> Is that global statistics or statistics for "over policed" neighbourhoods the article and discussion is about?

Neither, it is US statistics.

Getting specifically the "over policed neighborhoods" statistics is a little trickier but when you do some location of arrest lookups you do see a higher percentage of arrests coming from those neighborhoods, so it would make for a reasonable assumption that those neighborhoods are a representative demographic of this dataset (if you're drawing from two jars and you draw 80% from one jar and 20% from the second jar, the marbles you have is a better representation of the contents of the first jar than they are of the second jar).