Assessment of statistical procedures used in papers in the Australian Veterinary Journal.

One hundred and thirty-three papers (80 Original Articles and 53 Short Contributions) of 279 papers in 23 consecutive issues of the Australian Veterinary Journal were examined for their statistical content. Only 38 (29%) would have been acceptable to a statistical referee without revision, revision would have been indicated in 88 (66%), and the remaining 7 (5%) had major flaws. Weaknesses in design were found in 40 (30%), chiefly in respect to randomisation and to the size of the experiment. Deficiencies in analysis in 60 (45%) were in methods, application and calculation, and in the failure to use appropriate methods for multiple comparisons and repeated measures. Problems were detected in presentation in 44 (33%) of papers, with insufficient information about the data or its statistical analysis and presentation of statistics (appropriate missing or inappropriate shown) the main problems. Conclusions were considered to be inconsistent with the analysis in 35 (26%) of papers, due mainly to their interpretation of the results of significance testing. It is suggested that statistical refereeing, the publication of statistical guidelines for authors and statistical advice to Animal Experimentation Ethics Committees could all play a part in achieving improvement.