It's a question of correlation. Silver explains it
here in the context of highlighting the importance of North Carolina.
Long story short, if Trump comes back to win, say, Michigan, it would greatly increase his chances of winning several other states in the Great Lakes region, not only because he'd likely be outperforming current polls nationally but also because Michigan and Wisconsin and Ohio and Pennsylvania and even Minnesota are similar demographically.
So for example if Clinton is 75% to win three different battleground states that Trump needs to win, the chance of Trump winning all three is not 1.5% (25% x 25% x 25%). It's much much higher than that, because if the 25% chance hits in one state, whatever happened to give him that win would change the odds significantly on the other two. If those three states are similar demographically (say the Great Lakes states) it shoots up even more; if they're different (say Ohio, North Carolina and Florida) then there's less correlation and thus a Trump win in one doesn't necessarily mean a massive shift in the odds for the other two.