It’s likely that most p-hacking is innocent, meaning that the researchers don’t realize they’re essentially cheating. For example, if you survey the data as you collect it, you might decide that once you cross over the p = 0.05 threshold you can stop collecting data and publish.
