Marist is a known left leaning polling operation.
We have the same problem in Minneapolis, Minnesota. The Star Tribune consistently under estimates Pubbies at this time of the year by 4-8% and occasionally more. While the results tend to improve closer to the election, the Star Tribune usually still under estimates Pubbies.
IMHO part of the problem is sampling error (under estimating the true proportion of Pubbies)and the fact that the "undecideds" consistently in our State break toward the Pubbie column at the end of the election cycle. It would be interesting if thes findings are mirrored in other states.