If you look at original linked doc, you'll see that there are three 68-yr olds, that have on average 19.000+ points. That's why I only used in graphs age groups that have at least 100 developers.
Using a median instead of a mean might help w/ this.
Yes - median is a far better method to use in a situation like this, especially when what is being measured (reputation) has a bounded minimum and an unbounded maximum, and double especially when the sample size in each age bucket is relatively small.