Abstract In epidemiology many exposures of interest are measured with error. Random, or 'classical', error in exposure measurements attenuates linear exposure-disease associations. However, its precise effects on different nonlinear associations are not well known. We use simulation studies to...