Many different factors go into a decision like this. I'm reminded of the Bush campaign making a late play for California - a totally preposterous proposition - at the end of the 2000 race. Then and now, I believe the main goal there was to unnerve and demoralize Democrats. It was, in essence, a confidence play. Take a dead even race and make people think you know you're going to win to get inside their head and make your opponent's voters think there's no hope. It still seems possible to me that if Bush had focused on the critical states he might have won Florida fair and square. (Karl Rove's move here was actually the subject of the second post I wrote on TPM on November 13th, 2000.)
Another big factor is self-delusion. The Romney campaign was staffed by top rate professionals. But they managed to convince themselves that they were winning and were shocked at their defeat on election night. The Trump campaign is made up of know-nothings and charlatans. The chances of total self-delusion aren't just high. They're close to a certainty.
There's a good chance that Trump is doing this because he's an idiot. But even if he weren't an idiot, I think there might be a decent logic to the decision. Trump has a lot of states he absolutely has to win. If he loses just Florida, or just North Carolina or just several other states on their own, he's done. The problem is that even if he wins all those states he probably comes up just short. The must-win states by most measures simply don't get him to the actual win. To win, he really needs to pick off at least one and likely more than one states that people are now assuming are fairly solidly blue. I'm thinking Wisconsin, Michigan, Pennsylvania, Colorado, perhaps Virginia.
We still don't know the full impact of the Comey letter. It seems to have had a limited effect on the race so far. The Trump campaign has to assume that the modeling of the electorate chosen by most pollsters is at least somewhat off or that there is a late break in his direction. If either of those is the case, he might win Ohio, Florida, North Carolina, etc. But without a couple seemingly solid blue states, he still falls short. The better bet may be to assume a surge or misreading of the electorate and try to grab one or two blue states on top of that. If the surge or hidden voters are there, the battlegrounds will take care of themselves. If they're not, none of it matters anyway.