Because otherwise the data would be artificially lower in areas with more children.
For example, imagine a suburb in Utah filled with college educated software engineers with big Mormon families. If you count the kids, it might look like people there don't have degrees.
Doesn't a bachelor's take 4-5 years, with people starting around 18-19? I guess we're only talking about a year or two so the higher age is to help cut down on the noise (doubt there's many people with bachelor's dying before 25 to skew the results)
21-22 is the average age to complete a bachelor's degree, so I'd guess - other than eliminating children, who couldn't possibly have gotten degrees yet - just evening out the data a bit to account for later starters or longer programs? They probably had a target 90% of degree-receivers or something like that