Trump does best with working class whites and does worse with groups like college-educated and Hispanics. Ohio has lots of working class whites and not so many Hispanics. Florida and Nevada have lots of Hispanics. This leaves his support in Florida softer than in Ohio.
Note that Ohio's partisan lean has been just to the Republican side of the country. For this reason, it tends to vote for the winner. But in a close Democratic win like 1960, it can go Republican.
It could happen this year. Most projections of close elections have Ohio going for Trump even if Clinton wins. She'd need 320 electoral votes or so to win Ohio.
Some of the difference is just the polls that are being taken. For example, in Florida, the Monmouth poll has Clinton +5. But the previous Monmouth poll had Clinton +9. The JMC Analytics poll that just dropped out of the average had Trump +4. And the previous JMC poll had Trump +5.
House effects are when a poll has a consistent bias in favor of one side or the other. In this case, it looks like Monmouth has a pro-Clinton house effect while JMC Analytics is pro-Trump. At least in Florida. If JMC Analytics releases next week and Monmouth drops out, the polls will shift again.
Organizations like FiveThirtyEight measure pollster bias and compensate for it. This helps reduce the random noise movements.