Having a bad day, think regression to the mean

Regression to the mean, is one of those phenomena, that once aware, it tends to crop up everywhere.The term’s not new – it originates from Francis Galton’s paper on regression towards mediocrity in hereditary stature, published in 1877.

The basic principle of regression to the means is that measurements will tend to be closer to the average the next time they are measured. I think about, and explain the principle, in relation to having a bad day, applying the principle of regression to the mean, it is more likely your day will better tomorrow irrespective of what you try and do about your bad day today. But, if you are having a good day today, then enjoy it whilst it lasts, as it is likely to get worse tomorrow.

In healthcare, many measurements conform to the rules of regression to the mean, such as blood pressure, pain and depression: it takes about 12 measures to obtain an accurate blood pressure for diagnosis ann the first, although not always is likely ot be much higher than the average measure you finally end up with.

The major issue in epidemiology is determining real change from the change expected due to such variation. In observational studies, often multiple baseline measures are required, which are not readily available in the published literature.

Indeed, the field of observational epidemiology would be greatly improved by the presence of more repeat measures.

Consider the case of the sports illustrated cover jinx. There are at least 73 instances of sportsman and women having been on the front cover, and subsequently suffering from a worsening or drop in form. All of these can be explained by regression to the mean. They were an outlier in the first place, with a greater than expected performance leading to their front cover exposure. Subsequent regression meant they were back to their average: the next time their performance was analysed it was worse – the cover jinx

However, Michael Jordan was on the cover 49 times and it didn’t affect him – there are always outliers.

References of interest: Bland, Altman BMJ 1994; 309:780; Barnett, AG Int J Epidemiol (2005) 34(1):215-220