To make up an example:
Let's say that betting $1 in some way has a 4% chance of winning, and the payout is 20x bet, so if you win on one ticket, you get $20. So, you buy a lot of tickets (say 1000), and you pay $1000. Most of them don't win anything, but 4% of them win $20; $1000 x 4% x 20 = $800; you have won some, but you have $200 less than you started with. Not a good deal.
However, if nobody wins the jackpot, it is distributed amongst smaller payouts, which changes the equation: still 4% chance, but payout is now $30. Again, you pay $1000, buy 1000 tickets, most don't win anything, but the 4% that do, you get $1000 x 4% x 30 = $1200. You have $200 more than you started with. Profit! Of course, this will only work if you buy many tickets; buying one or two, you're unlikely to win anything at all.
The actual numbers were different in these cases, but the general principle is the same.
That's the important part here. You can always win the lottery if you can buy enough tickets. To match all (n) numbers, you need to buy one ticket for every possible combination. To match (n-1), you need to buy a lot less.
But in a normal game, the cost of doing so is more than the payout. So if I have a dollar-game "guess a number between one and ten", and you can win $6 - it's not worth buying ten tickets, even if it guarantees you a win. But if I say "one day only, double the winnings!" - now $10 guarantees you $12 back.
The odds don't change, just the profitability of them.
What's interesting with the rolldown mechanism is not just that the "expected payout" of a ticket becomes worth more than its price, but that the number of winners is such that you can actually reasonably expect to turn a profit by buying a modest number of tickets.