Letters

Clarification needed about possible bias and statistical testing

Consenting adults?

From BMJ USA 2003;Feb:93

RAPID RESPONSES FROM BMJ.COM Following are edited excerpts from Rapid Responses generated by this article, which can be read in their entirety at http://bmj.com/cgi/eletters/325/7377/1391.Editor

Clarification needed about possible bias and statistical testing

EDITORI have two concerns about this study. First, I would like to know exactly what participants were told about the study (perhaps the authors could make the letter sent to participants available on bmj.com). This is important because of the subjective nature of participants' responses. If participants had any reason to suspect the nature of the study hypothesis, this could have introduced substantial bias into the results.

Second, their choice of statistical method is rather strange. The t-test is designed for use with continuous, interval-level, normally distributed data. Use of a t-test on the discrete, ordinal data reported in this paper is questionable. A more appropriate statistical method would have been a non-parametric test, a comparison of proportions of positive and negative responses, or even a proportional-odds model. I think this matters, as I have checked the results of the paper by comparing the proportion of negative responses between the groups, and the statistical significance of the results becomes much less impressive when analyzed this way. Indeed, three of their five significant results (interest, importance, and relevance) become non-significant (P values of 0.054, 0.21, 0.054, respectively).

Adam Jacobs, director. 
Dianthus Medical, Ltd. London, UK ajacobs{at}dianthus.co.uk

Consenting adults?

EDITORSo, the editor of the BMJ is happy to coauthor and publish research involving undeclared deception of subjects. Three hundred BMJ readers were unwitting dupes. They took part without being informed about the real object of the exerciseno informed consent here, at least not by the standard of showing the subject the research protocol. This sort of research violates Kant's categorical imperative: Act as if this were a universal principle. It seeks payment in a coin that it debases for others. What is ironic about the paper is that the results only have meaning if the respondents were more honest than the authors. If you deceive your subjects, what right have you to expect they will not do the same to you?

Stephen J Senn, professor of pharmaceutical and heath statistics. 
University College London, London, UK stephens{at}public-health.ucl.ac.uk