unhealthy work logo, unhealthywork.org

Studies Using the Imputation Method

Studies of Job Strain Using the Imputation Method


(All reference numbers are from Schnall PL, Landsbergis PA, Baker D. Job strain and cardiovascular disease. Annual Review of Public Health 1994;15:381-411.)

Self report bias is a potential problem in many “job strain” studies, since exposure has often been assessed through questionnaires completed by study participants. Self-reports may be inaccurate descriptions of job characteristics or may be biased by personality traits such as “negative affectivity”. Concerns have also been raised about the need for more objective measures of “job strain” in intervention studies.

Therefore, in 13 “job strain” studies, researchers employed an analytic technique to overcom self-report bias and obtain more objective measures of job characteristics – the imputation of average scores for a particular job title to individuals in that job title. The average job-title score, free of the individual’s subjective assessment, then predicts outcome for the individual. However, while this strategy is often presented as desirable, it developed in the U.S.A.. because of a lack of databases containing both job characteristics data and health data – a weakness of past research. Large within variance exists in job characteristics (55% of reliable variance for latitude, and 93% for demands), since job titles such as nurse, machinist, secretary or teacher are somewhat heterogenous in skill levels, autonomy, or demands (102). As a result, in the U.S. studies, mean scores of job characteristics are adjusted for demographic covariates (e.g., age, race, education, marital status, region, urban vs. rural, and self-employment status) in the HANES 1 (59) when imputed to an individual participant (102). Despite this adjustment, the imputation strategy introduces (non-differential) misclassification and a bias towards the null. Thus, positive findings using the imputation method (4, 5, 36, 59, 68, 81, 109) provide strong support for the model, while negative studies may result, in part, from loss of power. However, individual level job data and health data clearly need to be obtained in future research.

Leave a Reply

Your email address will not be published. Required fields are marked *

© 2024 Unhealthy Work  |  For more information regarding this site, please contact us