable [22].

To gain a set of instant transitions matrices corresponding to the possible combination

of covariates, we adjusted the baseline matrices with a corresponding set of parameter

estimates from multi-state Cox regressions. This was done for each of the six subsamples.

Each Cox regression was stratiﬁed by transition and ridge regression was added to avoid

overﬁtting [23]. In addition, the Cox regressions were adjusted for the combined weight of

the inverse probability weights multiplied by the weight of the RoWA register. We used

the Chapman–Kolmogorov equation (Equation (1)) for multiplying the instant transitions

matrices (A(u)), to gain transition- and state-speciﬁc probabilities (P(s,t)) for each unique

covariate proﬁle in the data in the time interval from day 1 (s) to day 1461 (t). The instant

state-speciﬁc probability was estimated by one minus all instant transition probabilities of

leaving the state.

(

s, t

) =

∏

(

)

(1)

We estimated the transition- and state-probabilities for every unique covariate proﬁle

that we found in the six subsamples [24], including the corresponding standard deviations

using the Greenwood variance for the empirical covariance matrix used in the recursion

formula [25]. Next, we estimated the area under the transition probability curves and the

state probabilities curves, using one hundred random resamples within the corresponding

normal distribution of the mean and standard deviation. The distinct one hundred resam-

ples of the area estimates were assigned to all individuals matching the covariate proﬁle.

The area estimation was made from the survey answering date and in days to the end of

the four-year follow-up period. As the time axis, we used days from the survey answering

date (day one). We treated each data record/row as a late entry and censored only at the

end of the follow-up period.

The integral (E(h)), deﬁned by the area under the probability curves, expresses the

expected time spent in each state (Equation (2)). The state-speciﬁc area corresponds to

the expected state duration time, given that the person was in the state from the start of

the follow-up (in which: h = j, and h = (W, S, U or TO)). Likewise, the area estimate from

the transition speciﬁc probability curves corresponds to the expected duration time in

the “transition to state” (j), given that the person was in the initial state (h) at the start of

follow-up.

(

) =

(

s, t

)

(2)

We used the Beyersmann and Putter approach [26] to gain the non-restricted duration

time in each state of the model for every resample.

We used a standard analysis of variance to estimate the expected duration time for each

state, with a repeated statement concerning the resamples to obtain a bootstrap estimation

of the 95% conﬁdence limits [9]. All covariates included in the Cox regression analyses

were used as independent variables, and the proﬁle and non-restricted state duration time

were used as dependent variables. In addition to the gender and age-group stratiﬁed

analyses, we conducted a standard analysis of variance for each subsample for each of the

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

6 of 13

seven labor market afﬁliation states. The intercept parameter of the standard analysis of

variance expresses the absolute expected state duration time in days, for an individual

with variables matching the reference values. The individual parameter estimate expresses

the average expected number of days. These parameters are either added or subtracted

from the intercept value dependent on the value of the speciﬁc covariate [9]. We used SAS

version 9.4 with the Phreg and Genmod procedure for the regression analysis, otherwise

custom-made code was used to conduct all analyses.

3. Results

Sample Characteristics

A total of 17,022 men and 20,583 women were included in the study. The older

employees were the largest group and the middle-aged employees the smallest group

(Table

1).

Overall, stress was reported by 14% of the employees, ranging from 10.7% (older

men) to 16.9% (young women). On average, 8% reported work-related stress and 6% work-

and personal life-related stress either often or always.

Table 1.

Descriptive characteristics of the study population (n = 37,605).

Women

Group

Level

Young

Employees

(%)

6782

5638 (83.1)

550 (8.1)

594 (8.8)

4767 (70.3)

1155 (17.0)

860 (12.7)

197 (2.9)

3849 (56.8)

1204 (17.8)

632 (9.3)

900 (13.3)

577 (8.5)

2609 (38.5)

3572 (52.7)

24 (0.4)

2113 (31.2)

3938 (58.1)

731 (10.8)

3711 (54.7)

2175 (32.1)

896 (13.2)

2552 (37.6)

1620 (23.9)

2610 (38.5)

Middle-

Aged

Employees

(%)

6691

5649 (84.4)

539 (8.1)

503 (7.5)

5209 (77.9)

1220 (18.2)

262 (3.9)

78 (1.2)

3656 (54.6)

1721 (25.7)

932 (13.9)

304 (4.5)

535 (8.0)

2884 (43.1)

3257 (48.7)

15 (0.2)

1973 (29.5)

3805 (56.9)

913 (13.6)

4311 (64.4)

2071 (31.0)

309 (4.6)

2082 (31.1)

1902 (28.4)

2707 (40.5)

Older

Employees

(%)

7110

5974 (84.0)

686 (9.6)

450 (6.3)

5445 (76.6)

1422 (20.0)

243 (3.4)

107 (1.5)

3731 (52.5)

2048 (28.8)

932 (13.1)

292 (4.1)

1207 (17.0)

2844 (40.0)

3033 (42.7)

26 (0.4)

1652 (23.2)

4571 (64.3)

887 (12.5)

4326 (60.8)

2432 (34.2)

352 (5.0)

2054 (28.9)

2171 (30.5)

2885 (40.6)

Young

Employees

(%)

5530

4888 (88.4)

310 (5.6)

332 (6.0)

3934 (71.1)

1225 (22.2)

371 (6.7)

42 (0.8)

2755 (49.8)

1788 (32.3)

562 (10.2)

383 (6.9)

821 (14.8)

2509 (45.4)

2158 (39.0)

42 (0.8)

2806 (50.7)

1379 (24.9)

1345 (24.3)

4492 (81.2)

646 (11.7)

392 (7.1)

2025 (36.6)

2230 (40.3)

1275 (23.1)

Men

Middle-

Aged

Employees

(%)

5428

4776 (88.0)

373 (6.9)

279 (5.1)

4170 (76.8)

1096 (20.2)

162 (3.0)

7 (0.1)

2031 (37.4)

2423 (44.6)

795 (14.6)

172 (3.2)

704 (13.0)

2596 (47.8)

2089 (38.5)

39 (0.7)

2946 (54.3)

1174 (21.6)

1308 (24.1)

4939 (91.0)

269 (5.0)

220 (4.1)

1760 (32.4)

2181 (40.2)

1487 (27.4)

Older

Employees

(%)

6064

5416 (89.3)

431 (7.1)

217 (3.6)

4519 (74.5)

1319 (21.8)

226 (3.7)

12 (0.2)

2050 (33.8)

2816 (46.4)

940 (15.5)

246 (4.1)

1068 (17.6)

3002 (49.5)

1941 (32.0)

53 (0.9)

2909 (48.0)

1695 (28.0)

1460 (24.1)

5414 (89.3)

312 (5.1)

338 (5.6)

1791 (29.5)

2411 (39.8)

1862 (30.7)

Total:

Work-related

Work and

personal

life-related

Non-smoker

Smoker

Not

Available

Underweight

Normal

weight

Overweight

Obese

Not

Available

Low

Middle

High

Not

Available

Private

Public

Not

Available

Full-time

Part-time

Not

Available

2012

2014

2012 + 2014

Perceived

stress

Smoking

BMI

Education

Employment

sector

Work-time

arrangement

Survey year

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

7 of 13

Figure

shows the distribution of women and men at the start of the follow-up pe-

riod, according to the four recurrent labor market afﬁliation states (work, unemployment,

sickness absence, and temporarily out). The arrows illustrate the transitions during the

follow-up period, with the number of transitions made by women/men and the parenthe-

ses showing the percentage of recurrent events. The majority of individuals (92% of women

and 94% of men) started from a work state. Among individuals who started from sickness

absence and temporarily out states there were nearly twice as many women as men. The

most frequent transitions were between work and sickness absence states, followed by

transitions between work and temporarily out, and between work and unemployment

states. Women transitioned more frequently between work and sickness absence than men,

261,197 total transitions versus 115,165 total transitions, respectively. A total of 729 women

(4%) retired from work during follow-up compared to 428 men (3%).

Table

and Figure

show the expected average number of days spent in the four

recurrent labor market states (ELMA estimates) compared to employees not reporting

perceived stress. The table and ﬁgure are stratiﬁed by gender and age-group.

Table 2.

Expected average number of days (95% conﬁdence interval) spent in the four recurrent labor market afﬁliation

states (ELMA estimates) for perceived stress, stratiﬁed by gender and age group.

Gender/Age

Women

18–39

Work-related

Work and personal

life-related

40–49

Work-related

Work and personal

life-related

50–59

Work-related

Work and personal

life-related

Men

18–39

Work-related

Work and personal

life-related

40–49

Work-related

Work and personal

life-related

50–59

Work-related

Work and personal

life-related

. (-)

−

37.1

(

−

40.4:

−

33.8)

−

46.3

(

−

49.4:

−

43.1)

. (-)

−

37.2

(

−

39.7:

−

34.7)

−

117.1

(

−

122.5:

−

111.7)

. (-)

−

79.6

(

−

82.1:

−

77.1)

−

67.3

(

−

70.4:

−

64.2)

. (-)

7.3 (7.0:7.6)

26.7 (25.6:27.8)

. (-)

19.0 (18.0:20.0)

39.8 (38.2:41.4)

. (-)

22.0 (21.3:22.7)

45.1 (43.3:46.8)

. (-)

10.2 (9.7:10.7)

24.3 (23.2:25.4)

. (-)

1.9 (1.4:2.5)

18.2 (16.8:19.7)

. (-)

21.0 (19.9:22.0)

16.9 (15.5:18.3)

. (-)

−

16.8

(

−

18.3:

−

15.2)

−

21.7

(

−

23.6:

−

19.7)

. (-)

9.2 (8.2:10.3)

. (-)

−

50.8

(

−

52.8:

−

48.8)

−

61.9

(

−

64.0:

−

59.9)

. (-)

−

76.7

(

−

79.8:

−

73.6)

−

81.3

(

−

85.2:

−

77.3)

. (-)

−

103.1

(

−

105.7:

−

100.6)

−

53.0

(

−

55.8:

−

50.1)

. (-)

44.6 (43.6:45.6)

33.8 (32.8:34.7)

. (-)

51.5 (50.2:52.8)

41.7 (40.6:42.9)

. (-)

45.6 (44.8:46.5)

24.2 (23.6:24.8)

. (-)

12.1 (11.7:12.6)

5.6 (5.4:5.9)

. (-)

5.7 (5.1:6.3)

1.9 (1.5:2.4)

. (-)

10.6 (10.3:10.8)

13.1 (12.7:13.4)

. (-)

−

11.5

(

−

12.2:

−

10.9)

13.1 (12.6:13.6)

. (-)

13.5 (12.9:14.1)

1.5 (0.8:2.1)

. (-)

Perceived Stress

Work

Sickness Absence

Unemployment

Temp. Out

−

0.4 (

−

0.7:

−

0.2)

−

2.9 (

−

3.1:

−

2.7)

−

0.2 (

−

0.8:0.3)

. (-)

6.8 (6.0:7.6)

−

2.1 (

−

2.6:

−

1.5)

Among individuals reporting frequent work-related stress, the expected number of

working days lost increased with age, being the highest among older women (

−

103 days)

and the smallest among younger and middle-aged men (

−

37 days). More lost working

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ Res. Public Health

, x FOR PEER REVIE

Int. J. Environ. Res. Public Health

2021,

18,

4980

8 of

8 of 13

work- and personal life-related stress (2 d ys), respectively. While among men, older

days were found

work and personal

middle-aged men

had the largest differences in the

employees with

among young and

life-related stress

and women reporting frequent

work and

of sickness bsence days (45 days) nd young employees with work-rel ted stress

number

personal life-related stress than among the same aged employees reporting only

work-related stress.

had the smallest difference ( d ys).

Figure 2.

The expected average number of days spent in the four recurrent labor market states:

Figure 2.

The expected average number of days spent in the four recurrent labor market states: work,

work, sickness absence (sick) unemployment, and temporary out (temp. out). Comparison of indi-

sickness absence (sick), unemployment, and temporary out (temp. out). Comparison of individuals

viduals reporting perceived: work relates stress, and work and personal-life related perceived

reporting perceived: work relates stress, and work and

By gender and age-group.

stress with individuals not reporting perceived stress

personal-life related perceived stress with

individuals not reporting perceived stress. By gender and age-group.

Young and middle-aged men had more sickness absence and unemployment days

Among individuals reporting frequent work- and personal life-related stress, an N-

when reporting frequent work and personal life-related stress than the same aged men

shaped association between age and difference in the number of working days lost was

reporting only work-related stress. In contrast, women with frequent work and personal

observed, particularly for men (

−

46,

−

117, and

−

67 days among young, middle, and older

life-related stress h d fewer sickness absence and unemployment days than the s me aged

employees, respectively). Irrespective of stress source, different shapes of associations

women reporting only work-related stress.

between age and the difference in the number of sickness absence days were observed for

Young women reporting frequent work-related stress can expect 51 fewer d ys at

men and women (being N-shaped and linearly increasing, respectively). Among women,

work, 4

and

d ys of

differences in the

more d ys of unemployment and

the largest

smallest

sickness absence,

number of sickness absence days were

days less

seen in

of being temporarily out than young women not reporting stress. Overall, the number of

middle-aged employees with work-related stress (52 days) and older employees with work-

lost working days among individu ls increased with age.

among men, older

of sickness

and personal life-related stress (24 days), respectively. While

While, the number

employees

absence d ys

personal

among the middle-aged nd slightly declined among older

with work and

incre sed

life-related stress had the largest differences in the number of

women; the number of days of

and young employees with work-related stress had the

sickness absence days (45 days)

unemployment with the smallest difference w s observed

among

difference (7 days).

smallest

the middle-aged women.

Among women reporting frequent work and personal life-related stress, higher loss

Young and middle-aged men had more sickness absence and unemployment days

of working time w s found than among similar aged women reporting frequent

men

when reporting frequent work and personal life-related stress than the same aged

work-

related stress

work-related stress. In contrast, women with

and middle-aged women.

reporting only

only. This was particularly observed in young

frequent work and personal

Young men reporting frequent work-related stress c n expect 37 fewer days at work,

life-related stress had fewer sickness absence and unemployment days than the same aged

more sickness

only work-related stress.

women reporting

absence days, and 10 more d ys of unemployment th n similar aged men

reporting rare or no stress. Middle-aged men had more sickness absence

fewer days at

Young women reporting frequent work-related stress can expect 51

but less unem-

ployment days, while older men had more working

of unemployment, and

absence, and

work, 45 more days of sickness absence, 12 more days

time lost, more sickness

12 days less

lso more unemployment compared

women not reporting stress. Overall, the

no stress.

being temporarily out than young

to the same aged men reporting rare or

number

lost working days among individuals increased with age. While, the number of sickness

absence days increased among the middle-aged and slightly declined among older women;

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

9 of 13

the number of days of unemployment with the smallest difference was observed among

the middle-aged women.

Among women reporting frequent work and personal life-related stress, a higher loss

of working time was found than among similar aged women reporting frequent work-

related stress only. This was particularly observed in young and middle-aged women.

Young men reporting frequent work-related stress can expect 37 fewer days at work,

7 more sickness absence days, and 10 more days of unemployment than similar aged

men reporting rare or no stress. Middle-aged men had more sickness absence but less

unemployment days, while older men had more working time lost, more sickness absence,

and also more unemployment compared to the same aged men reporting rare or no stress.

Men had more sickness absence time and more unemployment time when reporting

frequent work and personal life-related stress than the same aged men reporting work-

related stress only. A similar comparison of middle-aged men showed that they lost more

than double the working time and had more sickness absence time. For the older men the

number of sickness absence days was even higher, while the working time lost was less,

when compared to the same aged men reporting work-related stress only.

Table S1A,B in the Supplementary Material show that only small differences of +/

−

days were observed in the expected time spent in three absorbing states (disability re-

tirement, retirement pension, and death) between employees reporting and not reporting

perceived stress. In addition to the perceived stress variables, Table S1A,B also show the

ELMA estimates for the additional covariates by category.

The Table S1A,B are interpreted like this: the reference group consisted of individuals

who reported no stress, with a middle education level, worked in the private sector in

full-time employment, and participated in the 2014 WEHD survey: young women can on

average expect 1227 days of work, 44 days of sickness absence, 18 days of unemployment,

and 135 days of being temporarily out. Men of the same age can on average expect 1189

days of work, 29 days of sickness absence, 24 days of unemployment, 91 days of being

temporarily out, and in addition to the women, on average, a death postponed by 2 days.

4. Discussion

In this study, we introduced the expected labor market afﬁliation (ELMA) method. We

illustrated the use of the ELMA method by estimating the impact of perceived work-related

and personal life-related stress on the average expected time spent at work, on sickness

absence, and unemployment during a four-year follow-up period. We used a unique

linkage of nationally representative survey data and national register data.

Overall, we found multiple changes in labor market afﬁliation for both genders during

the four-year follow-up period and for all age groups reporting frequent stress when

compared to individuals of the same gender and age group not reporting stress. Both

men and women with frequent perceived stress experienced a decline in their average

expected working time and an increase in time of sickness absence and unemployment.

For the young and the middle-aged individuals frequent work-related stress, alone and

in combination with personal life-related stress, was associated with a major decline in

working time and an increase of sickness absence days. A similar, but less pronounced

impact of perceived stress was seen for individuals reporting work-related stress only. In

older individuals work-related perceived stress showed a greater impact than work- and

personal life-related stress on the working time. While an increased time of unemployment

was more frequently observed among the male employees, an increased time of sickness

absence was more frequently observed among the women.

4.1. Advantages and Disadvantages of the ELMA Method

The ELMA method builds on multi-state methods for calculating working life ex-

pectancy or working years lost [6–8]. The main difference concerns the time period and

underlying time axis for which estimates are calculated and the way covariates are handled.

Working life expectancy estimates correspond to the expected time spent at work during a

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

10 of 13

life course when having a certain age. In contrast, the ELMA estimates are calculated for a

time-restricted period, which could be relatively short, and the method uses a time-axis

instead of an age-axis. In working life expectancy estimations a ﬁnal estimate is calcu-

lated and reported for each covariate included. The ELMA method will typically include

more explainable covariates than the working life expectancy method, and an analysis of

variance is used to sum up the inﬂuence of the covariates.

The ELMA method uses all information of time duration and transitions between labor

market states based on the multi-state model for the entire follow-up period. Analyses

with multi-state modeling of three or more states are typically characterized by a large

number of transitions, and may lead to very large tables for displaying the results [11,27].

The ELMA method is able to sum up numerous results from multi-state models without

compromising details on transitions, the number of states, confounders, or proportionality

assumptions concerning the survival analysis part. Though the arrangement of data,

the model assumptions, and the mathematical calculations in the ELMA method are

comprehensive, the ELMA results are easy to interpret and communicate to stakeholders,

due to their direct and absolute nature.

The ELMA method will not produce reliable estimates if there are only a few (e.g.,

three) individuals for any combination of covariates. It is, therefore, reasonable to use

only a restricted amount of measured covariates in the ELMA method, which will depend

on the sample size and the number of states. If the number of possible covariates is

large, some of them could be included as inverse probability weights. In the current

study, BMI and smoking were included in the analyses as normalized inverse probability

weights. The length of the follow-up period, in particular, has to be considered when

the multi-state model includes seldomly occurring events, e.g., transitions between work

and disability pension. Plotting the transition probabilities makes it possible to inspect

such “weak” events, and to assess the reliability of the curve by looking for relative

smoothness. In addition, plotting the transition probabilities will make it possible to check

the proportionality assumption concerning the Cox regressions. Like other working life

estimations based on the multi-state model, the ELMA method is a predictive method that

is based on the theoretical assumption that by cumulating the behavior of many individuals

over time, an average, proﬁle-speciﬁc behavioral pattern can be created. Such assumptions

are only valid for making predictions if the underlying conditions, such as the time period

and sample composition, are comparable.

4.2. Comparison with Previous Studies

To the best of our knowledge, this is the ﬁrst study to examine expected labor market

afﬁliation over a restricted follow-up period and including an analysis of multiple covari-

ates. Therefore, the results of the current study cannot be directly compared with earlier

studies. Typical studies related to this type of analysis use age as the underlying time

axis, and an entire life-course until retirement age in terms of analyzing the working life

expectation [2,3,6–8]. In comparison, Lie et al. (2017) used a high-dimensional multi-state

model and a simple time axis in a restricted sixteen-year follow-up period from age 20

until age 55 years and showed that low IQ and mental health problems were associated

with an increased risk of receiving a disability beneﬁt [11]. Based on earlier research,

showing that sex, age, socioeconomic factors, and health behaviors are associated with

labor market outcomes, we included ‘educational level’, ‘smoking habits’, ‘body mass

index’, and ‘working time arrangement’ [28–30]. Additionally, to accommodate potential

selection bias we adjusted for the individual selection in terms of survey wave and adjusted

for private/public sector, since the registration of sickness absence was less systematic in

the private sector. The study results are in line with ﬁndings on the association between

ill-health and working life expectancy. Previously, we showed a reduction in working

time, an increase in time of sickness absence, and unemployment among individuals with

self-reported depressive symptoms [5] and poor self-rated health [3]. Perceived stress

was shown to be strongly associated with an increased risk of sickness absence for both

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

11 of 13

sexes [17] and passive labor market participation (receiving sickness absence compensation,

vocational rehabilitation beneﬁts, permanent disability beneﬁts, or unemployment beneﬁts

among 20–21 year-old men [31]). Our results suggest, not to our surprise, that frequent

perceived stress at work and in personal life are particularly pronounced among female

employees in the life phase of family formation. To better interpret the ﬁndings more

information, e.g., on the presence of children and busy family life, is needed. Moreover,

further studies should include repeated measurements of perceived stress and investigate

if changes in perceived stress affect labor market afﬁliation.

4.3. Strengths and Limitations

The major strength of the present study is the large sample size with objective and

detailed longitudinal register-based data on labor markets states. The use of the ELMA

method allowed us to control for several confounders. However, the study limitations

should also be considered when interpreting results regarding the impact of perceived stress

on labor market afﬁliation. Data on short-term sickness absence (sickness spells less than

31 days) were not available for most employees in the private sector (approx. 63%), thus,

the number of sickness absence days in the private sector was probably underestimated.

Furthermore, frequent work- and personal life-related stress was assessed with two self-

reported questions. Though the used perceived stress measure established an immediate

condition and frequency, it might be prone to bias. Despite the large sample size, the ﬁnal

sample cannot be considered representative of all Danish employees. This is mainly due to

the low number of young individuals in the WEHD sample [32]. Sampling weights were

not used in the present study, due to considerations of not increasing the complexity of

the results’ interpretation. By neglecting possible transitions between disability pension or

retirement pension and death, the time on disability pension and retirement might have

been overestimated. However, since the follow-up was restricted to four years and only a

small fraction of individuals experienced either of the absorbing states, this overestimation

may have been minimal. The results of the present study are constrained to a Danish

context. The access to different labor market states depends on the rules and regulations of

the labor market. Depending on the particular country, an individual with the same stress

history may in one country be ‘long-term sickness absent’, but in another country with less

job security be ‘unemployed’. Our study illustrates the ‘Danish case’ and if compared with

other countries the differences in rules and regulations in the other countries have to be

taken into account in the interpretation. The results may encourage employers to focus

on work-related stress and work- and personal-related stress to decrease e.g., sickness

absence. However, to ﬁnd practical implications relating to stress prevention they should

look elsewhere, e.g., intervention studies on stress [33].

5. Conclusions

As more detailed data on labor market afﬁliation becomes accessible, more reﬁned

methods are needed to address complex labor market afﬁliation patterns. This study

introduced the new ELMA method to analyze complex labor market afﬁliation patterns,

while including covariates. The application of the ELMA method for estimating the impact

of perceived stress on labor market afﬁliation during a four-year follow-up revealed a

loss in working time among Danish employees aged 18 to 59 years with frequent work-

and personal life-related stress. In particular, women at the beginning of their working

life or mid-career may experience a considerable loss of working time and an increased

time of sickness absence if they experience frequent work- and personal life-related stress.

The ELMA method contains new ways of expanding the ﬁelds of labor market, public

health, and occupational health research; i.e., it can handle the complex and time varying

real world information on labor market states and transitions. For instance, an individual

may present with several episodes of sickness absence, before becoming unemployed and

perhaps ﬁnally leaving the labor market with a disability pension. The ELMA method

addresses all these labor market states and transitions in the same model, by calculating

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

12 of 13

the working time loss and the corresponding time in other labor market states. The ELMA

method may inform policy and practice with more detailed information about transition

probabilities and labor market attachment and may help to retain individuals in work.

Supplementary Materials:

The following are available online at

https://www.mdpi.com/article/10

.3390/ijerph18094980/s1,

Table S1A: Expected average number of days (95% conﬁdence interval)

spent in the four recurrent labor market afﬁliation states (ELMA estimates) for women divided by

age group., Table S1B: Expected average number of days (95% conﬁdence interval) spent in the four

recurrent labor market afﬁliation states (ELMA estimates) for men divided by age group.

Author Contributions:

J.P. wrote the original manuscript draft, designed the study and conducted

the analysis. S.V.T. and M.F.A. helped write the manuscript, contributed to the interpretation of the

results. S.S. and U.B. oversaw the study design, interpretation of the results, and helped write the

ﬁnal manuscript. All authors have read and agreed to the published version of the manuscript.

Funding:

The study was supported by the Nordic Council of Ministers (grant number 101250) (JP, SS).

The funder of the study had no role in study design, data collection, data analysis, data interpretation,

or writing of the report. The corresponding author had full access to all the data and had ﬁnal

responsibility to submit for publication.

Institutional Review Board Statement:

According to Danish law, research studies that use solely

Nationale Videnskabetiske Komité).

Informed Consent Statement:

Not required.

Data Availability Statement:

Data is available on the Researcher access at Statistics Denmark:

www.dst.dk/en/TilSalg/Forskningsservice

(accessed on 6 May 2021).

Conﬂicts of Interest:

The authors declare no conﬂict of interest. The funders had no role in the design

of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or

in the decision to publish the results.

References

Ilmarinen, J. The Ageing Workforce—Challenges for Occupational Health.

Occup. Med.

2006,

56,

362–364. [CrossRef]

Nurminen, M.; Nurminen, T. Multistate Worklife Expectancies.

Scand. J. Work. Environ. Health

2005,

31,

169–178. [CrossRef]

Pedersen, J.; Bjorner, J.B. Worklife Expectancy in a Cohort of Danish Employees Aged 55-65 Years-Comparing a Multi-State Cox

Proportional Hazard Approach with Conventional Multi-State Life Tables.

BMC Public Health

2017,

17,

879. [CrossRef] [PubMed]

Jagger, C.; Cox, B.; Le Roy, S.

Health Expectancy Calculation by the Sullivan Method,

3rd ed.; EHEMU Technical Report September;

Available online:

https://webgate.ec.europa.eu/chafea_pdb/assets/ﬁles/pdb/2006109/2006109_d5sullivan_guide_ﬁnal_jun200

7.pdf

(accessed on 25 March 2021).

Pedersen, J.; Thorsen, S.V.; Andersen, M.F.; Hanvold, T.N.; Schlünssen, V.; Bültmann, U. Impact of Depressive Symptoms on

Worklife Expectancy: A Longitudinal Study on Danish Employees.

Occup. Environ. Med.

2019,

76,

838–844. [CrossRef]

Robroek, S.J.; Nieboer, D.; Järvholm, B.; Burdorf, A. Educational Differences in Duration of Working Life and Loss of Paid

Employment: Working Life Expectancy in The Netherlands.

Scand. J. Work. Environ. Health

2019,

46,

77–84. [CrossRef] [PubMed]

Pedersen, J.; Schultz, B.B.; Madsen, I.E.H.; Solovieva, S.; Andersen, L.L. High Physical Work Demands and Working Life

Expectancy in Denmark.

Occup. Environ. Med.

2020,

77,

576–582. [CrossRef]

Schram, J.L.; Solovieva, S.; Leinonen, T.; Viikari-Juntura, E.; Burdorf, A.; Robroek, S.J. The Inﬂuence of Occupational Class

and Physical Workload on Working Life Expectancy among Older Employees.

Scand. J. Work. Environ. Health

2021,

47,

5–14.

[CrossRef]

Scheike, T.H.; Zhang, M.-J. Direct Modelling of Regression Effects for Transition Probabilities in Multistate Models.

Scand. J. Stat.

2007,

34,

17–32. [CrossRef]

Mansourvar, Z.; Martinussen, T.; Scheike, T.H. An Additive–Multiplicative Restricted Mean Residual Life Model.

Scand. J. Stat.

2016,

43,

487–504. [CrossRef]

Lie, S.A.; Tveito, T.H.; Reme, S.E.; Eriksen, H.R. IQ and Mental Health are Vital Predictors of Work Drop Out and Early Mortality.

Multi-State Analyses of Norwegian Male Conscripts.

PLoS ONE

2017,

12,

e0180737. [CrossRef] [PubMed]

Cohen, S.; Kamarck, T.; Mermelstein, R. A Global Measure of Perceived Stress.

J. Health Soc. Behav.

1983,

24,

385–396. [CrossRef]

Nieuwenhuijsen, K.; Bruinvels, D.; Frings-Dresen, M. Psychosocial Work Environment and Stress-Related Disorders, a Systematic

Review.

Occup. Med.

2010,

60,

277–286. [CrossRef] [PubMed]

10.

11.

12.

13.

BEU, Alm.del - 2020-21 - Bilag 328: Orientering om resultater af NFA-studie om arbejdsmarkedstilknytning for lønmodtagere, der rapporterer et højt stressniveau, fra beskæftigelsesministeren

Int. J. Environ. Res. Public Health

2021,

18,

4980

13 of 13

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

Holmgren, K.; Fjällström-Lundgren, M.; Hensing, G. Early Identiﬁcation of Work-Related Stress Predicted Sickness Absence

in Employed Women with Musculoskeletal or Mental Disorders: A Prospective, Longitudinal Study in a Primary Health Care

Setting.

Disabil. Rehabil.

2012,

35,

418–426. [CrossRef] [PubMed]

Theorell, T.; Hammarström, A.; Gustafsson, P.E.; Hanson, L.M.; Janlert, U.; Westerlund, H. Job Strain and Depressive Symptoms

in Men and Women: A Prospective Study of the Working Population in Sweden.

J. Epidemiol. Community Health

2013,

68,

78–82.

[CrossRef]

Mather, L.; Bergström, G.; Blom, V.; Svedberg, P. High Job Demands, Job Strain, and Iso-Strain Are Risk Factors for Sick Leave

due to Mental Disorders.

J. Occup. Environ. Med.

2015,

57,

858–865. [CrossRef]

Thorsen, S.V.; Pedersen, J.; Flyvholm, M.-A.; Kristiansen, J.; Rugulies, R.; Bültmann, U. Perceived Stress and Sickness Absence: A

Prospective Study of 17,795 Employees in Denmark.

Int. Arch. Occup. Environ. Health

2019,

92,

821–828. [CrossRef]

The Danish Survey the National Health Proﬁle from 2010. 2013; ISBN 978-87-7104-956. Available online:

https://www.sst.dk/

da/udgivelser/2018/~{}/media/73EADC242CDB46BD8ABF9DE895A6132C.ashx

(accessed on 25 March 2021).

Christensen, K.B.; Andersen, P.K.; Smith-Hansen, L.; Nielsen, M.L.; Kristensen, T.S. Analyzing Sickness Absence with Statistical

Models for Survival Data.

Scand. J. Work. Environ. Health

2007,

33,

233–239. [CrossRef]

Gran, J.M.; Lie, S.A.; Øyeﬂaten, I.; Borgan, Ø.; Aalen, O.O. Causal Inference in Multi-State Models-Sickness Absence and Work

for 1145 Participants after Work Rehabilitation.

BMC Public Health

2015,

15,

1082. [CrossRef]

Cpr.dk. Available online:

https://cpr.dk/english/moving-to-denmark/

(accessed on 18 March 2021).

Hernán, M.Á.; Brumback, B.; Robins, J.M. Marginal Structural Models to Estimate the Causal Effect of Zidovudine on the Survival

of HIV-Positive Men.

Epidemiology

2000,

11,

561–570. [CrossRef] [PubMed]

Heinze, G. The Application of Firth’s Procedure to Cox and Logistic Regression. Available online:

https://www.academia.edu/

2658217/The_application_of_Firth_s_procedure_to_Cox_and_logistic_regression

(accessed on 25 March 2021).

Lie, S.A.; Eriksen, H.R.; Ursin, H.; Hagen, E.M. A Multi-State Model for Sick-Leave Data Applied to a Randomized Control Trial

Study of Low Back Pain.

Scand. J. Public Health

2008,

36,

279–283. [CrossRef] [PubMed]

Andersen, P.K.; Keiding, N. Multi-State Models for Event History Analysis.

Stat. Methods Med. Res.

2002,

11,

91–115. [CrossRef]

Beyersmann, J.; Putter, H. A Brief Note on Computing Average State Occupation Times.

Demogr. Res.

2014,

30,

1681–1696.

[CrossRef]

Pedersen, J.; Bjorner, J.B.; Burr, H.; Christensen, K.B. Transitions between Sickness Absence, Work, Unemployment, and Disability

in Denmark 2004.

Scand. J. Work. Environ. Health

2012,

38,

516–526. [CrossRef]

Allebeck, P.; Mastekaasa, A. Chapter 5. Risk Factors for Sick Leave-General Studies.

Scand. J. Public Health

2004,

32,

49–108.

[CrossRef]

Barmby, T.A.; Ercolani, M.G.; Treble, J.G. Sickness Absence: An International Comparison.

Econ. J.

2002,

112,

F315–F331.

[CrossRef]

Sørensen, J.K.; Framke, E.; Clausen, T.; Garde, A.H.; Johnsen, N.F.; Kristiansen, J.; Madsen, I.E.; Nordentoft, M.; Rugulies, R.

Leadership Quality and Risk of Long-Term Sickness Absence Among 53,157 Employees of the Danish Workforce.

J. Occup.

Environ. Med.

2020,

62,

557–565. [CrossRef] [PubMed]

Trolle, N.; Lund, T.; Winding, T.N.; Labriola, M. Perceived Stress among 20-21 Year-Olds and Their Future Labour Market

Participation—an Eight-Year Follow-Up Study.

BMC Public Health

2017,

17,

287. [CrossRef]

Arbejdsmiljodata.nfa.dk. Available online:

https://arbejdsmiljodata.nfa.dk/metode.html

(accessed on 18 March 2021).

Egan, M.; Bambra, C.; Thomas, S.; Petticrew, M.; Whitehead, M.; Thomson, H. The Psychosocial and Health Effects of Workplace

Reorganisation 1. A Systematic Review of Organisational-Level Interventions that Aim to Increase Employee Control.

J. Epidemiol.

Community Health

2007,

61,

945–954. [CrossRef]