This bias is present in many studies and can sometimes completely invalidate the findings if the authors do not adequately consider it or control for it. Selection bias refers to bias in the data that results from who is included — or excluded — from a study. For example, if an infectious disease has a high rate of asymptomatic infections, then including only cases with symptoms in a study could lead to inflated rates of mortality or mobility. Consider an example from COVID-19:
Out of 1,000 people, let’s hypothetically say that 400 of them do not have any symptoms, leaving 600 with symptoms. Let’s say that 50 people die from the disease. If a study only includes people with symptoms, then it would appear that the fatality rate is 8.3% (50/600). But if all cases are included, then the actual fatality rate from infections is only 5% (50/1000). When it comes to counting deaths, that’s a big difference.
Deeper dive
Many different types of selection bias exist. If you only assess the outcomes of people who complete a study and not those who drop out or are lost to follow-up, for example, the study findings could be biased if the drop-out/loss-to-followup rate is high, called attrition bias. It’s important to see whether the characteristics of those who did not complete the study differed much from those who completed it. Similarly, sampling bias or non-response bias could occur if a study does not take into account differences between those who did and did not respond to a survey. If responders and non-responders differ substantially, it limits how generalizable the findings are.
In vaccine studies, a type of selection bias called the “healthy vaccine effect” has to be considered: people who get vaccines tend to be healthier and follow more health-related guidelines than those who aren’t vaccinated. This can lead to some positive outcomes to be attributed to a vaccine when it’s actually more likely to result from other behaviors or overall health of the person who received the vaccine.