The idea behind the gnome effect is simple. Let’s say you’re planning on releasing a hypothetical edition of D&D. You want to determine which races are important to the game, so you conduct a poll and find that only 10% of gamers play gnomes. That might make it seem obvious that you can safely cut the gnome without much trouble.
The problem with that line of reasoning is that we don’t play D&D by ourselves. We play with a group, and when looking at rules changes or any other alteration to the game you have to consider its effect on the group. Let’s look back at our gnome example. One out of ten gamers plays a gnome. However, let’s say your data shows that the average group consists of five players (not counting the DM). That means, roughly speaking, half the gaming groups have one player with a gnome character. That number is likely lower, since some groups might have more than one gnome, but it’s a rough approximation that serves to illustrate the larger principle. You cannot measure change and its effects on the individual level. You must look at it on the gaming group level. Delete the gnome from the game, or change it in a way that gnome fans dislike, and you’ve given about half the gaming groups out there a good reason to tune you out.