He may have worded it slightly awkwardly, but the point is that you're looking to predict results, not pluck them from datasets. If you start with data, and th

He may have worded it slightly awkwardly, but the point is that you're looking to predict results, not pluck them from datasets. If you start with data, and then look for patterns you can almost always find them. Whether they are predictive is what you're trying to understand. I would also note that he didn't say 'discard data that disagrees with the hypothesis', which I think is what you're commenting on. This is as you correctly point out dangerous, and not how you science.