This bizarre method of rounding is sometimes called "Banker's rounding". I
would love to see all examples of it expunged from all programming. For
scientific purposes, the up/down bias supposedly introduced by conventional
rounding is irrelevant, unless you very often come across exact x.5
numbers. Since money is quantized in rather large units, this happens
frequently. With physics data, exact halves tend to be very much the
exception, and making a special rounding case for them is silly.
Unless you are actually working with a banker (and using ROOT?) you
probably should just use conventional rounding (or use Nint and ignore the
odd behavior at 0.5, but I prefer to always have 0.5 round the same way).
Marcus Mendenhall