Survivor bias and how to overcome it
Professor Norman Fenton explains why claims of lower mortality rates for vaccinated may be just a statistical illusion. Worth reading in full. (Don’t worry; it’s not too complicated! Here’s a taste…)
In a previous article, we described the concept of survivor bias in studies that claimed better outcomes for covid vaccinated women in pregnancy: since the greatest risk to babies occur early in pregnancy, the babies of women who are vaccinated during pregnancy must already have survived the riskiest period.
In fact, a similar survivor bias more generally affects mortality rates for the vaccinated. If you see a study claiming much higher mortality rates of the ‘never vaccinated’ versus the ‘ever vaccinated’ you need to be sure it’s not just a statistical illusion due to survivor bias. This (7 minute) video provides an animated explanation:
The video shows this particular bias is avoided by using ‘person years in each vaccination category’ rather than people in each category. So a person who first gets vaccinated 6 months into a one year study and lives until the end of the year will be counted as 6 months never vaccinated and 6 months ever vaccinated.
The example is, of course, extremely simplified. Ideally, to calculate the correct number of person years in each category we need to know, for each person in the study, the exact date of each vaccination. And we also need to take account of the varying infection rate at different time intervals. That’s because the survivor bias is further exaggerated if (as was the case in most Western nations for the covid vaccines) the initial vaccine roll-out happened during the winter – meaning that fatality rates would inevitably fall anyway as more people were vaccinated. So, irrespective of the vaccine, more deaths were occuring at a time when more people were unvaccinated. Most of those classified as vaccinated would therefore already have survived the initial death peak when first vaccinated…
Norman Fenton
