2 things:
1. Significant oversampling of Dems skewing results
2. What did the polls say one month before November 2016 election?
Oversampling of Democrats, usually by over polling heavily Democrat districts, is one way these are skewed. Population is concentrated in the major cities but Congress (nor the Presidency) is not determined by population concentration.
Maybe not in this poll but for 2016 one of the polling faults was assuming independents would break 8 or 10 points for the Democrats when the reality is independents break pretty close to the same ratio as party affiliation. They also over-sampled independents in most of those polls, probably on purpose because they knew that would “look fair” on paper but give their results a strong skew to the Dem candidate.
Pollsters usually want to get results that get them headlines, attention and money (jobs/contracts). Accuracy is less important. They want to be hired not just for political polls but for all kinds of private enterprises, market analysis etc. And the final analysis it doesn’t matter much if they are 7% off in their “do you prefer Coke or Pepsi” polls. But in political polls they are trying to psychologically manipulate people because they think people will want to vote with the majority (or the majority as they are told it to be).