In 2015 and 2018, the British Journal of Pharmacology (BJP) published guidelines on experimental design and analysis (Curtis et al., 2015. The intention was to improve the credibility of papers published in BJP by the simplest means possible. It is all very well for a journal to elaborate a framework of best practice, with lengthy explanations for each issue considered, but if authors, reviewers and editors fail to adopt the framework because it is too complex or nuanced, then we fail as a journal. Consequently, unlike most other journals (Williams et al., 2018), BJP has opted for firm rules about a small number of issues, rather than generalized and lengthy 'best practice advice'. We focused on inconsistent reporting of P values (e.g. P < 0.05, P = exact value, P < different values), persistent and unjustified use of n = 3 (or fewer), grossly unequal group sizes and an absence of randomization and blinding (each of which typically occurs together in many papers) that are particular problems in our sector and contribute to the failed replication that is undermining the credibility of preclinical research. We received two letters that criticize some of our guidance and have written an itemized reply below.First, we make a general point. Most of the BJP guidelines are 'conventions', that is, pragmatic solutions to practical challenges. This is particularly relevant to BJP's requirements for group size selection. Setting n = 5 as the minimum allowable for comparing groups by statistical analysis (the 'n = 5 rule') is clearly a convention. We are not claiming n = 5 is sufficient and necessary for all studies. In some studies, group sizes much larger than n = 5 are necessary to reduce the risk of false findings, whereas in other studies, where the control outcome has been established repeatedly in previous published work, group sizes of fewer than n = 5 may be sufficient. In the main, BJP publishes papers on new drugs, or using new transgenic animals, or evaluating variables that have not been evaluated previously, often a combination of all three. Novelty is the key. When work is novel, it is extraordinarily rare for an author to include in their Methods section a clear statement that the data are known to be drawn from a normally distributed population (the necessary prerequisite for the type of parametric analysis typically undertaken) or that they have undertaken sample size calculations a priori that indicate that n = X would be adequate for their design. Consequently, it seems that deciding on an appropriate group size is done by after-the-fact power analysis using the data generated by a study to justify the group size used in the study (as opposed to a priori power analysis) or by 'informed judgement' (guesswork). Moreover, 'group sizes as small as possible' is normally the guiding principle. The resultant problem is that studies are often favourably treated by peer review if sufficiently novel, with no questioning of group size selection. This is not a problem that can be ignored. Most statistical software pr...