We found no significant impact overall, of implementing the IA Agreement on the risk of long-term sickness absence spells. When stratifying by sex, there was an overall positive impact of the IA Agreement among female workers, whilst no effect was found among male workers. Companies signing the Agreement were more likely to be large (≥50 employees) and were more often within the manufacturing and health and social work sectors. In large manufacturing companies, there was a statistically significant reduction in the risk of LSAS among both male and female workers after the implementation of the IA Agreement. As this was found in both intervention and control companies, it indicates no impact of the IA intervention. In large health and social companies there was, in contrast, an increase in the risk of LSAS after the introduction of the IA Agreement. The increase was lower in the intervention group compared to the control group, resulting in a positive impact of the actual IA intervention. This pattern was mainly evident among female workers in large health and social companies. In sum, the results indicate that the impact of the IA Agreement on risk of LSAS varied considerably depending on sex, industry, and company size.
One of the strengths of this study is the use of statistical analyses that take into account the difference in LSAS pre and post intervention. Difference-in-difference analysis is a causal inference method that can be applied to counter selection bias and confounding. The large study population also made it possible to stratify by sex, industry and company size. This stratification, in combination with the DID analyses, could therefore reduce the bias and confounding that may result from the significant differences in the distribution of company size, industry and sex between the intervention and control groups. This difference in the distribution of the IA Agreement by industry and company size was, however, a challenge in some strata where groups were small, leading to less robust estimates. In addition, self-selection bias may be an issue, as we observed that the employees in companies signing the IA Agreement had a higher risk of LSAS prior to the intervention, compared with employees in control companies. This might challenge the key assumption of exogeneity in DID, which posits that the selection into the intervention (the IA Agreement) should not be predicted by the outcome (LSAS) prior to the intervention. However, there is no evidence that individuals choose where they work based on the company’s IA status, and the companies do not only base their choice of signing the IA Agreement on prior sickness absence level (20). It is also worth mentioning that the NAV Working Life Centres had a recruitment campaign which mainly targeted companies with high SA, and we cannot rule out that this may have influenced risk of bias.
Difference-in-difference analysis is often used to counter selection bias between the intervention and control group, including for repeated cross-sectional data where the two samples are not the same in the pre and post period (21). The DID method can therefore account for changes within the groups over time, as long as the change is the same in both the intervention and control groups. Age is an example of this in our study, as the groups both age by 4 years (see Table 1). This change should therefore not influence our results, as the effect of age is assumed to be equal for both groups. This is supported by the supplementary analyses we conducted where the risk of LSAS was higher but the direction of impact of the IA Agreement remained the same, but slightly reduced (see supplementary Figure 2).
One of the many challenges in evaluating the IA Agreement’s impact on LSAS was that we did not have data on exactly when the IA Agreement was signed by a specific company (only yearly data). We did not have data on when they started introducing the different preventive measures, either. Other studies have shown, however, that most companies signing the Agreement increased their effort to lower SA following the start of the IA Agreement in 2001, regardless of the date they formally signed the Agreement (5). Even so, the specific preventive actions or activities they may offer is not available in our data, preventing us from evaluating the possible differential impact of activities on SA. To address this, we have used company size as a proxy measure as this can indicate differences in the means or resources they have available to use in implementing the IA activities, applying for grants and so on. Larger companies may have more resources available to make use of and benefit from the possibilities in the IA Agreement more than small companies.
Sickness absence is multi-factorial and influenced not only by the individual’s health status and work environment, but also by the regulations in the welfare system. A challenge often encountered when evaluating population-level interventions is other “interventions” outside the scope of the evaluation that impact the outcome. During the intervention period in our study, a sick leave reform was introduced. The reform was implemented in 2004, and a key element was the activity requirement. It required an individual to engage in work-related activity as early as possible (at the latest within eight weeks) in order to be entitled to sick pay. The only exception was when medical reasons clearly prevent it. SA figures from Statistics Norway show a marked decline in sick leave in 2004 and 2005 that coincides with the sick leave reform. It is most likely that the 2004 reform has an effect on LSAS in our study, but this effect is assumed to be equal in both groups as the change in regulations was implemented nationally and applied to all companies, regardless of the companies’ IA status.
Another limitation of this study is the focus on LSAS alone as the outcome, as the IA Agreement incorporates three goals that might influence each other. Reducing SA is only one of the goals. The two other goals are to secure recruitment of people with disabilities and vulnerable groups into the labour market and to prolong working life. These three goals may affect each other, as a company that increases the recruitment of people with disabilities and prolongs working life for older workers, might also experience increased SA due to this. It is therefore possible that we may present an underestimation of the impact of the intervention on LSAS in our study for companies with a high goal attainment on inclusion of disabled workers, but this is unknown. It is therefore important to have in mind that our results are only evaluating one of the goals of the IA Agreement.
Our results in light of other findings
Our finding that the impact of the IA Agreement on LSAS varied considerably depending on sex and industry contributes to limited literature on the impacts of this population based intervention. Evaluations of the IA Agreement have been published in some reports without peer review (5, 6, 22, 23), and have indicated a positive impact of the IA Agreement in manufacturing, as this sector shows a decrease in LSAS after implementation of the intervention. Earlier reports have also indicated a negative impact of the intervention in the health and social work sector, as LSAS increases in this sector in the same period (5). Our results were therefore a bit surprising as we found the opposite; there was no impact of the IA Agreement in manufacturing, and a positive impact in health and social work. These contradicting results can be explained by the fact that we use an analytical method that takes into account the intervention and control group differences, both before the IA Agreement was implemented and 4 years after. This explanation is strengthened by the fact that we get similar findings (decrease in LSAS for employees in manufacturing companies and increase in LSAS for employees in health and social work companies) when we use the same statistical methods as in the reports, without the use of DID method.
It is also important to bear in mind that how sickness absence is estimated, can lead to contrasting results when interpreting the impact of the IA Agreement. We evaluated the impact on the change in one-year risk of long-term sickness absence spells. We cannot rule out the possibility that a different result may have been obtained if a different measure had been used such as the mean duration of sickness absence spells, the annual number of spells or the use of graded sickness absence spells. However, other scientific papers on the IA Agreement use the same measure of SA as in this study, enabling easier comparisons of results. A study from 2011, on the impact of the IA Agreement on sickness absence (7) shows similar results as in this study, namely no impact in the overall sample. However, in contrast, we find that the impact varies by industry. This may partly be explained by the considerably smaller sample in the 2011 study, which impedes stratification by industry. In their analysis, they also used office workers as a reference, and did not include information on company size or take into account the baseline difference in sickness absence. Another study by Midtsundstad et al from 2014 showed, on the other hand, a positive impact of the intervention on overall sickness absence and a varying effect by industry (8). They used the same DID method as in our study and found a positive impact on sickness absence among IA companies in the public administration sector. This is partially in line with our findings; however, in our study, the positive impact in the public administration sector was only found in women and varied according to company size. They also did not find positive impacts of the intervention for manufacturing, construction and transport; although the overall sickness absence decreased in manufacturing, it was not due to the intervention which is also similar to our findings. This can indicate that the decrease in sickness absence may be related to other factors than the IA Agreement, for example company characteristics or the focus on sickness absence and work environment in the manufacturing sector as a whole, resulting in an impact for all companies and not only those signing the IA Agreement. We also found a positive impact among female health and social workers in medium and large companies, which was not found in the other study (8). A potential reason for this discrepancy in results may be that they focused on older workers (aged 50+) whilst our study included younger and middle-aged workers (aged 25-34). Beyond these two studies, there is little scientific knowledge on the impact of the IA Agreement on sickness absence that also takes into account the possible differential impacts by industry. Other scientifically based evaluations of the IA Agreement have been carried out, largely reporting a positive impact, however, they have evaluated other outcomes, such as disability benefits (10) and return to work after rehabilitation (9), and are therefore not comparable to our study.
The present study strengthens the evidence that the impact of the IA Agreement on the risk of long-term sickness absence spells varies greatly between industries and the size of the companies. Very few clear implications can be given based on this study, as we do not have data on the preventive measures used in different sectors. The varying impact by industry may however, imply that the IA Agreements measures (such as focusing on close follow up of those on sick leave, adjust work tasks to enable the employee to work even when sick) may suit the needs of some industries and not others. For example, in some industries the possibilities to adapt the work tasks are more limited than in others, such as manual workers in manufacturing or construction compared with office workers in public administration. These measures may therefore not be suitable for all industries.
It is, also evident that the impact varies greatly depending on the size of the company, which may imply that larger companies have more resources available to make use of and benefit from the IA Agreement’s activities compared to smaller companies. Based on this it is evident that there is a need for a greater focus on industry-related exposures and the possibilities for preventive measures addressing the specific challenges in each industry. Previous studies have found that 23-28% of long-term sickness absence is attributable to work-related exposures (24, 25). This indicates that interventions targeting the work environment can be considered an important method for decreasing sickness absence. According to the surveillance of work environment for the working population in Norway (14), health and social workers have challenges in terms of high emotional demands, role conflicts, job strain, unwanted sexual attention, violence and threats, working nights, neck bending, and awkward lifting. In contrast, manufacturing workers have more exposure to noise, vibrations, awkward lifting, squatting/kneeling, downsizing, and job insecurity. This may indicate that reducing sickness absence in these two sectors would warrant different preventive strategies and actions.