This study examined sample attrition in cohorts of permanent and non-permanent employees. The focus was on the associations of baseline SOC with non-response to follow-up survey. In addition, we examined whether SOC predicted contact failure due to exit from the work place among permanent employees and contact failure due to unlocated address among fixed-term and subsidised employees. Corresponding associations with sociodemographic factors and self-rated health were also studied.
We found that low SOC predicted low response rate in the cohort of fixed-term employees. Contact failure was not related to SOC among permanent employees, i.e. those who left work for "natural" reasons such as a move to another workplace, retirement or death, had only slightly lower SOC than the respondents. Among non-permanent employees contact failure tended to be more common in those with low SOC, but this association did not reach statistical significance.
According to Antonovsky [7, 8] SOC predicts and explains an individual's movement along the health-disease continuum. He viewed SOC as individuals' global orientation or disposition leading to a certain attitude towards their internal and external environment . A strong SOC means good resources to resist stressful encounters, including work-related stress and other exposures. In line with this, Feldt  found that employees with strong SOC had fewer psychosomatic symptoms and lower level of emotional exhaustion at work. Antonovsky's theory also suggests that SOC stabilises by the age of about 30, and the first years of employment are important in this development . The period during which individuals leave their studies to enter the labour market appears particularly important ; if the early career is characterised by insecure entry into work, SOC may never reach the optimal level.
In our study, the permanent employees were older than the fixed-term and subsidised employees, and it is evident that majority of them should have had a stable SOC. However, as most of the non-permanent employees were older than 30, and we may assume that their SOC was also relatively stable.
Research in the field of psychiatry has documented that psychiatric disorders and psychopathology increase sample attrition due to loss of contact rather than refusal [6, 18]. According to our results SOC does not follow this tendency. In other words it seems to be inappropriate to consider SOC simply as an indicator of mental morbidity or psychological well-being.
Poor self-rated health has been shown to predict sample attrition [1, 5, 19]. Our findings confirm this in the cohort of permanent employees, both with regard to non-response and to exit. In fixed-term employees the point estimate showed an opposite though statistically non-significant association with regard to non-response. On the other hand, among them responding was associated with high SOC. In all, these findings add one more paradox to the body of research concerning health and well-being of fixed-term employees [see e.g ].
Women tend to participate in follow-up more actively than men [19, 21–23], although there are some conflicting findings [2, 24]. In this study women also were more likely to respond in all cohorts. Several studies have explained this by pointing to gender differences in health. However, in our study such an explanation was unlikely because the analyses were adjusted for self-rated health at baseline.
The probability of responding is reported in most studies to increase with age [7, 10, 12], although there are studies of elderly cohorts reporting opposite findings [3, 12]. In our study the cohorts were relatively young. Thus, our finding of less attrition among older participants was to be expected.
Higher response rates have been reported among those with a higher socio-economic status [1–3, 6, 18, 19, 21–24]. This finding was replicated in our study but only among permanent employees.
This study, concentrating on associations between SOC and sample attrition during follow-up, does not permit conclusions of a potential SOC-related participation bias in the initial survey. It is possible that the effect of SOC is different in surveys conducted among unselected populations.