they sampled an incredibly small sample size. It's extremely easy to get fucked up results from assuming that you can make a poll representative of Americans as a whole. Like. where I live... most people in the state want him locked up (or you know. burned at the stake.) But, you go an hour out the cities and even the democrats there would be likely to express some hesitancy. Because it's trump country out there.
and that assumes the poll wasn't meant to get this result (for example polling in ways that get maybe more conservative democrats. or people simply lying and saying they're democrats.)
Actually the sample size checks out. I love it when people see "Smol number not as big as big number, therefore sample size bad" and I am going to pull a very elitist argument here and say that people at Harvard University likely know more about polling than you do, just saying.