Hacker News Reader

Comment by ryanwaggoner | Hacker News Reader

I find this really interesting, but I'm unclear on the proof that this made a big difference for Obama's campaign success. I'm sure Romney had people crunching the numbers as well, and he actually raised more money than Obama.

This was a pretty narrow victory, there are a lot of structural, ideological, and demographic biases at work here, and Obama did worse among most groups than he did in 2008.

Put another way, is there any evidence that Obama would have lost or even done much worse without this data crunching effort?

View on HN

There's at least some evidence that, in fact, the Romney campaign fooled itself into believing the "unskewed" polling models which simply (and incorrectly) disbelieved the demographic makeup of the polls. So yeah, it's hard to believe the republican campaign could have run a heavily data-driven campaign.

And the evidence that Obama would have actually lost is obviously not going to turn up. But you can look at the relative performance in "swing" states vs. the general electorate for hints. He clearly outperformed among the demographics that ended up deciding the election, as witnessed by the very large electoral victory on the basis of a very narrow (currently <2% I believe) popular vote margin.

Sorry, what is that evidence? The fact that they were saying they thought they would win? The fact that they didn't give up? Did you really expect them to? It's important not to confuse the public behavior of a campaign with their internal beliefs. And Romney really could have won this, had things gone a little differently.

The late attempts to to throw campaigning and advertising into states late Pennsylvania and Wisconsin (which, by the now-known-to-be-correct polls, were basically unwinnable) are probably the best evidence. There's also the more squishy testimony of journalists and pundits, who reported on genuine surprise coming from the senior Romney people about the exit polls. Karl Rove's freakout on FOX last night probably also counts.

None of that proves anything. But I think the burden of proof really goes the other way: the Obama campaign clearly seems to have played the moneyball correctly here. If Romney's folks knew they were going to lose, you'd have to show that.

And I don't know how to interpret "Romney really could have won this, had things gone a little differently". If you're talking about polling could have been different had he been winning, then of course: we'd be having the same argument in reverse. Though again, the evidence seems to say that Obama would still have outperformed the popular vote margin and thus have still been running a better data-driven campaign.

If you're saying that the data suggested by the end-campaign polls could have supported a Romney win, then no: that's just wrong, as evidenced by the discussion we're having. That kind of thinking is exactly the kind of "faith-based" notion that the Romney campaign is accused of relying on. The data showed a clear Obama victory. The data was right. Period.

I was interpreting your stance as a verdict on the Romney campaign's behavior over the last couple months, rather than just last night. And a month or two ago, they were behind according to the polls but they definitely could have still won.

Pennsylvania and Wisconsin I think are evidence of exactly the opposite: they knew they were going to lose Ohio and they had money to burn, so they threw a hail mary. Because why not?

Actually...

The Romney campaign is reported to have believed they were winning late into the day yesterday.

Based on what their model was telling them.

Not "faith", but data. A model. Math.

I have a strong suspicion in there's a good book in this election- how one set of Big Data led one team in the right direction, and another set of Big Data led the other team in the wrong direction.

This to me is the absolutely crucial thing to keep in mind. You can have your model and have your data, but that doesn't mean you've picked the right model and the right data.

Not the first time that lesson has been demonstrated, and it won't be the last.

The big difference in the models was their assumptions on voter turnout for each side. You can precisely measure what people prefer, but you can't precisely measure whether they will show up next week. So it was the 'art' part of modeling, not the 'science' of statistics, that the conservative pollsters blew.

It's hard to know, but one reason for the strong turnout might have been better get-out-the-vote efforts of the kind described here.

Follow-up articles seem to be indicating the Romney team was indeed wrong about turnout in each direction. Dems turned out more than they expected, Republicans less.

Not unreasonably. Most public polls supported that belief. http://www.washingtonpost.com/blogs/election-2012/wp/2012/10...

No, that's just wrong. Most "public polls" (rather: the median pollster -- obviously there were outliers in both directions) supported an Obama victory and a demographic and partisan makeup of the electorate that matched closely what was observed in the election. To claim otherwise is just silly, and perpetuating the "unskewing" nonsense that was proven incorrect.

If you're saying that the Romney campaign had "reasonable" justifications for disbelieving the data they saw, then I agree. If you're claiming that they actually had data the supported that position, then you're wrong.

We're talking turnout now, not preference. No one is talking unskewing.

The polls did in fact show that Republicans had higher enthusiasm, which is valid evidence for expecting better turnout. The better turnout did not materialize, but it was completely reasonable to believe that it might.

No, no, no. Polls include a prediction of turnout. That's the point of their likely voter models. They were telling you who was going to go to the polls already, yet the Romney campaign (and essentially all of the conservative establishment) chose to willfully ignore that data on the back of some cherry picked numbers (e.g. Rassmussen) and a set of "unskewed" models that were simply wrong.

There was never any data for that. It was all (incorrect) analysis. So don't say that there was data to support the position, there wasn't.

And the "enthusiasm" numbers were from early in the spring, before likely voter probing can be done. That stuff all disappears once the LV polls start coming out post-primaries.

No, it doesn't. Some polls ask about enthusiasm right up until election day.

Again, this is not about unskewing. This is about data as it was actually reported directly on the page.

Here's one of the polls directly referenced in the link I included in previous post. NPR Poll Oct 23-25, 2012. Just before the election. http://media.npr.org/documents/2012/oct/NPROctpoll.pdf

There's a 10 point scale that gives a 10 percent difference in enthusiasm to Republicans at the "10" level.

Of course that advantage vanishes if you include "8-10", which could be the sort of mistake Romney's internal pollsters made.

The poll only includes Likely Voters.

So there was data in public polls just before the election reflecting an enthusiasm advantage for Republicans. To say there wasn't is simply wrong.

I'm just at a loss. The poll told you how many voters it thought would appear, based on factors including internal modelling and the enthusiasm quesiton. And your argument is that you should throw the aggregate data out based on that one number, that they already included? That's insane; it's quite literally cherry picking.

And, of course, it's just flat wrong. That you would defend this behavior is just beyond me. It was a mistake, it wasn't ever a reasonable interpretation of the data, and plenty of smart people said so at the time.

The enthusiasm is a measure among the reported data. Not among the survey sample. The data was culled for likely voters, then the enthusiasm spread was among those that survived the cut. At the bottom of each page it says the results are all from the weighted sample.

The concept here is that "likely" is a continuum, and the Republicans had an advantage at the extreme upper end of it.

EDIT- Adding a note and then moving on from this. I understand you're saying the enthusiasm is baked into the final preference percentages, but that's not the case.

The Romney campaign’s equivalent was apparently some system called “Orca”, which involved volunteers on the ground sending information to a central location about the conditions that they saw. It wasn’t as extensive, and now we know that it gave misleading information to the campaign.

Two other things to keep in mind:

(1) The Republican leadership has been really really invested in the “fake it ’till you make it” school of persuasion: if you act like a winner, this will persuade the people who want to get behind a winner, and then you will win.

(2) Democratic operatives have known for decades that the demographic groups more likely to vote for them are also the groups that are less likely to vote at all. The Clinton-era solution to this problem was to pitch a message that was more likely to resonate with the population that was already voting; the Obama-era solution is to get more of those unlikely voters into the voting booths. Republicans don’t have as much headroom to improve their turnout.

It may have been a narrow victory in the popular vote, but it was an absolute stomp in the electoral college. Obviously we can't go back in time and try it a different way so you're never going to have any "proof", but there was clearly a strategic advantage by the Obama camp in the battleground states.

"I'm sure Romney had people crunching the numbers as well, and he actually raised more money than Obama."

Actually if you look at direct contributions, Obama raised nearly double what Romney brought in: http://www.opensecrets.org/pres12/index.php

SuperPACs allowed Romney to compete financially, but I don't think the number crunchers they talked about in the Time article were involved in the PACs

There is no way to disaggregate all the hundreds of factors that went into the 2012 election. So we'll probably never be able to say "X% of Obama's victory came from data mining and Y% came from hiring an addition Z Field Organizers".

But given the disadvantages Obama had going in, he definitely performed towards the top-end of what could be expected.

There is substantial and consistent reporting that Tuesday night's results were a surprise/shock to many Republican candidates and the associated outside groups. We know that they did a lot of number crunching themselves, so the implication is that the quality of Obama's data operation was superior--in that their predictions more closely resembled the actual results.

I think you assume that all "number crunching" is equal.

I absolutely do not, but how do you know which campaign had better number crunchers? You could go by who won, but you don't have a control; how do you know who would have won without the supposed advantage of the supposedly superior number crunchers behind the winning campaign?

I heard that Romney's number crunchers believed that they were winning till late evening Tuesday.

What other blunders did they make early on in the campaign?

>>This was a pretty narrow victory

What exactly is "pretty narrow" about a 303-206 electoral split?

I think he meant population-wise which doesn't reflect in the electoral count except in analyzing effects of 3rd parties on results.

Reply on news.ycombinator.com