David Moles Statistical Advisor to the BDJ responds: Aviva Petrie's letter lists a series of criticisms of the paper by Delilbasi et al 1 . In addition to the issues that are specific to the paper, Ms Petrie's letter also serves to stimulate consideration of the more general aspects of quality control in scientific publication. Ms Petrie criticises the Delilbasi paper in three areas: study design, analytical methods, and presentation of results.

Study design

The authors did not report the justification for their chosen sample size. If comparable studies have been published then it should have been possible to determine in advance a sample size that would provide an adequate chance of detecting clinically relevant differences between groups if they do in fact exist.

This is not to say that the authors did not do this, merely that they did not report it. If the authors were unable to undertake a sample size calculation because the requisite data were not available, then this study should perhaps have been more appropriately described as 'a pilot'. The authors administered the taste stimuli in increasing concentrations in order to elicit the threshold for their detection. Ms Petrie suggests that it would have been better if the concentrations had been given in a random order.

As the authors state in their reply, although this is theoretically true, it is not always appropriate to do so. If, for example, the administration of a particularly intense taste stimulus prevented or reduced a subject's ability to detect a subsequent lower intensity stimulus then it would not be prudent to administer the stimuli in a random order due to the 'carry-over effect'.

Determination of whether the stimuli should be given at random or incrementally requires an appreciation of the physiology of taste and is not solely a statistical consideration. As such the BDJ must rely on the guidance of its expert referees in this respect. Clearly the failure to consider the potential confounding effect of smoking on taste perception is a weakness in the study.

Analytical methods

Since the women under investigation had age-matched male controls, Ms Petrie is indeed correct in suggesting that the data would have been more appropriately analysed using a method that treats the observations as being paired rather than independent.

Presentation of results

There are several criticisms of the way the results were presented and the use of language. I agree that it would have been an improvement if the results were presented in the manner suggested by Ms Petrie. I could also offer some additional suggestions.

Indeed there is almost always room for improvement in presentation and clarity. However, there is a dilemma here in that journals must decide how prescriptive they wish to be in defining the format of acceptable presentation. Subtle rewording of the title to figure 2 would have reduced the potential confusion identified by Ms Petrie. It seems that the title needs to read with that 'pinch of salt'.

As suggested by Ms Petrie the BDJ does indeed strive hard to achieve the highest standards of publication. All manuscripts are reviewed by at least two referees who are chosen for their expertise in the particular subject area. Either the scientific or clinical editor as appropriate also reviews the manuscripts. Any of the people involved in the refereeing process may request that it be sent for statistical review.

On this occasion all of the referees were satisfied and so a statistical opinion was not sought. This raises the question as to whether a statistician should routinely review all manuscripts. This is a perpetual dilemma in scientific publishing. To do so has the potential to improve the standard of some publications, but there would be other consequences including that it would take longer for all manuscripts to complete the review process. These and related issues are being discussed and considered by the BDJ.

It is my personal perception that the quality of research in dentistry has improved steadily in recent years. Certainly more research is being undertaken in collaboration with multidisciplinary colleagues, including statisticians.

This is a positive trend that can only be beneficial; especially where there is genuine 'collaboration' rather than just 'consultation'. The BDJ will continue to promote the dissemination of high quality research and to learn the lessons of experience.

Until now the policy of the BDJ has been to attempt to avoid being overly prescriptive in its requirements of authors. In addition to the general instructions to authors, the BDJ has published comprehensive guidelines regarding good practice in both statistical conduct and the presentation of results 2 . These are based very closely on the excellent suggestions from Altman et al 3 . I should like to encourage prospective authors to give due consideration to these guidelines, and I thank Ms Petrie for giving me the opportunity to remind authors of their existence.