The validity of the risk index for comparing the accident risk associated with different work schedules

Although it is known that there is an association between working hours and occupational accidents, there are only few tools which allow for an assessment of the risk associated with different working time arrangements, none of them validated yet by independent samples. Promising tools like the risk index (RI), assessed via the fatigue and risk index calculator, thus need to be validated. Since analyses on the validity of the RI based on individual data found only moderate associations between the RI and the actual accident risk, the question arises, whether a validation based on grouped data yields more promising results. Thus, the validity of the RI was re-examined, using aggregated data, i.e. pooling several individuals into groups of respondents according to their RI-score and performing calculations on the group level in order to control for individual and random variation. The study sample (n=337) was derived from an Internet survey on working hours and health, which contained information on working hours over four weeks and on the occurrence of an occupational accident during the last 12 months. Two models of aggregation were used: a distribution based and a criterion based classification, resulting in four RI groups each. Accident rates for each group were calculated and compared via χ(2)-analyses and fitting trends to the data. Results show, for both models of aggregation, that the accident rate increased exponentially with an increasing RI. Risk was 3.7 times (distribution based aggregation) and 6.5 times (criterion based aggregation) higher in the highest compared to the lowest RI-groups. The exponential trend explained 92% (distribution based aggregation) and 98% (criterion based aggregation) of the variance, respectively. Analyses for homogeneous subgroups (e.g., service sector, and workers with high workload) showed similar trends. Thus the analyses based on aggregated data indicated a considerable and consistently higher validity of the RI than the earlier analyses based on individual data, irrespective of the model of aggregation. These results are quite promising although further analyses based on larger samples are needed to confirm these findings and in order to further develop the RI or similar indices.