Well, reading the entire article, there were a/b tests there, so the increasing experience claim doesn't explain the result.
A) 5$/h, 7.50$/h, 10$/h, 15$/h
B) 10$/h, 15$/hThe data comes from a source that needs to be double-blind but wasn't.
What would be interesting would be to do have one party write two separate emails talking about a different subject. Then randomly switch the name/gender of one of the emails.
Then have a second party "grade" the emails. With the "author's name" clearly marked on the email.
I'd like to see the results.