In the cases cited in the article, I strongly suspect “zero decimal places of percentage” is sufficient to show discrepancies (that discrepancies of full percentage point or more will be present, given the admissions rate for one group changed from 73% to under 50%).