Princeton Election Consortium

A first draft of electoral history. Since 2004

Iowa and Colorado: Is early voting showing up in polls?

October 29th, 2014, 9:43am by Sam Wang



Over at FiveThirtyEight is a new essay saying that there hasn’t been any concerted movement during this campaign season. Broadly, I agree with their objection that journalists and pundits overuse (and thoughtlessly use) the word “momentum.” However, in this case a close look at the data suggests there was strong movement from September through early October – which then stopped or reversed slightly.

The movement is easier to see in a polls-only snapshot, which is what is published here. In particular, Iowa and Colorado moved toward Republicans throughout September – and in the last few weeks, moved back toward Democrats. One possible cause is early voting – which should show up in surveys.

Speaking as a natural scientist, I would define momentum as movement in one direction over an extended period – and perhaps with the potential to continue. Here’s what a polls-only snapshot looks like:

The big jump in early September corresponds with Chad Taylor (D) dropping out of the Kansas Senate race. So just look at the September and October data. The average expected number of seats declined throughout September, with a big jump for the GOP late in the month. As many readers will recall, this movement corresponds to polling gains for the GOP in Alaska, Iowa, and Colorado.

The change is more vivid when expressed in units of Meta-Margin: how much polls would have to change (or be inaccurate) across the board to create a perfect toss-up for control:
Here, we can see a sharp drop of about 2 percentage points in late September. In the case of Colorado, it occurred shortly after the Udall-Gardner debate, when Udall made remarks claiming to speak for journalists who were killed by ISIS. There may have been other triggering events: for example, around that time President Obama announced that a change in immigration policy, which would have helped Latino immigrants, would be delayed until next year. That could have hurt Democratic candidates.

Since that time, in the last few weeks we’ve seen both the average Senate seat count move back toward parity. To put it more precisely, we now have a near-tie, with a Meta-Margin of R+0.1%. If polls are off by even 1% in either direction, then Democrats or Republicans are favored. We will find out after the election.

>>>

Recently, early voting has been the subject of speculative coverage. Molly Ball at The Atlantic suggests that early voting in Iowa is going poorly for Democrats. She points out that a similar number of registered Democrats and registered Republicans have voted, which is a real improvement for Republicans since 2010. Democrats counter-argue that they have done a good job of identifying and recruiting unaffiliated voters who tend to vote Democratic. Unfortunately, this evidence is fairly cloudy. Also, even if early votes trend one way or the other, a key point is whether they still would have been cast otherwise. If so, then early voting is mainly a strategic move to get votes safely stowed away, leaving more time and energy for Election Day efforts.

There is one possible route to resolving this issue: polling. Pollsters ask respondents if they have voted early, and if so, who they voted for. In principle, these questions should capture the effects of early voting. In a new Quinnipiac poll, Bruce Braley (D-IA) leads 57%-36% among those who have already voted. The same survey shows that Ernst leading overall by 4 percentage points. However, for such a narrow margin, it is better to look at the overall median among all pollsters, which shows Ernst leading by only 0.5%. It appears that this race is very close in the home stretch – too close to call.

The Colorado race is also narrowing, with Cory Gardner (R) leading by a median margin of 1.0%. This could be because of voting entirely by mail, which is new in Colorado this year. Or it could be because of genuine movement. Whatever the case, this race is also too close to call.

Tags: 2014 Election · Senate

69 Comments so far ↓

  • Matthew

    Just a thought, all the early voting numbers are being compared to 2010 a very good Republican year. These Senators won in 2008, a very good Democrat year. Of course 2010 was the last midterm so a comparison to 2010 makes sense but don’t we also need a comparison to 2008 since that is when the incumbents won?

  • atothec

    What is the lag time for early voting? I would imagine numbers posted today for instance are from ballots that were mailed at least several days ago, right?

    • securecare

      I can only comment on Washington state.

      Here one can mail in their ballot which must be postmarked NLT Tuesday OR one can drop it off at a designated location anytime up to poll closing time; 20:00 (8:00 PM) PST which is GMT/UTC -8 (since we “fall back” to standard time Sunday morning). I dropped mine off over a week ago.

  • Catherine

    REALLY dumb question. Is Sam predicting Dems hold Senate majority on Election day? I don’t understand the 45 +-15%

  • Olav Grinde

    “Thanks for the data, did some quick calculations with the assumption that the current 1,040,634 goes to 1,800,000 by election day.”

    Phil, it should be point out that the 1.8 million figure is total ballots in Colorado in 2010 – not just the early vote.

    That said, it will be interesting to see how things develop.

    • Phil

      Yes, sorry if I didn’t make that clear. If you compare Colorado to all the other early vote states you can see the percentage of 2010 total vote already cast is much higher in Colorado and Florida, two states that moved to mail ballots. So I’m making an aggressive assumption about turnout.

  • Matthew

    Byron York just tweeted “New numbers from Colorado Secretary of State: Dem ballots received: 337,897. GOP: 443,240. Unaffiliated: 259,947.”

    If we assume each candidate gets 90% of their parties vote, I think that is what we generally see in all races, Udall needs about two-thirds of independents for a tie. If Gardner has a problem with his base that would lessen the lead Udall needs among independents.

    Also for those expecting the Dems to continue gaining on the GOP the gap is now 10.1 points larger than the 9.4 point R lead reported on electproject a couple days ago. Pretty much the only hope for Udall is that Gardner is struggling with his base.

    • Phil

      Thanks for the data, did some quick calculations with the assumption that the current 1,040,634 goes to 1,800,000 by election day. If that’s true to get to my predicted R+5, if I holds a steady percentage, you’d need 292,103 D vs. 276,760 R, or 38% of remaining ballots to be D vs. 36% R. So far we have 43% R, 32% D, 25% I. Total composition of group would be 40% R, 35% D, 25% I. This is low for I for the final result, but my assumption is they are more likely to vote on election day.

      That doesn’t sound horribly unreasonable to me (total votes in 2010 were 1,828,323 and everyone is expecting the mail voting system to raise turnout by a few percentage points). If you prefer 1,600,000 then D’s need to outperform R by 40% to 35%. If it turns out I’m wrong next Monday/Tuesday I’ll certainly say so, but voting patterns and turnout are not linear. The first million does not determine the next 800,000. Specific county data where Ds are underperforming or Rs over can tell us something. My understanding is that right now Boulder and Denver are lagging, if that continues than Udall is certainly sunk.

      BTW, the opposite pattern is occurring right now in NC, where as of right now the D lead is 16. No one expects that to last. The key question with early voting is by getting someone to vote early did you just shift their vote or did you actually add to the tally? If it’s just a shift, then it doesn’t matter what order people vote in (which is the D argument in both CO and IA). The R argument is that they are actually adding to their tally and that the Ds overstate how many additional votes they add through early voting.

  • Olav Grinde

    Meanwhile, the number of Americans stricken from voter registration lists, and who have thus lost the right to vote, is astoundingly high. This is worth reading.

    http://projects.aljazeera.com/2014/double-voters/

  • biggest data

    Apologies for slightly off-topic, but are there publicly available data sets giving detailed vote breakdowns for past elections, especially the House? All I can find are in PDF or similarly unfriendly formats.

    • Sam Wang

      The House Clerk’s office is the most comprehensive source, but as you indicate, it’s PDFs. If you are willing to pay, then Dave Leip’s atlas of election data provides a rich source of information.

    • Olav Grinde

      A couple of years ago, I search long and hard for a breakdown of the 2012 election. In particular, I was eager to see the state-by-state overview of provisional ballots and cast ballots.

      To my astonishment,there does not seem to exist any overview, nor any final count. At least I wasn’t able to find it. If anyone can link me to a source, I would be most grateful.

  • Me

    I distrust early voting analysis. In most cases in the big elections it seems that all it does is make people vote faster lowering election day totals. It doesn’t matter if a party has an early voting lead if they give it all back on election day.

  • Josh

    Sam, have there even been any studies of recent (last 2 cycles) polls and the ages of respondents? I often wonder whether or not polls tend to underrepresent certain age groups, especially now with so few millenials owning land lines and preferring text-based communication. The methodology of the recent Harvard Kennedy Institute survey would seem to be moving the right direction: http://www.hks.harvard.edu/news-events/news/press-releases/iop-youth-voter-survey-fall-2014

  • Amitabh Lath

    Colorado’s ballot has the following questions:
    Recognizes unborn children as persons, K12 funding by gambling, open school board meetings for collective bargaining, and GMO labeling for food.

    I can imagine the “personhood” ballot question being as hot button an issue as the Senate race. And for some, the GMO question as well.

    • Phil

      The personhood amendment is unfortunately the third consecutive time, I believe, that this has been on the ballot–it’s way to easy to get stuff on the ballot in Colorado–and it’s probably going down in flames. It might help get some people to the polls, but I doubt it. There has been 0 advertising for it, pro or con.

      The GMO question is actually pretty complicated–it’s a shoddy proposition that exempts large swaths of food. I don’t know what the polling is like on it, but it has no juice behind it (a ton of anti-ads but no pro).

      The gambling question has a ton of money on the pro and con sides (since you’ve got casinos lined up on each), but I don’t think it’s going to boost turnout at all. We have gambling in Colorado in a few mountain towns.

    • Sod

      Luckily, most of the libertarians tend to vote no on the personhood thing. But these same people would tend to vote Republican so unfortunately it is not motivating only the left.

  • Canadian fan

    Given Colorado’s history of unexpected surprises, we can expect big surprises on Tuesday. For four reasons : 1) Democratic strength has generally been underestimated by pollsters between 2 % and 3 % by election day. 2) Spanish-speaking Latinos – they tend to be more Democratic than their English-speaking counterparts, harder to locate on surveys, and ultimately elusive to English-speaking surveys. 3) All-mail balloting. Analysis of Washington State’s before and after shift revealed a resulting uptick of 2 % to 4 % in increased state-wide voter turnout. The largest percentage of this came from infrequent voters – generally those who only vote in presidential elections. 4) Democrats unprecedented ground game, which has specifically targeted infrequent voters. With all these factors in play, Colorado could be set to deliver an unexpected punch.

    • Olav Grinde

      I wish I shared your optimism. The running tally of Prof. Michael McDonald’s Early Voting statistics currently shows that 905,500 Coloradans have already voted. This figure is equivalent to one-half of the total ballots cast in 2010.

      What gives me pause is that Republicans have 41.9 % to 32.5 % lead over Democrats. That is almost 10 %!

      Granted, over 25 % of the early votes are from independents and people with third-party registrations – but incumbent Senator Mark Udall would have to obtain the lion’s share of those to survive this challenge.

      I cannot help but wonder whether the 2014 Democratic ground game has failed in Colorado.

      Perhaps Colorado will deliver “an unexpected punch”, as you say, but I am not betting on it.

    • Phil

      You’re assuming the next 1,000,000 votes are going to be like the first, it won’t happen. The question is how much will the early vote window narrow before election day. You’re also assuming that the 2014 vote total will be roughly equal to the 2010, which may or may not happen. A robust ground game isn’t one that delivers all of the votes in the first few days you can return a ballot. Those people were almost certainly going to vote without anyone contacting them. Colorado is going to rise or fall on how many people get cajoled into returning ballots this week or voting on election day itself.

      My prediction is that by election day the R advantage on ballots returned is somewhere around 5%. Rs outnumber Ds in this state, BTW, and Obama’s win in 2012 was with more registered Rs voting than Ds, so that doesn’t mean Udall is doomed.

      A final point: Gardner doesn’t necessarily lead Udall by 9 points right now in early votes cast (he might, but we have no way of knowing). Same is true in Iowa, and any other state where people seem to be automatically assigning all of the R and D votes to the respective camps. It’s close enough to give us an idea of where things stand, but in a race that is a couple of percentage points apart, it’s dangerous as a guide.

  • Olav Grinde

    The closeness of this election is insane. It might be decided by:

    – Democratic (or Republican) GOTV efforts in a single state
    – Failure to process voter registration
    – Restrictions or reductions of early voting
    – Better GOP turnout in a run-off
    – Fears of Obama’s communist Islamic regime
    – Anger at the Koch brothers
    – Ebola fears

    And last but not least:

    – American courts

  • Amitabh Lath

    Has anyone looked at the Public Questions on the November ballot? Sometimes they can drive turnout.

    Looking at my sample ballot (NJ), I see a ballot question on giving courts the option to deny bail, state money to environmental groups, and a couple of Middlesex county proposals including one to partially restore funding to women’s health care groups that were cut by the state.

    I can imagine there will be NJ voters driven to the polls mainly by these questions. Are there questions on other states’s ballots that could drive turnout?

  • ptuomov

    Some thinking out loud below.

    As of 10/28, 143,000 registered Democrats and 138,000 registered Republicans had voted. The no-party votes are 66,000. That’s D 41%, R 40%, and I 19%.

    According to the same poll that Sam Wang quotes above, Ernst has a 9-point lead among the independents, 54.5% to 45.5%. If we apportion the 19% independent absentee ballots by those fractions and assume that non-independent voters vote their party, we get an estimate of R 50.4% and D 49.6% for the absentee ballots already cast. There are a lot of hard numbers in this calculation so it shouldn’t be too far off.

    Now, the data point in the poll that Sam Wang quotes is that the poll says that “Braley leads 57 – 36 percent among those who already have voted.” First of all, let me expose my ignorance about polling: That’s a really weird statistic since you’d think that the percentages in votes cast would add up close to a 100%, not to 93%. There are no other relevant candidates in the race, did those people not check a box or did they forget who they voted for?! But beyond that, since the relatively hard numbers suggest that the early voting is about even, isn’t by far a more likely explanation to the poll result that the poll happened to by chance have Democrats overrepresented? If so, doesn’t it mean that the poll’s headline result, Earnst 48% and Braley 46%, likely has an error in the direction that understates Earnst’s and overstates Braley’s numbers?

    Advance thanks for the education!

    • Edward G. Talbot

      @ptuomov – you have hit on the problem. We need to know the party breakdown of the already cast ballot voters in order to really judge that 21%. And it does seem likely that it oversampled dems. But we really have no idea – I emailed Quinnipiac to ask them the breakdown of those early voters and I have not heard back yet.

      I would agree that if the early votes, which are a third of those polled, are way off, it certainly calls into question the rest of the results too. But I wouldn’t necessarily assume Ernst is understated. We do not know what kind of fudging Quinnipiac does with their samples, and the detail of the poll says the percentages are “weighted”.

      All the more reason not to hang your hat on a single poll!

    • Davey

      Ptuomov – the big problem with your numbers is that they are skewed by one thing – an independent voter isn’t the same as a voter registered with no party preference. An unknown lump of that 19% are voters who would call themselves Democrat or Republican to a poll, but for various reasons don’t register as such. There are certain perks to this in Iowa’s open caucus system. We can’t therefore reliably assume that 19% reflects the same voting patterns as independents poll. I can’t speak for iowa, but here in California a huge chunk of non-party voters are Republicans, which is why the GOP still (sort of) competes here despite only 27% of voters registering Republican.

    • Matthew

      Usually subsections of the electorate, like how many blacks will vote for candidate x or Hispanics or single women etc, have significantly higher MOEs that the headline number. A lot of times the sample size is too small for the result to actually be meaningful. I suspect that is the case with asking early voters who they voted for. If they polled only early voters about who they already voted for that would make a great survey but the MOE is probably to large to make any definitive statements about it.

  • Jay Wilson

    I don’t want to change the subject, but …

    This Alaska Hellenthal poll from days ago seems unique. http://www.scribd.com/doc/244305469/AK-Sen-AK-Gov-Hellenthal-Associates-Oct-2014

    “Of the 403 interviews, 318 (79.0%) were conducted on cellular phones and 85 (21.0%) on land-lines.”

    …snip…

    “10. (IF LANDLINE PHONE, ASK…..) Is your residence accessible by cell phone?”
    Yes………397……..98.6%; No……….6………1.4%”

    “11. (IF CELL PHONE, ASK…..) Do you have a wired land-line at your residence?
    WIRED LAND-LINE FREQUENCY PERCENT
    Yes……..249……..61.8%; No…….154……..38.2%”

    “(COMPUTE FROM PRECEDING TWO QUESTIONS)
    Land-line only at residence……….6………1.4%; Cell phone only at residence…….150……..37.2%; Both cell phone and land-line……247……..61.4%”

    It seems most of the polling houses address the Cell phone question by figuring out how many people cannot be reached by a land-line phone. Then they call some fraction of of cell phones. It does not appear that they are addressing the other similarly important question, how many people cannot be reached by cell phone. Nor are they dealing with the middle, those who have both.

    Should polling houses be calling more Cell phones than they call land-line phones?

    If: Population constants are (C)= Cell only fraction, (i.e. 0.372 for 37.2%) (L)= LLine fraction; (Bc)= 1-(C), and (Bl)= 1-(L)
    (N)= sample size; (CC)= calls to cell phones; (LC)= calls to LLine phones.

    (CPC)= Cell phone only households contacted; (LPC) LLine phone only households contacted; (BCLC)= both Cell and LLine households contacted.

    Then: (CPC)= (C) * (CC); (LPC)= (L) * (CC); (BCLC)= (Bc) * (CC) + (Bl) * (LC)

    And: (N)= (CPC) + (LPC) + (BCRC), and (N)= (CC) + (LC)

    It’s interesting what happens to (BCRC) as either (CC) or (LC) approaches (N).

    It’s also interesting what happens when (L) approaches zero.

    This approach is messy, and there is no perfect solution for it, a best fit is as good as it gets. However I do think this approach is the correct one to model the population in Alaska. Is it the correct one to model the population in the lower 48?

    In order to interview those 37.2% of people who only have a cell phone fully, should all of the calls be made to Cell phones, and ignore those 1.4% of people who only have a land line?

    What if in the lower 48, in Iowa and Colorado, (L) and with it (LPC) is also approaching zero, ?

    I don’t know anyone who has a land line phone, who does not also have a Cell phone.

    • wendy fleet

      Jay –
      “I don’t know anyone who has a land line phone, who does not also have a Cell phone.”

      Moi . . .
      Call me a Passenger Pigeon I reckon, a bird with pretty feathers about to be extinct . . .

  • Laurie Roberts

    New CO numbers out this morning:
    According to the latest data from the Secretary of State, 905,500 ballots have been returned to the clerks across Colorado’s 64 counties. That total has increased by 250,000 ballots since Monday.

    Registered Republicans have returned more than a third of the ballots so far, accumulating a total of 379,250. Meanwhile, Democrats have returned 294,648 ballots.

    Colorado is purple on the map of registered voters-lots of independents. Our Gun Lobby got really fired up by the mandatory background checks in the reform bill passed after the Aurora movie theater shooting. The Lobby wanted more mental health services and no background checks for person-person sales, but of course the legislature didn’t want to FUND that. This law combined with the massive shortage of and price increase for ammo has really fired up a lot of people. Plus our economy is good, so why not go all Kansas on it? Yikes! On my way to make phone calls now!

    • Jim Bossert

      There has been little mention of gun control in the Colorado campaigns. Udall has come across as not engaged in issues of importance – almost exclusively appealing to single women with abortion rights message. I’m one of the many unaffiliated voters in the state and pay attention to the candidates’ messaging, which in Udall’s case has not been strong.

    • John

      I think something that isn’t being paid attention to but could make a difference is the political firestorm in Jefferson County of school curriculum. As mentioned earlier, the R advantage in JeffCo is below 2010 levels, and I’d hazard to guess that some of that is a more engaged D base that is seeing the effects of not fighting harder in last year’s School Board elections. Jefferson County is Colorado’s fourth largest, so a decent swing there could have an impact on these close races.

  • Glenn Reider

    Reading the 538 article you make reference to, the “trend” is gauged based on their own probability model, which includes fundamentals. Therefore, a static poll situation would appear as a trend, too.

    My question for the past is whether the switch to likely voter screens was the main factor that caused the polls to move 2% to R in September. The modeling for these screens was possibly overly influenced by 2010, I believe.

    My question now is whether the emerging trend (last few elections as more and more people drop their landlines) for polls to miss approximately 2% towards R will not create big surprises this time next week.

    Oddly enough, if control of the senate is known next week this becomes bad news for Nunn and Landrieu if they are involved in runoffs.

  • Olav Grinde

    It shocks me that Colorado Republicans have a 10 % advantage in early voting. Are Democratic GOTV efforts failing in Udall’s home state?

    Does anyone know whether independent Coloradans tend liberal or conservative?

    More than 25 % of early voters in Colorado have no party registration / other party.

    • Phil

      It’s the mail ballot effect. This isn’t early voting like it exists in most states, this is everyone in the state having a mail ballot in their hands early–so the comparison to make is not to Colorado in 2010 but to Washington, which has a similar system. Most early voters didn’t actually go to the polls and vote, they simply either mailed or dropped a ballot off.

      The Democrats are saying that the early lead here is simply election day voters shifting to early voters. The number to watch prior to election day is the comparison to 2010 total ballots casts. A lot of the pushback on the “Colorado polls are wrong” theory has centered around the idea that Bennett’s 2010 campaign already maxed out the off-cycle D turnout in this state. If pre-election day totals approach the 2010 total vote (which incidentally is what is being tracked at Elect Project, not the 2010 early vote), then that indicates the Democrats are correct that the all mail ballot system is going to boost participation in the election.

      It is also unknown which party the election day voters will favor–will it be like it used to be in Iowa, where D’s use early voting and R’s vote on election day? Or will so many of the R’s have shifted to early voting that the D’s have the advantage. I think the latter is more likely to be correct but we won’t know until Tuesday night.

    • securecare

      Yes we Washingtonians are voting by mail and have been doing so for a few cycles; any comparison of Colorado behavior this cycle might be compared to our past behavior (or not).

      IMO the “system” is on borders between “attractor basins” or on a “knife edge” in more common language.

    • Davey

      Olav…I think Colorado is a real wild card, which is why I love it this season. This is their first election with mail ballots, so there’s no way to predict what the new turnout will be. Thus far (as reported by Michael McDonald at HuffPost) returned ballots as of 10/25 equal half of the total votes cast in 2010…900k.

      Add to this that it’s legal for the parties to canvass and pick up ballots from voters. Democrats say they can knock on 150,000 doors a week…as we enter the home stretch we would assume both parties will have canvass volunteers offer to take the ballot to a drop point. When canvassing California I was able to drive about 20 voters (elderly) to the polls on Election Day…how many can a volunteer pick up, asking the voter to fill it out while they wait?

      If the race were ten points apart, this might not all make things seem uncertain. But 1% of the 2010 Colorado vote was 16,000 votes. If either party is truly speaking to 150,000 voters this week and launching a massive Election Day GOTV effort, it’s easy to see how the new laws could result in a wider victory than expected for either candidate.

    • Sod

      I live in Boulder County (super liberal land) and even here I am seeing as many Gardner signs as Udall signs… They (the conservatives) certainly seem more motivated than the liberals. The weirdest part is that we’ve been getting 1-3 letters a day about our state representative race (Tinlin vs. Primavera)!
      Independents in Boulder County are generally liberal, but I am right on the edge of Weld County, and independents in Weld County are more the Libertarian type, hunters, ranchers and gun lovers in general. They would mostly hold their noses and vote Republican if motivated. Denver independents would generally be liberal but the southern, affluent suburbs are Republican and if you go to Colorado Springs, you might as well be entering the Bible Belt.
      I’m hoping the polls are wrong with this whole mail-in ballot thing. Sadly, I have a friend, maybe 26, who is on the fence. He’s so intelligent, but somehow still can’t parse out the crap from the real issues! I don’t understand how the death penalty became the #1 issue in the governor’s race and how Gardner can somehow be perceived as moderate when he was loudly throwing Tea into the harbor in 2010 and 2012.

    • Franklin

      Here is another consideration: Back in 2010, the Tea Party election year, most polls had R ahead for most of that year. EV predicted D-46/R-48 for CO on the eve of the election. But the result was D-48/R-46. The real result was completely flipped from the prediction.

      As of 10/31, EV has D-44/R-48.

  • Alan Koczela

    Thanks for another good post.

    I would rather playing the R hand, too. I know we are only suppose to look at polls; but, looking at the betting odds suggests the Ds are the underdogs. Folks who bet are putting money where their preferences are, so you can’t discount ‘em out of hand.

    Looking at Paddy Power, the current betting odds for R taking over outright on Nov. 4th (i.e., no runoff stuff) is 1/8. The odds for Ds retaining the Senate are 9/2. Please excuse any math mistakes, but, I believe these odds translate into R at 88.9% and D at 18.2%. Since the house claims 5%, Paddy Power implicitly assumes R at 84.5% (i.e., 88.95 * 0.95) and D at 17.3%. This is not a comfort if you’re a D.

    BTW, Paddy Power lists three individual races: CO, NH and NC. Again, please excuse any math mistakes, however, the mistakes should not distract from the larger point made by the odds. The odds and percentages are listed below with the house odds in parentheses:

    CO
    Udall: 3/1, 25% (23.8%)
    Gardner: 1/5, 83.3% (79.1%)

    NH
    Shaheen: 1/2, 50% (47.5%)
    Brown: 6/4, 40% (38.0%)

    NC
    Hagan: 4/6, 60% (57.0%)
    Tillis: 11/10, 47.6% (45.2%)

    Again, a mixed bag for Ds, but CO looks like it’s in serious trouble.

    • Sam Wang

      Those agree with poll-based estimates – as of course they would. These bettors have all the same information that PEC readers do. They would have presumably bet against Reid and Bennet in 2010.

    • securecare

      In any betting system there is a self selection issue since one, “in general”, must have some financial comfort to be willing to risk monetary resources which would tilt the “sample” against the poorer demographics.

      Of course some poorer folks would be willing to bet, similar to any other gambling situation etc.

    • Andy

      This is poor logic, when you consider that a) in political betting as in sports betting, people love to bet favorites, and b) the public is seldom right. It is well-known that upsets have a far greater likelihood of occurring when 80% or more of the public is betting on the favorite…I remember well when Obama in late 2007 was listed as a 1,350 to one underdog against Hillary to get the Democratic nomination…the only bet I regret having not taken.

  • Gregory Primosch

    I find the use of the term momentum to be problematic. It implies a gradually increasing or decreasing trend. This is NOT what happened here. There was a discrete shift at the end of of September (just like there was a discrete shift at the beginning of September, end of July, etc). The polls shifted to be sure, but once the shift is over they stabilized, and no further trend can be discerned.

    So, why not discard “momentum” and use something liked “recent advantage”?

    • Amitabh Lath

      If you want to misuse physics terms, how about “impulse”?

    • Olav Grinde

      “Momentum” is a PR term.
      It has nothing to do with physical or political reality. ;)

    • SFBay

      Maybe “realignment” is a better description. There was a shift but now it’s stabilized.

    • securecare

      One versed in chaotic systems terminology might use the phrase “basin shift” and/or “attractor shift” which would probably confuse many even more. But these shifts do appear to be “non-linear”.

    • GEinCT

      I think the KC Royals have “momentum” going into tonight’s WS Game 7. However, you know what they say about momentum – “(it’s) the next day’s starting pitcher”.

  • Franklin

    Look at the early voting results for LA. Dems are surpassing Reps 53% to 34%, and so far the early vote total in that state is approaching double that of 2010.

    I have to say that at this point — with less than a week to go, it’s the early voting totals, and comparing them to the 2010 early voting results, that may give us a better prediction of the outcome. The polls seem to be “stuck” at this point and not reliable to divine out what will happen for either party.

    • Olav Grinde

      Surely that feedback from those who have already voted is better than any poll!

      If Braley really does enjoy a 21 % lead amongst the 349,000 Iowans who have already voted, and not just Quinnipiac’s sample, then Iowa is looking Blue.

      349k is equivalent to 30.8 % of all 2010 ballots. Ernst would have to get a very hefty share of the ballots yet-to-be-cast to make up the difference.

    • Edward G. Talbot

      Yeah, I agree that early voting number is puzzling. If Braley really does have a 21% lead among the early electorate (which has about equal number D and R voters right now), it would probably take huge turnout for Ernst to win. In the Loras survey that came out yesterday, Braley had an 8 point lead among early voters.

    • Matthew

      Colorado is looking “bad” for Democrats. Colorado is still 300K voters behind 2010′s total and Republicans have widened their lead from 6.1 points to 9.4 points. In Iowa the early vote totals are about equal with last year and Republicans have reduced the Dem advantage from 5.7 points to 1.5 points so I’m unsure how to read this.

      You are right about LA, almost double the vote from 2010 and a much higher percentage are black however North Carolina doesn’t look so great. Early voting is down 40% but the racial composition is more favorable to Dems. But with early voting being down so much and Reps having a huge advantage on election day if I were Hagan I would be worried. In Georgia, early voting is down but the racial composition does appear to be favorable to Dems. (I am comparing the current numbers reported on electproject.org to the final 2010 numbers reported on electproject.org so the analysis could change.) All in all, and I am not an expert in this, it appears that early voting is mixed, good for Reps in Colorado, good for Dems in LA, mixed in Iowa, NC and GA.

    • Olav Grinde

      Whether Braley’s lead is 21 % or 8 %, that is very significant! Note that roughly the same number of Democrats and Republicans have voted (41.1% vs 39.6%), this must mean that a vast majority of independents (19.1%) are breaking for Braley.

      Either that, or both polls are way off!

    • Matthew

      Olav,
      If early voting totals end up about 31% of the total electorate and Braley is ahead by 21%, Ernst would have to win about 55-56% of the election day voters to win the election.

    • Phil

      Just a correction for Matthew, Elect Project is comparing the 2014 early vote number to the 2010 TOTAL vote number. So right now Colorado has hit 50% of the total votes cast in 2010, with a week of early voting and election day itself to go.

      The R lead is also dropping, not rising. It started out over 10 and is now at 9. Nate Cohn, who has access to the voter file, also noted on October 26th that 43% of the submitted votes so far were from people over 65.

    • we_are_toast

      That segment of the Iowa Quinnipiac poll looks like a complete mistake. Braley would have to be carrying 80+% of Independents, which seems silly.

      I looked at CO. county by county, and believe CO may have the highest turnout in the nation. The one thing that sticks out is that if you extrapolate the current early votes, the R’s will exceed their entire active R registration. This looks like R’s have done a fabulous job at getting their people to vote early, but they are most certainly cannibalizing votes from election day and their numbers are already dropping off. 1% in the last 2 days. If R’s start losing ground at an increasing rate, might mean they haven’t kept many of their 2012 voters and D’s will pull it out.

      Late new registrations (last month) seem to be favoring D’s, and at current pace, early vote for bellwether Jefferson county looking to come in under that 6.1% R advantage from 2010.

    • Insidious Pall

      @Olav Grinde At the end of the day yesterday, Dems led in returned ballots by a point and a half, 41 – 39.5%. You can track daily returned ballots at the Iowa SoS here:

      http://sos.iowa.gov/elections/pdf/2014/general/absenteestats.pdf

    • Davey

      Matthew – disagree that Colorado is looking bad. I also won’t say it looks good for Dems. A total change in election law, primarily shifting to mail ballots, should utterly change early voting (and possibly all voting) patterns. We have no baseline at all to judge what’s coming in.

  • Dan Ferrisi

    Sam,

    Thanks for the thoughtful analysis. In a time when water-carrying pundits and partisan shills pollute the airwaves and the ‘net, it’s refreshing to see people approach elections from a genuinely science-based position.

    Since your model is, at this point, substantially more bullish than the other forecasting models (as of 8am this morning, pretty much dead even), I’m wondering in whose shoes you’d rather be at this point: the Republicans or the Democrats. Since you’re model gives the GOP a 0.0-point Election Day advantage, would you have no preference relative to whose cards to be playing?

    • Sam Wang

      By a tiny margin, 0.1% effectively today (and somewhat larger over the last week), it would be better to be a Republican based on polls. But we are talking half a percentage point or less, and as I have said, polling errors at a national level are routinely far larger.

      There is one hidden advantage for the GOP not captured in my calculation: the Georgia race, where a runoff seems likely. This should have lower turnout, and therefore favor Republicans. Then again, if the Georgia race is the single race that determines control of the Senate, that could be very motivating for both sides.

    • Edward G. Talbot

      One other hidden advantage for the GOP is Orman, right? The model is assuming Orman caucuses with the Dems?

    • Sam Wang

      That’s an advantage for the seat count because 51-49, a case where the GOP is clearly in the majority, turns into 52-48 if Orman does what he says he’ll do, i.e. caucus with the majority.

      For overall win probability…both sides are acting as if a 50R-50D/I split will lead Orman to caucus with the Democrats. If you disagree with both major political parties…you can correct for that yourself by subtracting it from the histogram.

    • Edward G. Talbot

      Good point – Georgia is a definite hidden advantage while you are correct that Orman using a deciding vote to caucus with the republicans is speculative at best given how both parties are acting.

  • Canadian fan

    It is fascinating how everything seems to be getting back to Iowa and Colorado, for many the perennial center of the campaign. Indeed – as Sam points out – there is much that recent polls are revealing about early voting, and much that they are not. The big unanswered question is – how successful have Democrats been in their long-stated goal, of targeting voters who did not vote in 2010 ? The majority of the senate will hinge on the answer to this question.

    • Bert

      I think Iowa and Colorado are really important in that they show whether the national electorate is shifting. These are states that have been key to the national electoral advantage for Democrats. If Republicans pick off Senate seats here, then it really will give the GOP hope for putting the electoral college back into play for conservatives in 2016. If Democrats pull it out in Iowa and Colorado this election, then it’s clear that the national voting demographic still favors Democrats.

      Of course what would be interesting is if Democrats lost in Iowa and Colorado but replaced those losses with 2 wins in Georgia, Louisiana, Arkansas, or Alaska. That’s the kind of scenario that would signal a sort of chaos or instability in the electorate.

    • Davey

      Bert – it’s never too early to talk electoral college (which because of mystical qualities is far more interesting than Senate elections.) As a history buff, I’m fascinated that 2014 looks similar to 1914, when one of the political parties centered a majority of their power in a Southern strategy that offered good representation prospects in Congress, but a distinct disadvantage in the electoral college. In the first half of the 20th century, it was the Democrats, and now we see something very similar with the Republicans. And the math mirrors quite strikingly (the Washington Post has some great articles back in April). In the early 20th, Democrats only had a 25% chance of an electoral victory with 52% of the popular vote. Today republicans face about the same odds.

      As we head toward 2016, it would be great to see more analysis of this effect, and how we arrived here despite shifting electoral power in the states.

    • Matthew

      Davey,
      And yet the Democrats won in 1912 and 1916. I know 1912 was the year TR ran on a third party but I suspect WW would have won anyways.