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 control: Three factors to watch in 2014

July 14th, 2014, 4:22pm by Sam Wang

Here (in beta-test version) is the Senate polling snapshot for this year so far.

I used the same polling-snapshot method as in past years: I calculated a compound probability distribution of all contested races. The graph shows a history, over time, of the probability of Democrats/Independents getting 50 or more votes in an election based on today’s opinion polls. On Election Eve, opinion polls closely track final outcomes. Therefore, consider this a snapshot of Campaign 2014.

A few weeks ago, I wrote about different kinds of Senate models: Type 1 models based on fundamentals, vs. Type 2 models based purely on polls. The graph above is a snapshot of polls, and is pure Type 2. Can it be turned into a prediction? Yes, although this year will be especially difficult because it’s such a tight race – as close as the Kerry v. Bush presidential race in 2004, or Gore v. Bush in 2000.

Why is prediction hard? Because the probability of Democratic control is highly sensitive to swings in national opinion. The change you see above, from 90% in April to 52% today, could be reversed by a tiny swing in national opinion. Right now, six Senate races are close to dead heats: Alaska, Colorado, Georgia, Iowa, Kentucky, and Louisiana. If the Democratic-vs-Republican margin in these races swung by two points in either direction across the board, the probability of a takeover by would go to 90% in either party’s favor.

Where will things go by November? As far as I can tell, three factors are going to matter. Two of them I’ve written about before:

1. Consolidation of GOP support behind their nominee. We’re waiting for GOP primaries in Georgia (July 22) and Alaska (August 19). In addition, Louisiana has a peculiar system in which many candidates appear on the November ballot, with a possible runoff in December. In each case, once Republicans settle on a single nominee, he will start to consolidate support. But by how much?

2. President Obama’s approval rating. In the graph above, do you see that precipitous drop in May? It corresponds to a subtle shift in the President’s approve/disapprove numbers. The similarity is only visible with aggregation:

Swings like this occurred in 2012, when the Senate and Presidency seemed to move together. I’ll analyze the possible relationship between Presidential approval and Senate races in coming weeks.

Finally, we have one factor that really limits the Meta-Analysis for now: an absence of data.

3. Alaska, Alaska, Alaska. The biggest problem with the polling snapshot right now is a near absence of information about the Alaska race. Senator Begich (D) is locked in a close race with the probable (but not definite) nominee, Dan Sullivan (R).In other close states, there’s at least a little June data. The last published Alaska poll that went into today’s snapshot was completed on May 11th (seeĀ HuffPost), before Sullivan made a possibly-significant gaffe, accusing Begich of not bringing home pork-barrel spending – in an advertisement shot on top of a building that Begich got built.

Sullivan is also fighting off fellow Republican Mead Treadwell in the primaries. On the general election ballot, Begich and the eventual GOP nominee may also face competition from Libertarian and Alaska Independence Party nominees, who would probably draw net support away from the Republican nominee.

Normally I don’t pay attention to single polls. But for the reasons I’ve given, I was super-interested in this survey by Ivan Moore. Depending on what it says, it would move the overall Senate-control probability by as much as 10% in either direction. However, I note that even this data point would be hard to interpret. Alaska is hard to poll because of its physical size and rural and indigenous populations.

Other polling nerds like to complain about the absence of good polling data. Alaska is the place where a single well-done survey could give the most information.

Tags: 2014 Election · Senate

22 Comments so far ↓

  • Steven

    I am looking forward to your analysis of the Presidential and Senate relationship. It seems to me, that in presidential year, the two would be highly connected because it is the same people voting in the election. However, this is not the case in off-year elections. I am not sure I would expect movement in President approval to correlate too much with Senate votes. Thus, I am interested to see what you discuss.

  • C.S.Strowbridge

    That’s a huge swing. I’m hoping for a few GOP gaffes to get the numbers to swing back.

    • Sam Wang

      Actually, that is incorrect! The swing is quite small, only a few percentage points in terms of opinion. It will only take the tiniest of movements to drive the probability in either direction.

      I see that my presentation is too clear. This problem came up in 2004. It seems like a good idea to switch back to the units I used in 2008-2012: Senate votes, or Meta-margin.

      Thank you for your reaction – this is very useful.

    • C.S.Strowbridge

      “The swing is quite small, only a few percentage points in terms of opinion.”

      It’s a tiny change in the opinion polls, but a huge swing in the outcome.

      “It seems like a good idea to switch back to the units I used in 2008-2012: Senate votes, or Meta-margin.”

      If small changes in opinions can cause huge swings in outcomes, then it might be better to use a system that prevents people from freaking out over the inevitable noise.

  • Bill

    The rapid change is reminiscent of the rapid change in the presidential profile after the first Obama/Romney debate in 2012. The question is, is the change permanent, or an oscillation about a stable median?

    • Sam Wang

      It’s almost certainly temporary. However, what we don’t know is which way things are headed.

      My current feeling is that this will be closely linked to the Presidential approval number. Midterm elections are generally thought to be a referendum on the incumbent President. Of course, if the House majority makes it be about them (i.e. with a government shutdown), that would change matters. In this respect, House Democrats and the Tea Party types want the same thing, but for different reasons.

  • William Ockham

    Alaska polls are useless. Or maybe I am suffering from confirmation bias. Look at the data and see if polls in Alaska have the same predictive power as other state polls.

  • Jim Hafford

    You recently identified several senate races with the highest contribution leverage. Could it be argued that democratic support of third party Alaskan candidates provides the highest leverage of all?

  • Billy

    I find it absolutely fascinating that the NYT shows several states as being essentially solid R for months but closer to a tossup on both here and places like RCP.

    • Franklin

      The problem is most of the polls they include are Republican sponsored. Eliminate them and you will find that the results tend to favor the Democrats slightly. In fact, it looks to be 52/53 for the Democrats.

      Remember what happened in 2012 with the Republican polls?

    • Sam Wang

      In 2012, at a state level, aggregation got all races correct – without adjusting or dropping polls. For this reason I don’t do it.

      Your claim about the outcome, after adjustment, is not correct. See my calculation in yesterday’s comments. However, adjustments like the +2% tool I offered in 2012 would let you to see for yourself. It might need a setting for adjust-R-pollsters-only and adjust-D-pollsters-only.

  • Tony Roberts

    Professor Wang, can you provide any snapshot assessment of The House of Representatives in terms of which party is likely to gain seats and what that gain might be?

    • Tony Roberts

      I know gerrymandering presents a tremendous obstacle for the Democrats, but I have not seen a generic poll not showing the Democrats with a lead, albeit small ones, in seemingly months.

  • Amitabh Lath

    I don’t understand how Obama approval rating adds information above and beyond what you get in the state level polls.

    • Sam Wang

      No, it’s not that it adds information. Polling data is probably best, since it is the most direct measure that we have.

      I am basically speculating that Presidential approval reflects some deep variable that drives all the other variables. The fact that both turned in late May is very interesting to me. It’s an echo of what happened in 2012. If true, it somewhat contradicts the perception that each Senate race is sui generis. But it does fit with narratives in which there are wave elections, national moods, that kind of thing.

    • Amitabh Lath

      So, hidden variables? Be careful Sam, even Einstein got bit by those.

  • bks

    Senate predictions table from NYTimes. I don’t think this particular compendium has been mentioned here on PEC, yet. If it has, apologies:


  • Bob Grundfest

    It’s interesting, and a bit sad, that the polling universe has become so fractured that data is characterized as either Republican or Democratic.

  • Stuart Levine

    I’m surprised that you didn’t add a fourth factor: “It’s the economy stupid.” (No insult intended.)

    Today we have a report that the GDP, annualized, went up 4%. The ADP employment numbers were good, although not Clinton-good. The weekly unemployment claims are falling and some economic analysts are saying that the unemployment rate will likely fall by about 0.5–0.6% a month. The July numbers, to be issued on Friday, are likely to be good (again, not Clinton-good).

    Finally, the ACA has been a great success.

    So why the bad approval numbers for POTUS which translate to bad Congressional vote numbers for the Democrats?

  • Mr. Universe

    Well, I’m hoping you beat Nate’s projections but that’s because I’d like to see the Republican Party banished from Congress for awhile.

    I do like Nate’s pollster weighting methodology. Anything to get the bias out of it.

Leave a Reply to Jim Hafford (Cancel)