The authors reply in plain English
Sir,
We would like to thank Drs Thiébaut and Kipnis for their thoughtful comments. We agree that random reporting error is always present and results in attenuated associations. It is clearly necessary to recognise all types of error and their potential impact on epidemiological associations.
As the commentary correctly points out, we were not focusing on random error in our paperReference Heitmann and Lissner1. Even if in some cases random error may overwhelm biases from systematic error, this may not always be the case. Whether true associations are overestimated or underestimated depends on the magnitudes of these two types of error, as well as on the direction of the bias in relation to the underlying association. The purpose of our paper was to illustrate, not to prove, that non-random error can in theory inflate an association.
The effects of non-random errors on diet–disease associations are not always appreciated. For instance, in a recent re-analysis of data from the OPEN (Observing Protein and Energy Nutrition) study by a team including ourselves, Dr Kipnis and other researchers from National Cancer InsitituteReference Lissner, Troiano, Midthune, Heitmann, Kipnis and Subar2, we concluded that obesity-related reporting errors require much further investigation. Although the OPEN data are indeed unique in being able to characterise both types of error, it must be kept in mind that they are based on a highly selected study sample. Clearly, we need to improve our knowledge about person-specific and other non-random errors, as well as our ability to communicate about them.
