If, let's say, a typical Elf in the population is on average two points higher in dexterity than a typical Human, then if PC Elves are considered to be on the exact same 3-18 dexterity bell curve as PC Humans the overall setting population isn't being reflected well at all. The Elves are being mechanically forced to be more Human-like (and not the reverse; as the same lack of baked-in adjustment applies to Dwarves and Hobbits etc. whose typical stats vary in different directions from Human than do those of Elves), thus significantly reducing the difference between PC-playable species.
Not that a flat + or - number is the best reflection either; far better is to determine where in relation to the 3-18 bell curve the low and high extremes for each species' stats would be, and then adjust the whole bell curve to that. So, if one decides that the bell curve for Elf dexterity runs from 6-19 instead of 3-18 (3-18 being the universal Human baseline) then a roll of 3 in dex gets boosted to a 6, a roll of 18 goes up to 19, and a roll of 11 goes up to 13. Similarly the range of Hobbit strength might only be 3-16, meaning a rolled 18 gets knocked down to 16, a rolled 10 goes down to 9, and a rolled 3 stays as a 3.