Study population

The study was conducted as a cross-sectional survey. Figure 1 shows the flow chart for the WEHD 2012 populations. The source population for WEHD 2012 was all individuals living in Denmark, aged 18–64 years, who were employed between 30 November 2011 and 31 January 2012, worked ≥35 hours/month and earned at least 3000 Danish Krones (Kr) per month (~US$460).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flow chart for Work, Environment and Health 2012. 

Two samples fulfilling these criteria were drawn by Statistics Denmark. The first sample of employees (the stratified workplace sample) was drawn from 1060 selected workplaces (with at least 10 full-time equivalent employees). These companies were sampled stratified according to industry (five categories) and workplace size (small workplaces with 10–99 full-time equivalent employees; medium-sized workplaces with 100–499 full-time equivalent employees; large workplaces with at least 500 full-time equivalent employees). This target population consisted of 198 789 individuals, of whom 29 027 had a secret address or a so-called ‘research protection’ in the Civil registration number (CPR) registry (a protection of own information, that any individual in Denmark could request before 2014 and that implied that researchers were prevented from getting the contact information needed to invite potential participants), leaving 169 762 individuals for the final population. A total of 10 employees were selected randomly from each of the small workplaces (10–99 full-time equivalent employees), 20 employees were selected randomly from the medium-sized workplaces (100–499 full-time equivalent employees) and 30 employees were selected randomly from the large workplaces (at least 500 full-time equivalent employees) resulting in 15 767 eligible individuals. Of those, 10 died and 82 emigrated between 1 April and 30 September 2012, resulting in a final sample of 15 675 employees.

The second sample (the random sample) was drawn from the population of 2 068 696 employees living in Denmark and not employed at the 1060 workplaces in the first sample. Of these, 318 430 had a secret or ‘research protected’ address, leaving 1 750 266 individuals for the final target population. A random sample of 35 039 employees was drawn from this population of which 24 died and 210 emigrated between 1 April and 30 September 2012, resulting in a final sample of 34 805 employees. To ensure generalisability, the primary results of the article are based on the random sample of employees, and results from the analyses on the stratified workplace sample are included as online supplementary material.

Data sources

In April 2012 potential participants were invited by a postal letter, including a link to a web-based questionnaire. Non-responders of the first contact received a second postal letter with a link to the questionnaire and an enclosed paper questionnaire. After these two contacts, non-responders were contacted up to two times by phone by a commercial research institute. They were asked whether they were still working and offered a new paper questionnaire.

Participation was defined as answering at least one question in the questionnaire, regardless if the questionnaire was the online version, the paper version or answered by phone.

Questionnaire data

The questionnaire comprised 51 questions on physical and psychosocial working environment, health-related behaviours, working ability and health. The questions related to current employment (eg, length of employment, transportation, working hours, job security), psychosocial working environment, work accidents and safety, noise, vibrations, skin exposures, physical demands including lifting, self-rated work ability, seniority and pension, sleep, vitality, pain, chronic disease including mental disorders, smoking, alcohol use, height and weight.

Registry data

Information about country of origin, cohabitation, education, annual personal income and type of job was extracted from Statistics Denmark using the unique, Danish civil identification number18 given to every Danish citizen. The registers contain historic information that is collected continuously for the entire Danish population from the civil registration system, taxation authorities and all Danish worksites for whom it is compulsory to report certain information about their employees.19 Age was defined as age on 15 March 2012. Information on job type was collected from the Danish E-income Registry and was based on the first digit of the DISCO-08 classification code, the Danish, and slightly modified version of International Standard Classification of Occupations 2008 (ISCO08).20

Definition of variables

The outcome (participation) was defined as having answered at least one question in the questionnaire.

Job type was defined as one-digit ISCO08 number; sex had two levels (men and women); age was categorised in five categories (18–24, 25–34, 35–44, 45–54 and 55–64 years); annual personal income was categorised in three categories (<250 000 Kr; 250 000–430 000 Kr; >430 000 Kr); education in six categories (compulsory school system, upper secondary education and university preparatory schools and vocational education, academic level, missing data on education (Danish origin), missing data (Western origin), missing data (non-Western origin)); cohabitation was categorised in three categories (married or unmarried couple, living alone, more families living together); country of origin was categorised in three categories (Denmark; Western country, not Denmark; Non-Western country); and regarding geographical region of residence, the five Danish regions were used (Copenhagen, Zealand, Northern Jutland, Middle Jutland, Southern Jutland).

Statistical analyses

In the initial descriptive analyses, frequencies of characteristics of participants and non-participants were presented for both the random sample and the stratified workplace sample. The following results are based only on the random sample (to ensure generalisability as mentioned in Methods).

Logistic regression models were used to estimate the OR with 95% CI for participation. OR for each independent factor was estimated in four models: (1) model 1—unadjusted, (2) model 2—adjusted for age by sex, (3) model 3—model 2 further adjusted for personal income, education, cohabitation, country of origin and geographical region of residence and (4) model 4—model 3 further adjusted for job type. The full combination of age by sex (age in 1-year categories) was included in the analyses. The association of participation with income was modelled as a linear spline on the log-odds scale with knots in the quartiles: 262 000; 337 000; 427 000 Kr. The other factors of interest were included in the logistic regression models as categorical variables categorised as in the descriptive analyses. The associations with age by sex and income were displayed graphically and the associations with the other variables were presented in tables.

Sensitivity analyses

We excluded the 1400 individuals with no information on education and evaluated whether this changed the results.

Software

The statistical analyses were performed using the GLIMMIX procedure in SAS V.9.4 (SAS Institute) and the plots were performed using R V.3.1.2 (The R Foundation).

Patient and public involvement

There is no patient and public involvement in this study.

Characteristics of the study population

Table 1 shows the characteristics of participants and non-participants in the random sample of 34 805 eligible employees as well as in the stratified workplace sample of 15 675 eligible employees. A total of 17 591 employees (50.5%) participated in the random sample of the survey. The two most prevalent job types in the random sample were professionals having the highest participation of 60% and elementary occupations having the lowest participation of 41.6%. The usual pattern for participation was seen for the other socio-demographic variables: women were more likely to participate than men, participation increased with age, personal income, higher education, cohabitation and Danish origin. Similar results were found for the workplace sample.

Results from regression analyses

Results from the logistic regression analysis of participation status in relation to job type and other socio-demographic factors can be seen in table 2 and figures 2 and 3. Table 2 shows that compared with professionals, the low participation among armed forces occupations (OR 0.46), craft and related trades workers (OR 0.52) and skilled agricultural, forestry and fishery workers (OR 0.59) was clearly increased by adjustment for sex and age (model 2: OR 0.85, 0.72 and 0.72, respectively). Further adjustment for the remaining socio-demographic variables generally attenuated the association between participation and job type, but the association with job type remained statistically significant (model 4: p<0.0001). The adjustment for job type slightly attenuated the associations with education and cohabitation (model 4 compared with model 3), whereas adjustment for the other socio-demographic factors was more important regarding the association with country of origin (model 3 compared with model 2). However, all socio-demographic factors remained statistically significant in all investigated model. Similar results were found for the workplace sample, see online supplementary appendix table 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

OR (95% CI) for participation by sex and age in 1-year categories for employees in the random sample of Work, Environment and Health 2012. Model 1: Crude OR by age and sex. Model 3: OR adjusted for country of origin, cohabitation, education, region, age*sex and annual income (as a spline function). Model 4: model 3+job type. Reference value: men aged 45 years.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

OR (95% CI) for participation by annual, personal income (as spline function) among 34 805 employees in the random sample of Work, Environment and Health 2012. Reference value: the median income, 353 000 Danish Krones (~US$50 000). Model 1: crude OR by income. Model 2: model 1 adjusted for age*sex. Model 3: model 2 further adjusted for country of origin, cohabitation, education, region and age*sex. Model 4: model 3 further adjusted for job type.

Figure 2 shows OR for participation by sex and age in 1-year categories. The crude ORs (model 1) were higher for women compared with those for men for all age groups, except for employees aged 64 years, where men and women had similar ORs. There was a steadily increasing participation with age among both men and women, except that men aged 20 years or younger had a higher OR for participation compared with men aged 20–26 years. When adjusted for all socio-demographic factors except job type (model 3), the lower participation among 18–25-year-old individuals compared with same-sex older individuals disappeared, but the difference between the two sexes remained. Additional adjustment for job type (model 4) made almost no difference. For individuals above 26 years of age, the ORs in the crude and adjusted analyses were similar. Similar results were found for the workplace sample, online supplementary appendix figure 1.

ORs for participation by personal income with the median income as reference can be seen in figure 3. In the crude analysis a gradual increase from OR 0.5 for employees with the lowest income to OR slightly above 1 for employees with the highest income was evident. Adjustment for age by sex slightly attenuated the association below the median income (model 3). Further adjustment for education, cohabitation, country of origin and region clearly attenuated the associations across the whole range (model 3), whereas additional adjustment for job type made no difference (model 4). Similar results were found for the workplace sample, online supplementary appendix figure 2.

Results from sensitivity analyses

Excluding the 1400 individuals with no information on education did not change the results substantially.

Key results

The results of this study showed that participation in WEHD 2012 varied between job types. This variation was partly explained by differences in sex and age for armed forces occupations, craft and related trade workers, and skilled agricultural, forestry and fishery workers, and by personal income, education, cohabitation, country of origin, and geographical region for all job types. All socio-demographic factors remained statistically significant when mutually adjusted.

Strengths and limitations

The major strengths of this study are the large sample size, the national coverage and the access to almost complete registry data available for both participants and non-participants. Potential limitations are the relatively crude job groups, and missing information on education for 1400 individuals. The sensitivity analysis addressed the potential limitation on missing data. Unfortunately, we did not have sufficient statistical power to conduct analyses with more detailed job groups.

Furthermore, some job groups were small. For example, the number of employees with no DISCO code is 84; consequently, the result for this group could potentially be affected by sparse data bias and should therefore be interpreted with caution.21

In a cross-sectional study like this, all the variables are measured at the same point in time; therefore, cross-sectional studies are not well suited for drawing any conclusions about causality. However, in this study our scope was to characterise participants and assess associations between individual, characteristics based on registry data, and therefore, temporality is not a major concern.

Previous studies

Our results confirm previous studies with regard to the importance of socio-demographic factors like sex and age, education and socioeconomic indicators for survey participation.6 7 10–17 Furthermore we were able to show that job type can be added to this list as an independent predictor of survey participation, at least in a study on work environment and health.

Studies have shown that follow-up procedures,22 a relevant study aim (potentially affecting the individual in question), and monetary incentives can be used to increase participation rates.23 However, the effect of monetary incentives was related to education and income.24 It has also been shown that applying different data collection methods on different population subgroups can increase participation rates and produce more representative samples. For example face-to-face recruitment and data collection,25 telephone interviews or printed letters can be used as an alternative to mail/web-based surveys or vice versa.26–31 In such case, the researchers need to weigh the benefit of higher participation against the possibility that participation as well as actual answers to a given question may differ according to the method used. In contrast, long and exhaustive questionnaires, difficult accessibility to the questionnaire, or both, are typically associated with lower participation.32

Therefore, questionnaires should first be designed to be easily accessible and then different incentives and data collection methods (including stratified sampling) may be considered. However, it is important to note that low participation not necessarily means a high level of bias and that, vice versa, a high participation not necessarily means a low level of bias. Some efforts to increase participation may in fact increase bias if the consequence of this effort is selective participation, or unwillingness to participate in future studies.

Concluding remarks

In this population of Danish employees, we found a relatively small but independent effect of job type on participation. We also found socio-demographically imbalanced participation as reported in previous survey studies. The low participation in the age groups below 25 years of age could be explained by the socio-demographic factors, personal income, education, cohabitation, country of origin and geographical region of residence. Further adjustment for job type did not change the estimated association with sex and age.

Future studies using questionnaires may consider using different data collection methods and incentives for different (sub-)populations defined by job type and other factors, for example face-to-face interviews in populations with low education or elementary occupations, and monetary incentives in the younger age groups.

The authors believe that the results of this study on participation in the WEHD2012 can be generalised to the general Danish work force aged 18–64 years because they are represented by the random sample of this study and because registry data are used.

Ethics

Because the study did not include biological samples, no further approval was required. According to the Helsinki Declaration participants were informed that participation was voluntary and that they could withdraw themselves (and their data) from the study at any time.