36% isn't great in absolute terms, no. But if you're willing to accept improvement and not demanding utopia-or-nothing, reducing the rate at which members of a

36% isn't great in absolute terms, no. But if you're willing to accept improvement and not demanding utopia-or-nothing, reducing the rate at which members of a subpopulation are being arrested repeatedly by ~1/3 would be a big deal from a public policy perspective. My criticism of that statistic would be different. Ones that are structured like "X% reduction in Y happening more than threshold level Z" smell of p-hacking. A simple "Gets arrested X fewer times (p=0.15)" is more trustworthy to me than "X% less likely to get arrested more than 5 times (p = 0.03)".