I don't think it's a good ground to rejected observational data. If you want the published data to be less biased you should introduce as little arbitrary (not connected to the quality of data itself) barriers as possible.