X hits on this document





111 / 156

  • Studies with a large refusal rate (low participation rate) are unlikely

to receive a “yes”.

Studies with comparison groups

    • Consider whether blinding was used in the allocation of workplaces to intervention/control groups

    • Differential workplace refusal rates and differential workplace dropout rates in designs with comparison/control groups are indicators of bias arising from sample selection and maintenance.

    • Whenever groups have been created through non-random means, there is a threat to internal validity. If groups were selected through a matching procedure, the threat would be less.

  • 9.

    Are you confident that the potential confounders were adequately considered, and then either well controlled or appropriately discounted as a source of bias in the I/E portion of the study?

  • In the comment box following your response, note the rationale for your selection.

  • This question addresses confounding bias: distortion of the estimated effect of an exposure on an outcome, caused by the presence of an extraneous factor associated both with the exposure (i.e., intervention) and the outcome, but is not a mediator between exposure and outcome.

  • Before-after and time-series designs area unlikely to receive a “yes” rating, because they are always susceptible to “history bias” (another term from the quasi-experimentation literature), which occurs when something else takes place in the workplace or extra-workplace environment that could bring about or alter the effect observed. A “partially” rating can be achieved, if efforts are taken to assure the reader that the observed effect can not be explained by another event taking place in the organization simultaneously.

  • Demonstration of similar distribution of known confounders among comparison groups is required for a “yes” when groups are created through non-random means. If created through random means, this demonstration is important when there are small numbers in the comparison groups.

  • We are concerned in this question about all potential sources of confounding, whether they arise through the sample, coincident events, differential measurement, etc.

  • Control of confounders can be either through design or statistical analysis.

  • Important confounders at the workplace level: management commitment to OHS, size of enterprise, other organizational change initiatives that could affect implementation/effectiveness, changes in the process technology

Institute for Work & Health


Document info
Document views415
Page views419
Page last viewedFri Dec 02 22:44:46 UTC 2016