* subject to far less stringent controls
* much less likely to be participating in a lengthy appeals process
* subject to corrupting influences such as for-profit prisons
* vastly dominated by "plea agreements" in which people are convicted via coercive influence rather than any evidentiary process (and far worse if socioeconomically disadvantaged or non-white).
If 4% only refers to death row errors, the overall error rate for wrongful convictions is likely in the double digits.
Also, the error rate probably goes up again for non-incarceration results which let the system "slap people on the wrist" since that is seen as less serious (despite high incidental cost to life, career, mental & physical health). It's the legal system's equivalent of closing a support ticket.
And this is only classifying "correct judgement" vs. "wrongfully convicted." It fails to model cases where a criminal conviction was a really stupid way to deal with the problem, which could have been addressed via non-criminal means.