“Public sentiment is everything. With public sentiment nothing can fail. Without it nothing can succeed. He who molds opinion is greater than he who enacts laws.” – Abraham Lincoln
With the midterm elections approaching on Tuesday, it seemed time to weigh in with some predictions.
Before launching out on any limbs, let me say that I fully embrace the Abraham Lincoln quote above. It is, at present, more important to move public sentiment in the right direction, than to have this or that political party in power. (This is not always true as a matter of course. Surely Germany would have been better off in the 1930s had they staved off the rise of the Nazis – even if public opinion was not yet molded into an egalitarian view of the world. But for now, in America, I certainly hold it true.)
I like to dig through what little numbers and prediction sources are available to the public. I have a sense that the political insiders have access to much better data and even models. Even still, there is some data available to the public, and enough public prognosticators willing to apply their modeling capabilities to the problem. The sources we’ll pull on today include Nate Silver at fivethirtyeight.com, Real Clear Politics, and the Intrade prediction markets. The sources are all significantly different. Fivethirtyeight is a gargantuan modeling effort by Silver, and takes a pretty comprehensive view of outcomes, polls, national sentiment, etc. RCP is a great place to aggregate information, but doesn’t apply much in the way of modeling. Their “predictions” (if you can call it such) merely consist of averaging all available public polls – not scientific at all. Finally, intrade allows people to make bets on what they think the outcome will be.
The House of Representatives
Let’s start with the House. The current makeup of the House is 255 Democrats, 178 Republicans, and 2 vacancies – 435 total. A party needs 218 for control.
The last time I looked (this morning), fivethirtyeight had a Republican distribution of outcomes that was approximately normal (Gaussian, for the mathematicians in the audience) with a mean of 233 seats and a standard deviation of 15 seats. That’s an average gain of 55 seats – enough for control.
The intrade standings imply a slightly better situation for Republicans, with the “50-50” outcome being around 237 Republican seats and a slightly lower implied standard deviation.
RCP showed 163 safe (R), 16 likely (R), 42 lean (R), 43 tossup, 26 lean (D), 24 likely (D), and 121 safe (D). If we assume 100% chance of winning a safe seat, 95% chance of winning a likely seat, 80% chance of winning a “lean” seat, and 50% chance of winning a tossup, the average Republican outcome is 240 seats. The standard deviation of the outcomes depends heavily on what we assume for correlation of outcomes. (Clearly, the outcomes will be positively correlated. Republican wins in lean or likely Democrat seats imply a national wave of support – just as would Democrat wins in lean or likely Republican seats.) A correlation coefficient of 0.1 (fairly small if there truly is a “wave”) would give a standard deviation of results similar to that of fivethirtyeight, around 14 seats this time.
The figure below plots all of these results. The x-axis is the number of Republican seats held after the election, and the y-axis showing the percent chance that the Republicans will hold at least that many seats. The orange curve shows the results for fivethirtyeight, the blue for Intrade, the black for RCP with no correlation, and the gray for RCP with correlation of 0.1.
As a fundamental, this is not a good chart for Democrats. The prognosticators in question are predicting between a 55 and 62 seat average pickup for Republicans. Interestingly, using the 0.1 correlation coefficient gives reasonable agreement between RCP and Intrade, each of which indicate a modest chance (20%) of around a 75 seat pickup.
All of this leads to an obvious question to my mind: “what if the RCP numbers are reasonable, but the correlation is actually higher?” If we assume a 0.25 correlation we then get a 25% chance of a 77 seat pickup (that number is significant because the Republicans would then have the same number of seats, 255, that the Democrats currently hold in overwhelming majority).
So, what do I make of all of this? Well, first, I believe we are dealing with a highly volatile situation. Trying to predict outcomes this far away from the current state is a difficult prospect. I also believe in a high degree of outcome correlation – this is a wave election. That means I will not be surprised to see an 80 seat pickup for the Republicans – nor will I be surprised to see a 38 seat pickup, leaving the Democrats in control. My gut tells me that it will be closer to the former than the latter, and I’d look for about a 70 seat pickup.
More interesting will be to see the outcome of several “important” races. Word on the street is that Barney Frank, Steny Hoyer, and a host of other high-placed Democrat leaders are in trouble. I would not expect any of them to lose, but in a wave, or surge, or tsunami election you never can tell.
So, my predictions for the House:  72 seat pickup for the Republicans, giving them an even 250 seats in the next House. (It’s as good a number as any, right?)  Outcomes in the tails of the distribution, whether less than 30 seat pickup, or more than 85 or 90, will result in immediate calls for inquiry into vote fraud. Don’t be surprised to see this … and don’t be surprised to find that there may well be voter fraud out there (there always is, but who really knows how much?).
The Senate is a little easier to discuss, simply for the limited number of races that are even in question. The figure below offers the same curves as before, but for the Senate.
Both fivethirtyeight and Intrade put the median outcome at 49 (R) – 51 (D), with dramatic drop offs in either direction as we deviate from that split. But, my sense is that the Senate, as with the House, is more volatile than expected. Moving to 45 (R) – 55 (D) or 52(R) – 48(D) would not shock me. (Well, the last one would be quite surprising – implying the Republicans win every single RCP tossup.) Ever the optimist, I have a hope that the country is responding quite negatively to the Reid-Pelosi-Obama power grab.
My predictions:  51(R) – 49(D). The Republicans will win PA and CO easily, and IL in a closer affair. Also, Reid will go down in Nevada by a stunning margin (maybe 7 or 8 points). While my gut tells me that the other three tossups, CA, WV, and WA will go 2-1 Democrat, leaving us with a 50-50 Senate, it is no longer the case in American politics that the House changes hands when the Senate does not. So, I will not mess with the streak and assume a 2-1 Republican win in the three tossups.  Looking ahead, the Republicans will control the Senate in 2012 and 2014 having an unimaginable upper hand (very few seats to defend).  Calls of voter fraud are already in the can – regardless of the outcomes.
A couple of historical notes. We have seen three state-wide elections since the 2008 general election: Virginia and New Jersey governors, and Massachusetts Senate. All of those saw massive shifts from 2008. In Virginia, Obama won in 2008 by 6 points; McDonnell won by 19 (a shift of 25 points). In New Jersey, Obama won in 2008 by 15; Christie won by nearly 4 (a shift of 19 points). In Mass, Obama won by 26; Brown won by 5 (a 31 point shift). These are dramatic shifts in spread.
More interesting, the polls in the governor races were off. McDonnel was up 13 in the RCP polling, but won by 19. Christie was up 1 in RCP, but won by 4. By the time the Mass senate race came about, the polls appear to have caught up in their turnout models. Who has the right turnout model this time? I don’t know, but I suspect it will be a volatile shift. The Tea Party is more organized now, a plus for conservatives. However, one cannot expect turnout amongst Democrats to be as bad as it was in the special elections – when they were still floating in a euphoric high over the election of Obama. As a net, I’d suspect that is advantage Democrat.
But, I also think the undecideds are breaking against one-party rule, which is why I have given an edge to Republican over the prediction sources we’ve used. Come Wednesday morning, I suppose we’ll know …