Measuring working hours in the Norwegian Labour Force Survey

We have linked survey sample data with employee data from administrative registers. The combined data sources are used to assess the quality on measured contractual working hours in the Norwegian Labour Force Survey (LFS). The data files are linked at the job-level. Information on contractual hours of work in the LFS is given by the employee himself or by a member of his household. In the register data, the employers report information on contractual hours of work to a central governmental agency. We therefore have two independent sources of contractual working hours. Measurement error is estimated from the differences between the two independent working hours variables. The quantitative assessment is based on the assumption that the register data is correct, but this assumption is discussed at some length. The results indicate that LFS survey-based mean contractual working hours generally are higher in the survey than in the register. Compared with register data reported from employers, some groups of employees seem to over-report contractual working hours. The differences vary with working time length and type of working time arrangements between industries and occupations. The measurement differences are greatest in jobs without a fixed number of hours per week, and generally in jobs with few hours per week. Proxy interviews ¹ result in even larger divergences.

• Overall, the survey based contractual working hours are systematically longer, and vary more, than the register based.

• Systematic measurement error, as we have defined it, is generally much larger than random measurement error.

• For jobs with "normal" working hours and the same number every week, the relative measurement errors are quite small.

• For jobs with shorter hours, the survey seem to overestimate working hours, while it seems to underestimate the working hours for jobs with long hours.

• Jobs without an agreement of working hours have by far the largest systematic measurement errors regardless of the number of hours reported.

• Teachers in primary schools have somewhat larger systematic measurement errors than other occupations at this level. The pattern is the same as the general on: the overestimating shorter working hours, and underestimate the longer working hours.

• Teaching professionals have about the same overall systematic measurement errors as other professional occupations with fixed working hours, but a larger underestimation of the longer working hours.

• There is great variation in systematic measurement errors between industries, also when controlling for contractual working hours length and working hours arrangements.

• Proxy interviews seem to cause more overestimation of contractual working hours, especially in jobs with varying contractual working hours and long hours. The observed differences can be attributed to measurement error in the survey, given that the suggested evaluation method is valid. We therefore discuss other factors that can offer alternative explanations of some of the divergences. One such factor is errors in the administrative registers, for instance due to errors in reporting to the register or errors in the punching of reported data. Another factor is what we might call evaluation error, for instance due to error in the linking of survey data with register data. In this pilot study we focus on discussing the evaluation method and some descriptive statistical results. This is meant to form a basis for further studies, in which more details and some fundamentalcauses of measurement errors could be investigated.

¹ _{Proxy interviews, whereby a spouse or parent are interviewed on behalf of the subject, are used in the Norwegian LFS in order to increase response efficiency. Proxy interviews constitues about 15 per cent of the responses.}