Using the amount of total registered voters in each respective state and a 95% CI, we get the following margins of error
MI: ±3.939%
PA: ±3.723%
WI: ±3.811%
Depending on the exact lead (NYT only shows round percents, not specific numbers for each response), all of those are potentially within the top end of that margin of error.
Am I trying to claim that a swing from being down by ~4% to being up by ~4% means nothing and is indicative of nothing? Of course not. But man, most people really do not at all understand how statistics work, and I really wish people would stop talking out of their ass about it.