Vox's guide to the battle for the Senate

What's behind the change? All these models incorporate 'fundamental' factors in their forecasts that tended to advantage the GOP - things like incumbency and presidential approval ratings - but, as the election nears, they're relying more heavily on pure polling, and Democrats are doing surprisingly well in the polls.

Meanwhile, two of the models that most favored Democrats - Sam Wang's and HuffPostPollster's - have moved slightly toward the GOP. (The Daily Kos model has basically stayed the same.) Those models already relied on polls alone. So another part of what's happening is that the models are converging. As more polls come in and give everyone more data to work with, outlier projections are moderated.

These changes shouldn't be overstated - in a probabilistic forecast, 60-40 looks a lot like an ordinary coin flip, as the Upshot's Amanda Cox explained recently. But with this year's battle for the Senate looking so close, small differences like these could end up determining which party has control for Obama's last two years. Here's how things are looking now.

1) Democrats are favored in the purple states

In early September, I outlined how some models showed North Carolina's Senate race as a toss-up, but others gave a clear edge to Democrat Kay Hagan. The summer polls had shown 'a small margin but an extremely consistent margin' in Hagan's favor, and the Post's model accordingly gave her a very high chance of winning , its co-creator Eric McGhee told me. Other models were more skeptical of her chances - either because they treated that small margin with more uncertainty, or because they were more skeptical of the polls themselves. Seven of the 10 recent polls had come from avowedly partisan pollsters, some polled registered voters rather than likely voters, and some came from firms with poor track records in past elections.

Since then, eight new polls of the race have been released - six showing Hagan ahead, one showing a tie, and just one showing her challenger Thom Tillis with a narrow lead. This seems to have cleared things up: Hagan is ahead. The more uncertain models' forecasts have moved accordingly:

This is part of a larger trend that's been occurring in polls from presidential swing states ('purple states'). Since my early September post, all four new pollsof the Colorado Senate race have showed Democrat Mark Udall ahead. And in Iowa, where the race between Bruce Braley and Joni Ernst looked very tight in the late summer, the three newest pollsreleased show Braley ahead. (Update: Wednesday morning, though, a new poll showed a sizable lead for Ernst.) Meanwhile, Republicans hoped to put Democratic-held seats in Michigan and New Hampshire in play, but haven't yet had much success in the polls (though the New Hampshire race has tightened a bit) .

It's still far from assured that Democrats will manage to sweep all these purple state races. In Iowa particularly, the Democratic advantage in most models is a small one. B ut if they do manage to do so, Democrats will have 49 seats - and, with Vice President Joe Biden prepared to cast tie-breaking votes, they'd need just one win in a red state to keep their majority.

2) The red states look good for Republicans, though polls have been sparse in some

Is a red-state sweep possible for the GOP? First, they'd have to hold on to their own seats in Kentucky and Georgia - which looks very likely. Despite the media-friendly nature of the Kentucky race, polling has clearly shown Mitch McConnell ahead and election forecasters view him as the overwhelming favorite. In Georgia, Democrat Michelle Nunn also trails, and unless she can manage to win over 50 percent of the vote in November, she'll face the added difficult challenge of winning a runoff election in January.

Second, the GOP would have to knock off Democratic incumbents Mark Pryor (AR) and Mary Landrieu (LA). Beating incumbents is rarely easy, but these two Democrats have had little good news in the polls lately. Most recent polls shown Pryor trailing Rep. Tom Cotton, and as a result, he's moved from a favorite to an underdog in Sam Wang's snapshots. Polling has been rarer in Louisiana, but Landrieu also have to win a runoff if she gets less than 50 percent of the vote, as looks likely. And she's led very few of the head-to-head polls against Rep. Bill Cassidy conducted this year. So Republican odds look good in both of these.

Then there's the Alaska race between Democrat Mark Begich and Republican Dan Sullivan. Here, the polling has been extremely sparse - but the only two nonpartisan polls to have come out recently have shown Democrat Mark Begich behind. (They've moved him from an overwhelming favorite in Sam Wang's late August snapshotto an underdog.) Additionally, Nate Silver wrote recently that Alaska polls have overestimated how well Democrats would perform in every single competitive race there since 2000.

3) Forecasters are still trying to figure out what to make of Kansas

If the GOP manages to defeat the three red-state Democratic incumbents while losing all the purple state races, control of the Senate will hinge on the strange election in Kansas, where Sen. Pat Roberts (R) is pitted against independent Greg Orman. (Democrat Chad Taylor dropped out of the race, and is trying to get his name taken off the ballot.)

Before this week, only two polls had attempted to measure what the race might look like without Taylor. First, before Taylor even dropped out, a poll pitting Orman against Roberts showed Orman up 10. Second, a poll after the news gave Orman a one-point lead - but the respondents were told that Taylor had dropped out, even though they wouldn't be told that in the voting booth.

The bizarre turn of events has proved a challenge for the forecasters, and most have responded with caution. By Monday of this week, most models still gave the advantage to Roberts or viewed the race as a toss-up. But one had a very different view:

Sam Wang emphasizes that his race-to-race numbers are 'snapshots' of current conditions rather than projections for November, which sets his numbers apart from other modelers. But Wang's results for individual races matter - because while his model has gotten a lot of attention lately for being the most favorable to Democrats, much of that advantage now hinges on his strong estimate that Orman is ahead. And, on Monday, he was basing that estimate on just two polls showing an Orman lead.

'My general philosophical stance is that polls are more informative about voter opinion than any other parameter,' Wang told me recently. Essentially, he believes that the best measure of the state of a Kansas Senate race is a poll that tries to measure the race. He argues that anything else - whether it's fundamentals or added state-based measures of uncertainty - is ' special sauce' that dilutes a direct measure of public opinion.

The other modelers had many plausible reasons for discounting those two Kansas polls. Voters might not know much about Orman yet, and he hasn't yet been exposed to many negative attacks. Taylor's name, if left on the ballot, could draw precious votes away from the independent. Many undecided voters remained. And Kansas is so conservative that it's difficult to imagine a Republican incumbent losing his Senate seat.

But on Tuesday morning, a new poll was released showing Orman up seven. Two polls became three, and most of the models moved in Orman's direction. We'll keep an eye on what happens next.