Critical appraisal of the statistical aspects of an article can be taxing for anyone without expert knowledge. It is tempting to seek out statistical “rules” that can be used to identify flaws in a study, but few situations are sufficiently cut and dried to allow such a crude approach. It is worth bearing in mind that statistics is a specialist field. The idea of a clinician dabbling in statistics should alarm us as much as the idea of a statistician dabbling in clinical medicine.

Statistics should help readers, not baffle them. The findings of nearly all quantitative medical research are subject to a degree of uncertainty arising from random error, as described in the first article in this series. Appropriate use of statistics should help the reader to understand how the findings of a study may be influenced by this uncertainty. The most useful way of appraising the statistical aspects of a study is therefore to try to work out what the statistics actually say. Example 1 shows how some statistics can actually mean very little.

Of course, the least helpful statistical analysis is none at all. In this situation the authors have ignored the potential influence of chance. It is always worth asking of even the simplest article, where are the statistics? An example of a study where some statistics would be very helpful is shown in Example 2.

There are broadly two ways in which statistics can be used to address uncertainty, depending on the way the research question is asked:

1. Hypothesis testing (the p value)

2. Estimation (the confidence interval)

There is no reason why both approaches cannot be used in the same analysis. Indeed, they often complement each other. However, a number of journals prefer confidence intervals to be reported than p values.

HYPOTHESIS TESTING (THE P VALUE)

The research question is …