The Severinghaus errors are known to accumulate with age. In essence, toothwear aging almost never over-estimates age, but does under-estimate age, and the error increases with age. A great study done by Mick Hellickson at the King Ranch, with known age deer (ear-tagged as fawns) found that toothwear aging of yearlings was accurate, about 25% of 2 1/2 year-old wear-aging was a year too young, 50% of 3 1/2 aging was a year too young, 75% of 4 1/2 aging was too young (with about 25% of the incorrect ages being 2 years too young) and 100% of 5 1/2 aging was too young (with about 50% of those being two years older than toothwear indicated).
Personally, I use a combination of toothwear and body conformation to produce an age (for bucks). Does, I stick with toothwear, knowing the older does are being under-aged.