Beta test: Senate snapshot 12 July 2014

July 12, 2014 by Sam Wang

Greetings, everyone. This is for hardcore readers. I’m going to dispense with bells and whistles. We’re building things, so I’m not very chatty! I just thought I’d show you where things are at. Bottom line, Democrats have a 55% chance of control in an election held today. That is as close to a toss-up as it gets. The median result is 50 D/I seats, 50 R seats.

I am treating the Senate race using the same strategy that I used in the 2004/2008/2012 Presidential races. As I wrote yesterday and also for POLITICO, Election Eve Senate polls are fairly predictive, maybe as good as Presidential state polls.

To determine Senate control on a day-by-day basis, I’ll calculate the exact distribution of Senate party seat counts, using the same methods as the Presidential Meta-Analysis, except that each race determines 1 seat, not N electoral votes. Vice-President Biden will break ties to support Democrats, so Democrats need 50 or more votes to control the chamber. Republicans need 51 or more votes.

As of today, the most recent 3 independent polls (or last 6 weeks, whichever is more polls) gives the following snapshot.

That is about as close as it gets. Note that I would score any probability in the 20-80% range as a knife-edge situation, too close to call. As far as I can tell, we’ve been in the 20-80% range for months. In coming weeks and until Election Day, we will give a time-series plot, updated daily.

Here is where key races stand today.

State median+/-SEMD win%
Alaska(tie) +0.0/-3.4%50%
ArkansasCotton +5.5/-2.0%<0.5%
ColoradoUdall +1.0+/-2.1%68%
Georgia(tie) +0.0+/-4.1%50%
IowaErnst +3.0+/-3.1%17%
KentuckyMcConnell +1.5+/-3.8%35%
Louisiana(tie) +0.0+/-3.4%50%
MichiganBraley +5.0+/-2.3%>99.5%
North CarolinaHagan +5.0+/-2.3%>99.5%

In other races: Safe D, HI, MN, NH, OR, VA. Safe R, KS, MS, MT, SD, WV.

Finally, a few notes on what’s happening.

Arkansas. Right now, Cotton (R) is ahead of Senator Pryor (D). However, all recent data come from Republican pollsters. That bears watching.

North Carolina. Senator Hagan (D) leads Tillis (R). She’s been helped by the entry into the race of Sean Haugh (Libertarian), who has drawn net support away from Tillis. This race is likely to narrow in the home stretch as third-party support fades.

Obama approval/disapproval. In 2012, the Obama/Romney EV estimator went up and down in close correlation with the Senate control probability. I suspect that a slight decline in Senate Democrats’ chances has paralleled a slight slip in President Obama’s approval/disapproval numbers. In the coming weeks and months, I will examine this question.

This calculation used the MATLAB scripts Senate_estimator.m and Senate_median.m, and the data file polls.median.2014Senate.txt.


Tony Roberts says:

Sam, very briefly, are you able to offer any insights as to the Democrats chances of gaining seats in the House? I have no illusions that they will flip the House.

Jpell says:

Dr. Wang,
Another race that may require a disclaimer is Iowa. I don’t find the Ernst lead convincing-if you look at the polling results on Huffpo that you’ve linked, literally every poll that Ernst leads in (3 of them) came out the same exact day! These three polls consist of polling taking entirely in the two days after her primary win. Since every poll before these and the one poll since have Braley in the lead, it bears watching whether that was just a brief post-primary win bump.
Thanks, Really enjoy the site

Sam Wang says:

Right now is when the overall analysis is most sensitive to small differences. Let’s imagine hypothetically that the Republican pollsters really are skewed.
If I take all the Republican pollsters currently used in my analysis and shift the margin 3 points toward the Democratic candidate, the Democratic control probability rises a tiny amount, to 58%. It’s equivalent to a national swing of 0.12% in opinion.
In my view, such an adjustment, even if we had a data-based approach to justifying it, is not worth the increased uncertainty. Nor is it worth the loss of credibility that comes from “adjusting” what we think the polls ought to be saying.

Sam Wang says:

Maybe. But other states fluctuate in the other direction. It is a risky practice to only second-guess the measurements you don’t like. Don’t look at the sausage too closely!

Amitabh Lath says:

Just to play devil’s advocate: There are several ways to interpret the plot of poll margins vs. final results from your previous post.
One is to ascribe it to a non-linearity that manifests away from zero (ie, pollsters underestimate blowouts), as you do in the discussion.
However, notice that a straight line of slope 1, but with intercept shifted by about 2 points Democratic also fits the points pretty well. In other words, a fixed pollster bias towards Republicans.
So if I go ahead and put a +2D on each of the means there (and what the heck, add 2% in quadrature to all the uncertainties) :
Alaska: Beigch +2 +- 3.9
Arkansas: Cotton +3.5 +- 2.9
Georgia: Nunn +2 +- 4.6
Iowa: Ernst +1.0 +- 3.7
Kentucky: Grimes +0.5 +- 4.3
Louisiana: Landrieu +2.0 +- 3.9
The uncertainties get a little larger so only Cotton in Arkansas is >1 sigma.
I know, you did something similar in 2004 with Kerry and got burned, but whatever this systematic is, it is worth looking into. It does have an effect in close races and there seem to be several of them this time around.

Amitabh Lath says:

I just realized that even if the mean moves by +2, the median may not. I suppose to estimate the shift in the median you might look at the affiliation of the pollsters. If a state has similar number of R and D pollsters, the median is probably fine.

bks says:

In 2010 there were about 260,000 ballots counted in Alaska. A hundred votes there could swing the entire Senate. –bks

Bob Grundfest says:

Dr. Wang:
I’m happy to see that you’re analyzing the Senate races and am particularly interested in Kentucky. Grimes looks to be a good position so far. Looking forward to further updates. Did you happen to see the NYTimes Upshot article on the pollsters?

Sam Wang says:

If it was the July 10 piece, I replied to it that day. It was interesting but a bit overblown. The polls are fine and a few Republican pollsters don’t have much effect on the overall picture.

Leave a Reply

Your email address will not be published. Required fields are marked *