Nate, Karl, and Orca

In the weeks since the election, three separate but related stories have emerged. Unless you work hard to keep your life politics-free, you’ve seen at least one version of each.


The first story is that the polls were basically right, and that predictions based on perceived momentum, gut feelings, and the horse sense of old pros were basically wrong. Nate Silver’s work at the New York Times got the most publicity, but similar analyses by Sam Wang at Princeton and Andrew Tannenbaum (yes, the Minix guy) were also quite accurate. And why wouldn’t they be?

Well, as Silver said in a post shortly before the election, the poll-based predictions could be off if the polls had some sort of systematic bias in them. Because, apart from a bit of poll weighting and regional correlation, the fundamental assumption of all of the Silver-like analyses is that the poll results come from a random, unbiased sample of voters. They treat the polls like those textbook problems in which someone draws a blind sample from a well-mixed urn of blue and red marbles. Had their numbers been way off, it would’ve meant that the urn wasn’t mixed well or that the draws weren’t blind.

Actually, I thought the results were too close to the predictions. Obama’s 332 electoral votes was Silver’s most probable outcome, but he assigned a probability of only 20% to that result. Had number been 303, which was the next most likely Obama total in Silver’s analysis, would that have been considered a failure? It shouldn’t have been.

In fact, Democratic crowing about Silver’s accuracy after the election (“math works!”) was just as misguided—well, OK not quite as misguided—as Republican dismissal of his work before the election. Silver wasn’t really saying “this is who will win,” he was saying “these are the likelihoods of various outcomes.” It just so happened that the most likely outcomes were all in Obama’s favor. That said, when Florida finally went Obama’s way and pushed his electoral count to 332, it must have done wonders for the sales of Silver’s book.

Now, you have to believe that there are Republican-leaning analysts who can read the polls as well as Silver, Wang, and Tannenbaum. Where were they, and why did the Republicans who did make predictions end up looking so stupid? That leads us into the second post-election story.


I happened1 to be watching Fox News when it called Ohio for Obama, and Karl Rove brought the network to a stop by claiming it was premature. He was, he said, in contact with Romney’s people and they had numbers that suggested that Ohio was still in play.

Let’s put aside the fact that Rove’s Crossroads Super PACs are supposed to be independent of the campaigns themselves. Only five members of the Supreme Court even pretend to believe that tissue of lies. Instead, let’s focus on why Rove thought it worthwhile to keep hope alive.

This wasn’t like 2006, when, on the verge of his party getting pummeled in the midterms, Rove famously put on a confident look in the face of terrible polls and said he had “the real numbers.” In that case, the election was still upcoming, and he had to keep up appearances; no one expected him to say the Republicans were in for a shellacking. This time the voting was over. There wasn’t even a theoretical downside to accepting the truth.

Which raises a topic I’ve always wondered about: How much of what educated Republicans say is what they really believe and how much is just sop for “the base”? In some cases, it’s clearly sop. George H.W. Bush didn’t believe in supply-side economics, but he knew that pretending to believe it got votes. And just this past week, we saw Marco Rubio deathly afraid to admit that the Earth is more than several thousand years old.

But those are obvious cases. Two decades after Bush I, Republicans who know the difference between real economics and campaign strategies may well be extinct. And as the GOP center of gravity moves ever closer to the center of the Bible Belt, it becomes more reasonable to believe that the party’s rejection of biology and geology is sincere.

Still, denying certain facts of economics and science is one thing, denying fundamental political realities like polls and demographics is something else. One makes you bad at government, which can be seen as a feature rather than a bug; the other makes you bad at winning elections, which is suicidal. This leads to the third story of the election: the Romney campaign’s apparent belief that they were going to win and that their Orca computer/network/GOTV system was going to overwhelm the Obama campaign.


The failure of Orca and the success of Narwhal have become the established truth of the past few weeks. The tech press, in particular, has been tripping over itself to heap praise on Obama’s cadre of nerdy recruits from internet startups and scorn on the buttoned-down consultants retained by Romney from companies owned by his campaign staff.

It does appear that a lot of both the praise and the scorn are deserved. The extensive pre-election testing by Obama’s team kept Narwhal rolling, while Orca—somehow never put through a dry run before Election Day—ground to a halt. But I’m skeptical of the stark black-and-white differences in how the teams have been portrayed. It serves the popular narrative, but it’s unlikely to be the full truth.

And the full truth is important, at least to the 2016 campaigns. Among the interesting questions raised by Orca and Narwhal are: Can Obama’s success be transferred to the next Democratic candidate? Will Republicans accept that their system was flawed and make the necessary changes? Will the electorate feel as comfortable being data-mined by candidates as it is by grocery stores and Amazon? After all, presidents don’t offer 50¢ off your next purchase.

That’ll be in 2020.

  1. “Happened” hell. I was in full-on schadenfreude mode, hoping to see them crushed and hear the lamentations of their women. 

8 Responses to “Nate, Karl, and Orca”

  1. Lukas says:

    It also seems to me that the first and the last story, as told by most people, are not consistent with each other. If the election results were so exactly predictable, how is it possible that they effectively occurred in significant parts due to a difference in technological prowess that we only found out about after the elections?

    Either Orca and Narwhal had little effect on the eventual outcome, or Silver wouldn’t have been as accurate if both candidates had ended up with working systems.

  2. Paul says:

    I forget where I saw this (The Atlantic or Mother Jones, probably), but it seems that Narwhal is actually only a tiny bit of the Obama “tech arsenal”. More to the point, this battery of tools - whose integration is one of the highlights of the stories dr drang is refering to - were brought to play well in advance of election day, very probably influencing the polls that Nate and Friends were so diligently tallying. On the other hand, Orca was just an Election Day tool - built to make sure republican voters did come out and cast their ballot. In which case, its influence would be zilch on the pre-election polls. In that light, the stories are consistent: Narwhal et al. had an influence on the outcome, which showed in the polls. Orca had no influence on election day …

    In any case, the broader point is still interesting: how does the prediction business affect voter’s opinions? I suspect in the past two elections, Nate and Friends were enough under the radar that their work had no influence on the results. That is probably dead: some people will now probably pay attention to their predictions. Apart from that, I’m not well versed enough in statistical analysis to have an opinion, but I sure will like to hear interesting people talk about it.

    As an added data point: I’m French. In France, it is illegal to publish opinion polls 48 hours before an election, and it is illegal to post estimates before ALL polling stations have closed (8pm election day, usually.) Of course, this doesn’t work really well in the twitter era - the French speaking belgians (immune to the law) where posting predictions starting at noon time ;-) The idea is to not sway voters. It’s so ingrained that I’ve always had trouble imagining people in California taking the time to vote - after all, the predictions from the east coast are already in when they get up! In practical effect, I suspect it really doesn’t matter. But I’d really like to see some numbers backing that up.

  3. Alan says:

    Paul: you might be interested in the 1980 elections, in which Jimmy Carter conceded before California was done voting, potentially affecting down-ticket races on the Pacific coast:

  4. Dr. Drang says:

    I wanted to make a connection between the Orca and Nate sections but couldn’t think of a succinct way to do it. Your comment does it.

    As Paul said in comment 2, Obama’s system was much more than just an Election Day app. It could be that its identification and motivation of voters was reflected in the pre-election polls. Republicans tend to believe that minorities are overrepresented in polls, and they have some historical justification for that, but it didn’t work out that way this time.

    The networks here don’t give projections for a state until the voting in that state has stopped, no matter how much information they have from exit polls. So you’ll see states that aren’t close, like South Carolina and New York, called early in the evening, but not before Californians roll out of bed. This is a gentlemen’s agreement, not a law, and has been in place since the 80s.

  5. Lukas says:

    Ah, that explains it. I thought that Narwhal (like Orca) was mainly a system used to motivate voters to the polls on election day. But now I remember reading that it was also used to canvas people, so the system might have influenced, say, whether people were likely voters, which might have been reflected in the polls.

  6. Michael says:

    I said it to Macdrifter, and now I’m going to say it to you: you are not interesting enough for me to tolerate your political trolling. Expertise in one area does not prove expertise in other areas, and only a self-satisfied technocrat would think it does.

    If you want to talk about politics, then talk about politics. Persuade me. Let me see if your opinions are actually worth anything. This is empty, contemptuous snark.

  7. Paul says:

    Alan, thanks for the link. Instapaper’d for a rainy day.

    Dr Drang, Thanks for the clarification. Surprised the gentleman’s agreement has never been called out in court, given the general legal-trigger-happiness over there.

  8. Clark says:

    In fact, Democratic crowing about Silver’s accuracy after the election (“math works!”) was just as misguided—well, OK not quite as misguided—as Republican dismissal of his work before the election.

    I’m glad I’m not the only one who noticed that. The original anti-Silver R pundits were idiots. The triumphant pundits celebrating Silver were just as bad and ironically using the same logic as the pundits they attacked.

    Not sure what was up with Rove. A meltdown of epic proportions but as you noted 2006 there was some indication of this. He’s not an idiot by any means but how he managed things seemed to bespeak a horrible strategy from early on.