Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Senate horserace update: what does a Meta-Margin of R+0.4% mean?

October 23rd, 2014, 6:00pm by Sam Wang

Update, Saturday: with two new surveys in Alaska, one showing a tie and the other showing a Begich (D) lead, the predicted Senate Meta-Margin is D+0.0%, a perfect tossup. Statistically this is not different from the minuscule difference yesterday…but it does emphasize how closely fought the battle for control is. To see more, click on the seat graph at the right or the Meta-Margin graph.

Now that Greg Orman (I-KS) has a median lead of 1.5%, the Senate Meta-Margin has gone to R+0.4%. For those who are new, let me explain what this means.

The Meta-Margin is defined as “how much swing would it take to create a perfect toss-up for control.” In other words, if polls swung by a scant 0.4% across the board, then a 50R, 50D/I split would become about as equally likely as a 51R, 49D/I split. Recall that across-the-board midterm polling errors (i.e. in the Meta-Margin) are typically 2-3%, and are five times as large as presidential-year polling errors. I would say we do not know who will control the Senate!

Here is another way to think about it: Democrats and Independents are reasonably assured of winning 45 seats. A relatively likely outcome (the mode) is to add Hagan (NC), Shaheen (NH), Orman (KS), and Nunn (GA), to make 49. At this point, if Republicans lose Alaska, Arkansas, Colorado, Iowa, or Kentucky, then they will be left with 50 seats. Of course, with 10 races having margins within 4%, this is just one of 2^10=1024 likely combinations. The Princeton Election Consortium’s meta-analysis considers them all.

Tags: 2014 Election · Senate

67 Comments so far ↓

  • Cal State Disneyland

    While you indicate Nunn in the above list as a “relatively likely outcome”, the reality is that she may not reach 50% of the vote in a couple of weeks and then lose in January.

    I am also assuming that the 45 Democratic seats do not include Louisiana.

    Therefore, I am more of the mind that the Republicans would have to lose two of AL, AK, CO, IA or KY in order to drop down to 50 seats.

    • Stu

      Part of this may be that partisan pollsters who have had two fingers on the scale need to be more realistic as election day gets closer, otherwise they end up with egg on their face. So, as you get closer to “final” predictions there may be less shenanigans with the weights used to adjust the raw results. Several weeks ago there were a heavy batch of polls showing bigger leads for the GOP and it is difficult to tell if these house effects are shifting as the election gets closer or whether these newer results represent a “real” shift back toward Dems.

    • Louis Hirsch

      the most recent poll in Georgia recognizes the necessity of a runoff and has polled that with Nunn ahead 51-47.

    • Edward G. Talbot

      Lou – any links to that poll? I’d be curious how they’re determining likely voters accurately. It’s hard enough to determine likely voters before the general.

  • Edward G. Talbot

    That’s as close as it gets! Am I wrong or does Nunn +1.5 not really take into account the potential runoff situation? It seems like it’s not possible to do so with any real objectivity given the fact that no one is polling a runoff (and many polls aren’t even polling a two way race at all). Until the first election is done, determining likely voters for a runoff would be suspect anyway.

    • Sam Wang

      No allowance is made for the runoff – we just use the two-candidate margin. I have not thought of a better way to address this issue.

    • Olav Grinde

      Meanwhile, another good poll for Nunn today, CNN/Opinion Research placing her at +3.0 %.

  • Joe

    I’ve noticed for the last few days it’s been creeping back downward in the Dems direction. Is this the model coming to the conclusion we all know, which is control is a coin flip at this point, or is it possible with 12 days to go, we are seeing a national shift towards Dems in all of these races?

    Also, the GA question is a good one. None of the polls, except for PPP that I have noticed have asked a follow up “if a runoff were held…” question of voter preference. Does the model account for that?

    • Edward G. Talbot

      Most likely explanation is not a shift, just that it was always close and it doesn’t take much to make it even closer. GA and CO alone have moved enough to make a noticeable difference.

    • Matthew

      Unless I’m missing something the only race that has moved towards Democrats is Georgia. Of course that is a big move because Republicans were counting on holding all their seats and a loss of a Republican seat requires them to pick up two Democrat seats, one to maintain their current seat count and one to expand their seat count.

  • A New Jersey Farmer

    None of Ernst, Cotton or Gardner are polling at 50%, which at this point tells you that these races could go either way. Cotton is in the worst shape because he hasn’t polled above 45% in more than a month.

    • Sam Wang

      I had not noticed the high undecided numbers in Arkansas. That is a good catch.

    • Matthew

      Pryor’s RCP average is 41.3% and Udall’s is 43.0%. Pryor is behind by 5.5 percentage points and Udall is behind by 4.0 percentage points, according to RCP averages. They both have to win 70%+ of the undecided in their respective states to win, assuming the polls are accurate. Do you really think an incumbent of the same party as a very unpopular president can do that? Even if the president weren’t unpopular do you think an incumbent can do that?

      In Iowa, again using RCP numbers, Braley is down by 2.5 percentage points and at 45.0%. He would need to win 2/3 of the undecided, assuming the polls are accurate. He isn’t the incumbent and the race is closer than in Colorado or Arkansas but the fundamentals are essentially the same, he needs to win 2/3 of the undecided voters when he is in the same party as a very unpopular president. It certainly could happen but doesn’t seem very likely.

    • Matthew

      Also in Arkansas, according to RCP, there’s only been two polls since September that had Cotton below 45. Of the ten polls listed on RCP which have ending dates in September, two had Cotton below 45, three had Cotton right at 45 and five have Cotton above 45. Three of the four polls listed on RCP which were released in October had Cotton above 45 and the other one had Cotton at 45. I don’t know where you got your information about Cotton polling below 45 but it’s wrong.

    • A New Jersey Farmer

      Matthew: Thanks for catching my glaring mistake. I meant to say that Pryor hasn’t crossed 45% in a month. That a big error on my part, and my point was similar to yours. Pryor has his work cut out for him but an opponent polling under 50% does not have his race sewn up.

  • Sean Patrick Santos

    “In other words, if polls swung by a scant 0.4% across the board, then a 50R, 50D/I split would become about as equally likely as a 51R, 49D/I split.”

    Maybe I’m picking nits, but having the Republicans win 50 seats is of course not the same as the point where the probability of 50 seats is equal to the probability of 51 seats. Admittedly, for a very symmetric distribution, this distinction doesn’t make much difference.

    • Sam Wang

      That is true. It is why I used the word “about.”

    • Charles Stanton

      Not sure if this is the same side of the coin or not, but could it also be said that changing the makeup of the electorate by .4%, aggressive Democratic GOTV, would also affect the outcome? Or would that also be lumped in with mid-term polling error of the likely voter model?

  • 538 Refugee

    PEC is once again “bluer” than 538 and all is right with the world. Are we seeing a true shift or are pollsters with an agenda slowly adjusting so they can point to their final number as the ‘one’ that they will hang their ‘accuracy hat’ on? You can only validate polls close to the election. Who is to say that early polls that looked like outliers were wrong and not a true reflection at that moment in time? For what it’s worth, nobodies right if everybody’s wrong. ;)

    • Joseph

      IMHO, the difference is in the decision that Sam made to continue using a much longer time span for generating the Meta-Margin than 538 uses. Will the Meta-Margin now swing back to the mean of a few weeks back? If so, it would seem to justify the longer time span approach. If not, it would seem to justify recalibrating when you get within a month or so of the actual election. Time will tell. This will be a very interesting two weeks….

  • Amitabh Lath

    The drift back towards the Democrats seems consistent with a random walk. The steep drop in the margin during the 3rd week of September seems even more strange now. Usually such large jumps are caused by some event like the first Obama-Romney debate, or the 47% video. Any idea what the late-September surprise could have been?

    • pechmerle

      Amitabh, I still think the steep drop in late September is related to the appearance of the ISIS beheading videos. Of course, the pres. has very slight influence on the current situation in the Middle East (and none on the degree of brutality ISIS chooses to engage in), but I think the public as a general proposition does hold the president accountable for the status of foreign affairs. And those have deteriorated — at least in perception — markedly.

    • Sam Wang

      Someone also suggested Obama’s September 6th speech on immigration, which could have cooled a few Democratic voters. That’s all it would take.

  • Olav Grinde

    Sam, I could have sworn that last night I very briefly saw a new PEC post appear, entitled something like: “How to make a ripple into a wave election”. Seconds later it was gone…

  • Canadian fan

    What an indescribable joy to see the meta-margin now at 0.3 ! As New Jersey Farmer illustrates, Arkansas’ undecideds are unusually high for this stage in a campaign, and that’s very telling. And the Monmouth University poll in Colorado with Udall trailing by only one point is also telling. As that is a pollster that 538 accords an A – rating , that ought to give reasonable assurance that this race is indeed close. ( Being separated by an average of 2.1 % is another indication, of course). Harry Enten of 538, though, gave a dress-down of Democrats’ expectations for Colorado one by one. He’s a very good writer, and poses his arguments well, but one argument seemed especially dismissive to me. If Democrats did indeed outperform state polls in both 2010 ( a midterm during a GOP wave no less ) and 2012 by 2.7 % and 2.9 %, and the Huffpost places this race within 2.1 %, does it not seem more plausible than he gives credit that that gap could easily be filled, especially in a year with such an intense push by Democrats ?

    • Insidious Pall

      I think I recall Mr. Enten addressing Democrats’ outperforming the polls in the 2 previous cycles. His argument is that pollsters have likely adjusted methodologies at least after 2012 with 2 straight results to consider. He makes a fairly compelling case that “Udall is losing.” To that I would add Obama factors. One, there exists a fair amount of Obama fatigue in CO and elsewhere. Two, without Obama on the ticket the African American vote and possibly Latinos may not be as substantial.

    • Matthew

      As I pointed out to Farmer, Arkansas undecided would have to break 70%+ to an incumbent in a year in which the president of the same party is very unpopular. That is using RCP numbers. Pryor’s situation on Huffpost is not quite as perilous but he would still need to win 55% of undecideds. In Colorado Udall would need to win 70% of undecideds using RCP numbers and 65% of undecideds using Huffpost numbers.

    • EmmaAnne

      I have been paying a lot of attention to Colorado (my state) and my view has been that the non-polling factors (which PEC doesn’t adjust for, and shouldn’t) would start disappearing once people were actually voting. I have hoped that these favored Democrats – especially mailing ballots to every voter and the apparent under-polling of Latinos.

      Udall is still not favored to win, but the margin has dropped substantially recently – he is now trailing by only 2 points. So I am feeling cautiously optimistic that previous polling was slightly off in its likely voter estimate, and that now that people are voting, the results are more accurate. We’ll see where the polling goes in the next week or so.

  • A New Jersey Farmer

    Here’s an interesting analysis of statewide elections since the 1980s. The thesis is that Democrats tend to win close races overall, though this is tempered by swings during waves.

    I’m ignorant enough to be dangerous when interpreting the numbers, but my take is that if the races remain this close and the poll leaders are below 50% next week, then we truly will have a 50-50 election in November.

  • Kathy C

    This may just be my “crossing my fingers, hoping,” but there are a lot of Republican governors losing their seats this year at the hands of Ds, Rs, and Is. I can’t help but think that some of them voting, in one or two of these very close states, will tip the scale to the Democrat or the Independent. They are already not happy with Republican leadership in their states. Just a happy thought to get me through the nail-biting!

  • Olav Grinde

    In retrospect, with regards to strong Democratic control of the Senate, it is most unfortunate that the following Senators decided to retire:

    – Max Baucus, Montana
    – Tom Harkin, Iowa
    – Tim Johnson, South Dakota
    – Carl Levin, Michigan
    – Jay Rockefeller, West Virginia

    Montana, South Dakota and West Virginia are lost – perhaps needlessly so – while Braley is struggling to keep Iowa in the Blue camp.

  • Amitabh Lath

    I confess I do not understand the effect of President Obama on the races. Usually, the midterms tend to turn on local effects. But let’s say for some reason this time it’s different, Democratic candidates are somehow closely associated with Obama.

    Ok then. Does this work as a negative since Obama’s unpopularity now brushes off on local candidates? Or is it a positive, since being associated with Obama will draw some fraction of the youth and other subgroups that like Obama but tend not to vote in midterms.

    And how do Likely Voter screens account for the latter, if they consider it at all?

    • margaret simmons

      As I posted the other day, we are working very hard here in Colorado to register folks who are not registered, with emphasis on younger people. My group has registered over 300 new voters. I called nearly 100 of these on Tuesday and most of them had already voted or were ready to vote for the Democrats (mail balloting really helps!). I’m not sure Likely Voter screens account for this effect.

    • Amitabh Lath

      That is amazing. Thank you for your efforts Margaret.

    • Edward G. Talbot

      Actually, my understanding is that the midterms often DO turn on the national picture. 1994, 2006 and 2010 are particularly good examples. It seems more negative than positive or we probably wouldn’t even be discussing the possible/likely loss of the Senate. Whether there is a positive effect on Democratic turnout as well I don’t know – doesn’t seem likely. Which is not to say that various GOTV efforts won’t make the difference in an election as close as this.

    • atothec

      The DSCC does not seem to see the President being as big a liability as the GOP makes him out to be. In fact the DSCC has made some pretty bold statements about their Senate chances.

    • Amitabh Lath

      I was not referring just to the GOTV efforts, but the tendency shown by the punditocracy to label any association with Obama as entirely negative.

      Obama has run in two national elections and each of them was characterized by larger than usual turnout of minority and youth voters. So if a Senate candidate is seen as say, 25% of Obama, then does that mean that there will be 25% of the Obama effect on the electorate in that state?

      The pollsters may well be allowing for this effect, so I am NOT saying that polls need to be unskewed or anything like that.

    • Davey

      Margaret – win or lose, I really look forward to seeing how your mail ballots and GOTV efforts in Colorado affect our thinking on results there. Have polls built in these effects, or are we in for a night of surprises from the Rocky Mointain state? 538 wrote a whiny article on why everything is the same as its always been there, but I have a feeling that GOTV + mail ballots = something different.

    • Matt McIrvin

      The political media seem to be characterizing Obama as much more unpopular than he actually is, according to the polls. He’s several points ahead of where GW Bush was at this point in his administration, and Obama’s job approval seems to be holding more or less steady rather than dropping.

    • Matt McIrvin

      …nor, by the way, has he ever gotten as unpopular as Johnson was when the Vietnam War really began to go south, or remotely as unpopular as Carter got either just before or in the late stages of the Iran hostage crisis, to take two popular comparisons. Of course, we have two years and change to go, but he’s already hung on longer than either of them did.

      (Both Bushes eventually ended up in Carter territory, though for the elder Bush there was kind of a remarkable recovery during his lame-duck period.)

  • Canadian fan

    Yes, Mr. Enten does indeed put forth a theory that pollsters have ” likely “adjusted their methodologies. But on what basis does he actually know that they have, and on what basis does he know they succeeded ? The only way to prove the effectiveness of pollsters’ methodologies – improved or otherwise – are the elections themselves, and as there have been no other national elections since 2010 and 2012, it really seems Mr. Enten has gone out on a theory that has yet no scientific proof. We actually won’t know whether pollsters have improved their methodologies or not until after the elections themselves.

    • Amitabh Lath

      I don’t know. Pollsters tend to be a conservative bunch (small “c”). There is a lot of hot air around every election about how this time we will get out the unlikely voters, and it never really materializes for the midterms. So maybe if they see some actual turnout increase they may factor it in for the 2018 midterms.

  • Lewis Ramsey

    The Georgia Senate race has two polls out today. The CNN polls has Nunn up by three and the AJC poll has Perdue up by two. The AJC poll has women making up less than 50 percent of the electorate. Has there ever been a major state or national election when women made up less than 50 percent of the electorate?

  • Jeremy lehman

    Regardless of whether one’s team colors are red or blue (or green–how is it that green became the color of independents anyway Sam?), one thing I think will become more common is the tactical retirements of senators in all but blood red or ocean blue states.
    Take John McCain and Charles Grassley as two examples. With the duration of their incumbency, all but the most perfect of political storms would probably be insufficient to dislodge them next cycle, even factoring in a projected historic and well run Hillary candidacy for president. But in states that have shown they will elect statewide officials of both parties, can you imagine how much worse it would be for the Republicans to not only have to defend seats with freshman incumbents in purple and indigo states like PA, OH, NH, FL, WI, IL, but now they have open seats in AZ & IA to contend with too?
    Say what you will about Baucus and Rockefeller; they’d have sleepwalked to reelection and we aren’t having this conversation if they’d decided to wait for a presidential electorate to retire. Both parties I think will come to appreciate this dynamic rather quickly–and I doubt for that reason that you’ll see Paul or Rubio stick around long in the primaries; if they pass on defending their seat, they open the door to a decidedly less friendly electorate to their causes choosing their replacement.


    Mr.Wang, Here we go with another mass shooting in a school. With the deep pockets of the NRA contributing to conservative candidates, might this bring some “blowback” in some races? If so, how much this close to the elections?

  • atothec

    New AK poll has Begich 48-39. It’s an Alaskan pollster to boot.

    Also recommend watching DSCC Dir. Cecil give a rundown on Senate races – sounds like there might be some surprises for the GOP on election day

  • Joe

    Has anyone seen this out of Alaska today?

    Sam, any thoughts? Is this an outlier, seems like it, but also Majority Pac released one today showing the race tied.

  • securecare

    Holdin my metaphorical breath I am.

  • Sam Wang

    Poke around the other races. A 10-point spread is not uncommon. I am a bit relieved that they are not herding toward one another.

  • A New Jersey Farmer

    Over the past two weeks, Sullivan’s gone from 50% to 39% and now 44% in the latest Alaska polls, and Begich is at 49% after not breaking 45% for the past six weeks.

    Either some event is causing this movement or the story about Alaska being nigh impossible to poll accurately is, well, accurate.

  • Joe

    2 thoughts on Alaska now that this poll and the majority pac poll have allowed to seep in: One is a scandal involving the state national guard which involves both the Gov and Senate nominees, it’s worth a read:

    I’m not sure if this is having an impact on the numbers yet, I think another poll showing Begich ahead would be useful to see. Also, DSCC boasting a robust ground game in Alaska which according to him is the best there ever has been. Take that with a partisan grain of salt. Honestly, as exciting as the 10 point ahead poll is, and the tied one, we need another poll to say it’s a real trend. I see with it added, the meta-margin is almost at a dead even, first time in a few weeks it’s been that high.

  • Canadian fan

    The joy continues with the meta-margin. We are seeing some odd polls – Alaska, North Carolina, Georgia – that suggest wildly different trends than ones established. They emphasize Sam’s point about not focusing on any particular poll, but certainly make the race interesting. There is the war of internals. Rounds internal has him commanding a 24 point lead. And McConnell’s an 8 point lead. Alaska however, is – as Sam points out – an impossible state to poll. And we honestly don’t know what will become of Begich’s aggressive outreach to isolated Native communities that can only be reached by air. Some of these communities may never have seen a Republican canvasser, and are likely absent from any polling survey. We really don’t know what impact these will have. But what we do know is that they are rich with Democratic voters.

    • Olav Grinde

      The Meta-Margin at 0.1 %? That’s comforting!

      I would talk about Democratic “momentum”, but that word was so horribly abused by the Romney camp in the last election.

      Anyways, it’s an even and highly complicated fight. Truth be told, these days I’m spending more time reading Professor Michael McDonald’s 2014 Early Voting Statistics – actual votes! – than the dancing poll numbers. There is some fascinating stuff in the details:

  • CRM

    The Eskimo wave. Who knew? ;)

  • Olav Grinde

    Alaska, there are actually 2 NEW polls…

    That’s strange. RealClearPolitics doesn’t list either. I thought their poll results were supposed to be complete!?

    Anyways, I’m holding my breath waiting for some Blue movement in Colorado and Iowa. However…

    Colorado: The Democratic ground game leaves much to be desired. Republicans currently have a 43.8–31.7 % lead in early voting, which have reached 28.4 % of the total number of votes in 2010. However, which way will the 24.5 % Independents/other break?

    • Sam Wang

      RealClearPolitics is often incomplete. Do what I do, and click on entries in The Power Of Your Vote. They take you to our data source, HuffPollster. It is more reliable.

    • Olav Grinde

      Thanks, I have noticed that. But only recently did I realize how incomplete RealClearPolitics can be. Wonder why – sloppiness, or do they decide to leave out certain polls?

      No matter. Hereafter I shall rely more on HuffPo.

  • Insidious Pall

    There’s some cool stuff at Professor Mcdonald’s site (referenced in Olav Grinde’s above post). There is a daily accounting of early requested, sent, and returned ballots. In Iowa for example, Republicans have requested slightly fewer early ballots but retain a slight lead in returned ballots. Thus far it represents a significant departure from the 2 previous national elections. The significance in terms of final outcome remains unknown, of course.

  • G_Man

    I think RealClearPolitics throws out partisan polls, only uses the last 3 weeks of polling, and always uses the most recent poll a polling organization has put out. I think they also only use the more common or well known polling organizations i.e. they exclude some non-partisan polling organizations. I’m not sure where I recall this criteria, but I believe it’s roughly what RealClearPolitics uses. Given these rules, they would throw out the two new Alaska polls since one is a partisan poll and the other is not well known.

  • Art Brown

    How about a post on “How to think about an election day probability of 50 +/-15%?” Compared to, say, just 50%, full stop. If this were a case of balls in urns, I could understand an uncertainty on their expectation value, which would decline as more data arrived. But for a one-off event? How do you think about this? Thanks.

Leave a Comment