When returns from a class of classes of investment are measured, it is common to look at what investments in that class are available and how they have performed. The problem with this is that it excludes investments that are no longer available. This is survivorship bias. It is closely related to self-selection bias and backfill bias
Suppose we were looking at global returns on equities in the very long term (often done in the context of equity risk premiums). The obvious approach is to look at the world's equity markets and look at what returns an investor who invested in those markets at some past date would have made.
This only considers those markets that exist today. An investor who actually invested in the past may also have invested in markets that have since ceased to exist. If we look at returns over the course of the 20th century all the Eastern European markets, and some Asian ones, closed in countries that became communist. Investors lost all money invested in those markets. This exaggerates returns significantly as the worst performers are eliminated.
Survivorship bias can also distort calculations of returns on stock markets in much the same way, and a similar effect can distort returns on companies in an index (through changes in the companies included and excluded). The importance of the effect in the last case is arguable as the reason for most index changes is to keep indices representative of their markets.