I don't really know, but I'll guess. Maybe the estimate is done by taking the total for one day and subtracting out the total for the previous day? If the data

I don't really know, but I'll guess. Maybe the estimate is done by taking the total for one day and subtracting out the total for the previous day? If the data were a bit messy and revised down, this might result in an occasional negative value.