I think we may need to consider that people's parses are off, with such inconsistency in the data that we have.