The most important paper I never published


HOW MUCH DOES PRIOR INFORMATION INFLUENCE THE DIAGNOSTIC PROCESS? OBTAINING INSIGHT INTO OBSERVER VARIATION. This paper, downloadable from gives experimental evidence showing how the influence of background information sways the diagnostic process, to a degree that differs substantially between observers. I never succeeded in publishing this one, only presented at conferences, and it would need much more up-to-date references. But I commend it to your interest!



This is an interesting paper, and I’m surprised you have had trouble publishing it!

I’m not sure that you should use the word bias in the context of background information being integrated into the decision making process. Bias is, after all, a property of error. Prior information can change the modal diagnosis, and therefore changing the error rates associated with positive and negative tests. Central chest pain, for example, will have a different significance in a young patient with achalasia and an old patient with a history of smoking. This information will influence the interpretation of the symptom, but I wouldn’t use the word ‘bias’.

You rightly point out that

over-caution may result in gross over-diagnosis with excessive demands for further diagnostic or therapeutic interventions

but to balance this, under-diagnosis can lead to death of the patient. So in addition to the prior probability of a diagnosis, the balance of risks involved in errors of interpretation need to be taken into account. It is this consideration, I think, that requires information about previous convictions to be withheld from jurors, on the grounds that the perceived harm in convicting an innocent person is always held higher than the harm in letting a guilty person go free.

Please don’t give up on it – it is far more deserving of readership than many of the things I’ve had to review!



This manuscript with Kerry Lee is the most important paper I never published: Using logistic model calibration to assess the quality of probability predictions. D.R. Cox, then Editor of Biometrika said he would publish the paper if we could cut its length by 1/2, which I could never bring myself to do. I wish I had.



If you can’t bother submitting to yet another journal, maybe it would be a good idea to publish the paper as a preprint in arXiv, OSF Preprints, Zenodo etc. It might make it more easy for people to used them as references for their own research.