You're misreading. 538 says 97% chance of at least one prize. The EV calculation comes from the Federalist, which doesn't give an exact probability for a shared prize, but reduces the payout by 37% to account for it. This is based on another source, which also doesn't give a probability that I can see. (That probability might not be super helpful - you want the probabilities of one winner, two winners, three winners, etc. At some point it becomes rounding error, but I wouldn't be surprised if you want to go at least to three.)
But 538 guesses twice as many tickets sold as the Federalist does.
