Say you’re designing the character creation rules of a fantasy RPG (pen-and-paper or video game). Nothing fancy, just a few ability scores picked using a point-buy system and improved as the character gains levels. So far so good. But you also want to show off the amazing diversity of your fantasy setting, so you write something like this:
Elves are swift and gracious compared to clumsy humans. Elf characters get +2 Dexterity.
Seems reasonable on its face, and damn near universal in gaming. But what does it actually mean?
It doesn’t mean that “elves” – as in “the average person you meet in the elvish tree-city” – are swift and gracious compared to clumsy “humans” – as in “the average person you meet anywhere else”. All of these are NPCs. Their stats are determined not by the character creation rules but by an arbitrary decision of the GM (or the programmer, in the case of a video game). In fact, in any given campaign world, there will be maybe one person affected by this rule. Four in groups with an elf fetish.
And PCs are not a representative sample. That a human PC can be better at archery than the average elf doesn’t tell you much about the average human or elf. It only tells you something about the best archer in the world.
(Exception 1: if you’re building an MMORPG, where PCs will outnumber sentient NPCs 1000 to 1, then there’s nothing you can do to create a “swift elf” stereotype but to add PC racial modifiers.)
(Exception 2: if you’re designing Dungeons & Dragons 3rd edition, where every single NPC is built using full character creation rules and there are tables breaking down city demographics by Commoner class level, then disregard this post. You’ve got 99 problems but PC/NPC inconsistency ain’t one.)
It does mean that if a human PC and an elf PC both try to max out their dexterity, then at equal level the elf will have higher dexterity. But that’s really just passing the buck, forcing us to ask what “level” means. Instinctively, we think of something like “years of training”, and the rule seems sensible. Surely the human will have to train a lot longer before matching natural elfin grace. But, in modern RPGs at least, “level” actually means “two PCs at the same level should be equally matched in a fight.” (Or, if you want to get fancy, “two PCs at the same level should, on average, be able to contribute equally to the resolution of any challenge faced by their party.”)
So when we taboo “level”, we get “if a human PC and an elf PC who both specialize in dexterity are a match in a fight, then the elf will have higher dexterity.” Which is transparent nonsense. A PC’s dexterity affects how good they are in a fight, but not vice versa.
The major consequence of racial modifiers is that players who want to be skilled archers will have an incentive to play elves (which, again, has no significant effect on the number of elf archers in any given campaign world).
…Which is sort of a good thing, at least for bright-eyed newbies. It cuts down on decision-making time. Wanna play an archer? No need to wonder how a human archer differs from an elf archer, and which of them you’d rather be playing. The elf is better.
But it does considerable damage to more creative players who’d rather make
reverse stereotypes original characters. The game essentially tells them “you can fit in this mold, or you can be comic relief”. Admittedly, content-heavy games like D&D are more like “pick one of these 200 molds (then choose the colour, shape and texture of your bow-tie)”, but still: you have to use the game designer’s idea, not yours, or accept mechanical inferiority.
So, modest proposal: never use racial modifiers. I dare say that art like the above, and sentences like
Elves are swift and gracious compared to clumsy humans.
are all you need to help bright-eyes newbies figure out which race they want their archers to be. And if a player wants to be the one-in-a-million clumsy elf barbarian, more power to them.
EDIT: I will also note that racial bonuses can interact badly with ability scores rolled on dice with a bell curve, or with point-buy systems emulating them. E.g. in D&D, you save a lot more points by using your +2 to bump a 16 to an 18 than by bumping a 10 to a 12, which you can then use to make your character better in any way you please. But that’s an issue that can be solved with better math.