As a philosophy major, I am deeply troubled when I come across such disregard for the literature. What follows is my response to an email from the Obama campaign received today, and below that the email itself.

——–

Hello “Info,”

I am generally strongly in favor of your mailings and read them with zeal. Further, I campaigned for Obama in 2008 and intend to vote for him next November. However, I have major reservations regarding the content of this recent mailing. My difficulties are not political or tactical, but rather philosophical. Let me elaborate.

In your mailing, you present an argument that, upon analysis, works like this:

1) A mathematical model has been published in a reputable newspaper suggesting Obama will not win the coming election.

2) The reliability of predictive models depends upon a deterministic universe in which each state causally necessitates one and only one proceeding state.

3) We do not like the results of the model.

4) Therefore, we live in an indeterministic universe and the model is unreliable.

I hope it is apparent that the introduction of premise (3), and thus the move to the conclusion (4), the acceptance of which requires far-reaching metaphysical commitments strongly at odds with our current best picture of our world, is disreputable. The body of scientific research to date strongly suggests a deterministic universe, and the philosophical consensus is that even the admission of quantum relativity does not seem to offer much hope for free will. If we accept science and value consistency in our beliefs, we must define free will in a way that is compatible with determinism (a conceptual project with many distinguished proponents).

There is, of course, the possibility that the published model, now part of the deterministic system, will spur Obama’s campaign and its supporters to action, rendering itself inaccurate. This, however, is all in principle explicable through the lens of determinism. There are myriad subtleties to unravel here, but I found your mailing to disregard them with a wanton frivolity and reckless unseriousness.

For further elucidation on these and other matters, please refer at your convenience to the following resources, courtesy of the Stanford Encyclopedia of Philosophy and Wikipedia:

Thank you for your work in service of our nation. I am confident you will make every attempt to refine your language to meet the standards of rationality maintained by your supporters.

Best,

Peter Sloan

On Mon, Nov 7, 2011 at 3:24 PM, Jim Messina, BarackObama.com <info@barackobama.com> wrote:

Peter —This weekend, The New York Times Magazine ran a long analysis of the 2012 election headlined, “Is Obama toast?”It uses a mathematical formula to conclude who will win this race.In other words, it says neither you nor Barack Obama has a role to play in this election, because the outcome is essentially predetermined.We disagree.The outcome will depend on what we do every single day between now and November 6th, 2012. And I want to give you an idea of how we know that.Our Republican opponents, from Mitt Romney and Herman Cain to Rick Perry and Michele Bachmann, have endorsed the same set of Tea Party policies that drive the Republicans in Washington: letting Wall Street write its own rules again and giving special treatment to millionaires and billionaires while asking seniors and middle-class families to pay for it.All of them would return to the failed economic policies that led us into recession.

Yet the Times piece assigns each of them a score on an ideological scale, ignoring the obvious reality that there has been virtually no difference among the GOP candidates — or between them and the Republican congressional leaders who refuse to do anything to restore economic security for the middle class.

Whoever wins the nomination will no doubt try to appear more “moderate” as they compete for undecided voters in the general election. But they have all made their positions clear. And we will hold them accountable for that.

The only true difference in this race is between their agenda and President Obama’s. Facing historic challenges when he came into office, he has fought every day for a fairer economy where everybody who does their fair share gets a fair shake.

He’s stood up to credit card companies to ensure they can’t target consumers with hidden fees. He’s stood up to insurance companies, who can no longer deny health care coverage on the basis of a pre-existing condition. He’s stood up to Wall Street to end taxpayer bailouts and rein in the kind of risky financial behavior that nearly toppled our economy.

These dramatic differences between the Republican nominee and President Obama will be crystal clear to Americans as the 2012 election approaches, because our grassroots organization in all 50 states will be having conversations every single day with their friends, families, co-workers, and neighbors.

That grassroots organizational advantage is a critical factor in this election that the Times’ “formula” doesn’t consider at all.

More than 1 million people have already taken ownership of this campaign. Millions more are organizing their communities on behalf of the President, online and off. This weekend, we had our single biggest day of action of the campaign — more than 2,000 volunteer events took place across the country, and more than 10,000 volunteers participated.

This work is already having an impact across the country.

We expanded the electoral map in the last election, fighting hard for — and winning — states like North Carolina, Colorado, and Virginia so that the entire election didn’t hinge on the results in a single state, as it had in 2000 and 2004.

We have no intention of returning to the old electoral map. And the organizing you’re doing means we won’t have to. Today, we are showing signs of strength in states we didn’t win even in the watershed election of 2008 — states like Georgia and Arizona, where a recent poll had President Obama beating every potential Republican nominee.

The map isn’t as friendly to our opponents, who won’t be able to compete in traditionally Democratic states because their organization won’t compare to ours. Whether you measure donors giving or doors knocked, there’s grassroots enthusiasm for President Obama that the other side can’t match — but that theTimes doesn’t consider relevant.

The truth is this isn’t the first time you’ve been written out of the story by many in Washington and the media — and it’s not the first time they’ve been completely wrong about that.

In the 2007 and 2008 campaign, almost everyone in professional politics said it wasn’t Barack Obama’s “turn” to be president. But millions of people like you took responsibility for the campaign — knocking on doors, making phone calls, and donating whenever you could.

You proved everyone wrong — not just about who was going to win the election, but about the ability of everyday Americans to come together and change the course of history.

The entire premise of the Times article is that you won’t — and can’t — do it in 2012.

The election is now less than one year away. No one thinks it will be easy. But there can be no doubt its outcome depends on how hard you and I work over the next 364 days. Right now, we’re opening field offices in key states, hiring organizers, recruiting volunteers, registering voters, and getting ready for what’s going to be one hell of a fight.

So, is Obama toast? It’s up to you.

– Messina

Jim Messina

Campaign Manager

Obama for America

The unfortunate thing about Nate Silver’s move from an independent to a nytimes based blog is that he publishes a lot less of his raw data and models, which was what set him apart. The details are not clear, but essentially what he’s done is some kind of regression analysis where P(Win) = B1(EconomicGrowth) + B2(ApprovalRatings) + B3(IdeologyOfOpponent), where the value of the B terms are estimated from past election data. These past election data are thought of as “random samples” even though this isn’t true, and give us predictions on what the next random sample, Barack Obama, might do depending on his values for these 3 opponents.

Anyways, whatever arguments you might have about the murkiness or validity of his model, i wouldn’t say it’s deterministic. At least, its no more deterministic than claiming that the odds of drawing 1 red ball in a box of 1 red and 4 black is 1/5. Essentially that’s what he’s done.

Tom, you say the model isn’t deterministic. I’m not sure I can make sense of this. Don’t all predictive models assume a deterministic system that they are modeling? Even predictive models of chaotic systems work by offering as narrow a range of predictions as possible, which I understand as offering a prediction at the first deterministic level available (so, like, I can’t tell you how many quarks will be up or down when the ball flies through the window at 36MPH, but I can tell you the ball will fly through the window at 36MPH).

I’m econ-illiterate, though, so please educate me.

“You say the model isn’t deterministic”

later…

“Don’t all predictive models assume a deterministic system that they are modeling? ”

I don’t mean to answer for Tom here, but I have a few things to say about a “deterministic model.” First off, there are a few things this could mean, 1.) that the model itself unvaryingly produces certain values, given an input; and 2.) that (1) is the case and that those values have an immutable predictive correspondence with real things in the world. Let’s say we have a model called “How many cookies are in the cookie jar”: A – B = C.; A representing ‘how many cookies were in the jar to begin with’, B ‘how many am I taking out now’, and C ‘how many are now left’. There is no question of the model’s determinism via criteria 1. Criteria 2 has little room for doubt, although we could do some silly tricks here to break its predictive certainty. Say we imagine a pernicious little niece “Suzy”, that takes from the jar while we still have it open and aren’t looking. The model only accounts for me, not little Suzy, and thus gets ‘out of sync’ with the world. Although it is a silly example, one could say in this later scenario that our model doesn’t ‘predetermine’ real events in the world. This wouldn’t be equivalent to saying that the universe isn’t deterministic, rather that our model doesn’t reveal the actual future products of this ‘physics-churning’ that we assume is underneath it all, that our model does not fully describe the real “happenings” of the universe.

Now if we threw out all models that were not ‘congruent’ with the real “laws-of-the-universe” (as if we had a means to confirm that relationship), then we could again assume that predictive models are somehow always deterministic. But that sounds like a gross mistake if we understand Science to be a ‘useful-theory-finder’, rather than a ‘True-theory-finder’. It seems to me from the following statement that you have unnecessarily restricted “predictive models” to make uncertain claims:

“Even predictive models of chaotic systems work by offering as narrow a range of predictions as possible, which I understand as offering a prediction at the first deterministic level available”

From my understanding, solid state physics involves making probabilistic claims about what path a given electron/photon will make under certain conditions. If we cut out all of the uncertainty, we might not have much to say other than that ‘it will move’. But probabilities happen to be useful, not in predetermining something’s behaviour, but in telling us what’s likely, so I’m not willing to take away its status as a “predictive model.” To bring it all back to the main point, I cant imagine there exists a predictive model which could make such election predictions as an “available deterministic level.”

Hey Colin,

Thanks for your thoughts. My point was that making a prediction assumes determinism, not that the prediction causes later events.

Think about it this way:

Predictions involve two things: an observation of the current state of the universe, and a statement about a future state. The two are linked by some belief about causation.

Any belief in causation seems like it implies a belief in at least some degree of determinism. In other words, in a totally indeterministic system, causation is impossible.

Here’s a fancy graphic (arrows are logical entailment):

Prediction -> Causation -> Determinism

Thus, Prediction -> Determinism

Peter

@Peter: Tom is saying that the Nate Silver model (which I have not seen, btw) relies on a multivariate regression: P(Win) = B1(EconomicGrowth) + B2(ApprovalRatings) + B3(IdeologyOfOpponent), where:

— P(Win) is the dependant variable to be predicted by the model. It is the probability of the “Win” outcome as a dimensionless quantity (i.e. it has no units). Probability has no units quite like radians have no units. Similarly, just as you might say a dimensionless parameter has units of radians if it appears in the right trigonometric context, you could also say that the same parameter has “units of probability” if it appears in a context like the above regression equation.

— EconomicGrowth, ApprovalRatings, and IdeologyOfOpponent are the independent variables of the model. For simplicity, lets say these are also dimensionless. For example, and quite literally, IdeologyOfOpponent seeks to give a quantitative response to a question like “On a scale from 0 to 10, what is the ideology of the model cadidate’s opponent?” Presumably, 0 represents some defined extreme and 10 represents the extreme opposite 0’s archetype on a 1-dimensional spectrum.

— The values B1, B2, and B3 are the model parameters chosen to minimize the model error. While the other quantities mentioned above are true variables of the model, the B’s will actually turn out to be constant coefficients, not variables. Recall the linear equation y = m*x + b (slope-intercept form, to be exact). The variables y and x are like P(Win) and EconomicGrowth, respectively. The quantities m and b are like B1 and B2 (except that b does not multiply an indep. variable as B2 does). Actually the B’s in the regression model behave exactly the same way as m does in the linear equation. All else held constant, an increase of one unit in the indep. variable x causes the dep. variable y to increase by m. Similarly, all else held constant, an increase of one in a candidate’s EconomicGrowth measurement would *predict* an increase of B1 in that candidate’s probabililty of winning an election (note that B1 could be negative).

Anyway, that’s that.

“Also, you say the model isn’t deterministic” — I believe Tom means that it’s not deterministic in the mathematical modelling sense (see http://en.wikipedia.org/wiki/Mathematical_model ). It’s a stochastic model. Basically, that is just to say that the model cannot guarantee repeatable results. Two candidates with the same values for their independant variables may not be observed as having the same probability of winning, even though their *expected* probability of winning would be the same. Or, slightly differently, two candidates that differ only by a single predictor variable will not necessarily exhibit, even qualitatively, the predicted relationship. For example, all other predictor variables equal, there could exist a candidate A with a lower actual probability of winning an election than another candidate B, despite A having a higher EconomicGrowth score than B (assuming EconomicGrowth correlates positively with P(Win)). This is true of any regression model that represents an imperfect correlation.

I left out an important part of the equation, which is actually P(win) = b1(growth) + b2(approval) + b3(opponent ideology) + epsilon, where epsilon is the residual error term. Essentially, the Obama email is making a push to drive that epsilon error term up.