You can't really improve this until the lead author doesn't have a final say on how to treat outliers.
In any given study, there are going to be hundreds of special cases in the data that you didn't anticipate, and you have to decide whether to include or exclude them.
Any researcher will subconsciously be more sympathetic to arguments to exclude subjects that go against the principal theory, and less so to subjects that confirm it.
And it's a battle of reasonable arguments, most of the cases aren't bright line fraud or misconduct, they're just humans finding some arguments more compelling and the impossibility of escaping our own biases. (And yes, sometimes it's fraud, but fraud is just the tip of the iceberg if we're talking about genuinely improving the reliability of scientific findings.)
Prepublication is helpful, but more and more I'm convinced that the only way to do proper science would be to completely disaggregate study design from study execution.
In any given study, there are going to be hundreds of special cases in the data that you didn't anticipate, and you have to decide whether to include or exclude them.
Any researcher will subconsciously be more sympathetic to arguments to exclude subjects that go against the principal theory, and less so to subjects that confirm it.
And it's a battle of reasonable arguments, most of the cases aren't bright line fraud or misconduct, they're just humans finding some arguments more compelling and the impossibility of escaping our own biases. (And yes, sometimes it's fraud, but fraud is just the tip of the iceberg if we're talking about genuinely improving the reliability of scientific findings.)
Prepublication is helpful, but more and more I'm convinced that the only way to do proper science would be to completely disaggregate study design from study execution.