Two runners compete on a course with changing slope. One started earlier, the other later, and the judge, unaware of this, compares their speeds. So it is with virus variants: data about them arrive with a delay, and if the delays differ, the estimated advantage of one over the other may be wrong.

The advantage of a variant is usually estimated from the slope of a special curve that shows how quickly the share of the new variant among all infected grows. But this estimate assumes observations are accurate. In reality, days pass between infection and sample registration, and different variants may have different delays. Then the comparison happens at different phases of the epidemic, when growth conditions are no longer the same.

The authors tested this with computer models of epidemics and found that the error can reach tens of percent. Even when the curve looked almost perfect, the advantage estimate was biased by about a third. This means that from frequency data alone, one cannot confidently say how much faster one variant is than another without accounting for delays.

Now the judge knows: to fairly evaluate the runners, one must account for who started when. Similarly, scientists need to account for observation delays before drawing conclusions about the advantage of virus variants.