But pollsters have the luxury of cherry-picking poll data from specific demographic sources. In Wisconsin the Marquette Law School poll took samples from 802 Wisconsin Registered voters while the New Hampshire poll is "based on a probabilistic sample of 442 likely Democratic presidential primary voters in New Hampshire" The New Hampshire sample is roughly half the size of the Wisconsin sample.
Not only does size matter, even the period of time in which the sample was taken can make a difference depending on news developments. The Wisconsin poll was taken from Thursday August 13 to Sunday August 16, The New Hampshire poll was taken from Friday, August 7 to Monday, August 10.
August 8 was when two members of a group called "Outside Agitators 206" claiming to represent Black Lives Matter shut down Bernie's speech event in Seattle. It turns out the two women involved in the incident acted on their own without consulting other activists. On August 15, Bernie was in Iowa drawing huge crowds and sniping at Donald Trump over Helicopter rides. Who knows what the differences in the polls would be if these events didn't take place until after the polling was done?
So, polling is basically a crap-shoot that must be taken with a grain of salt.