ABSTRACT
Epidemiological studies require accurate measurements of both health outcomes, potential confounders and estimates of exposures that might drive associations with health (Finazzi, Scott, & Fassò, 2013). These may be measured with varying degrees of error. Considering measurements of air pollution for example, there is a true underlying pollution surface which will form the basis of the exposures experienced by the population at risk. However this surface is not directly observable and instead measurements are taken at locations over space and time. Differences between these exposure measurements and the unknown underlying field are often referred to as measurement error. Here the term measurement error is taken to refer to any difference from the underlying true values and what is measured. Traditionally measurement error has been based around the idea of repeated measurements of a value, for example, measuring blood pressure where repeated measurements will contain a component of error, often assumed to be random. In modelling exposures and in spatial epidemiology, error will possibly comprise a number of factors including monitor calibration error and random variation but also variations in the underlying pollution field over time and space which aren’t acknowledged in the analyses. This may arise for example when modelling assumptions are too simplistic for the complex surface of the pollution field.
