Congratulations, Bernie!

Congratulations, Bernie, on a surprise win in the Michigan primary! But my Bernie-phile friends: Please don't fall for the bad cognitive errors I've seen some supporters distributing in responses: binary and one-step thinking, and being misled by inapt metaphors.

First, "a win-is-a-win" carries a lot of associational baggage, some of which may be true but which is certainly worth some careful analysis, but it's a binary thinking error. In Michigan, Bernie beat Hillary by 1.5% of the vote. A win, right? But in delegate count – what matters in this primary election – Hillary took 70 and Bernie 67, increasing her lead. For the primaries and for the election as a whole, one needs to keep in mind that we live in a republic, not a democracy. That means we elect a representative government, and do not directly elect a president. Winning the total popular vote is not the point (just ask Al Gore) and this is reflected in both the Democratic and Republican primaries, though in different ways.

To see how this works, consider three districts of 10,000 voters. The winner of each district gets a delegate. Suppose candidate H wins two districts by 6,000 to 4,000 and candidate B wins one district by 9,000 to 1000. Candidate H gets a total of 13,000 votes, while candidate B gets a total of 17,000. A big popular vote margin for B (57% to 43%) but a win for H (2 to 1). While this feels unfair, it's a way of guaranteeing that the political process requires coalition building among diverse sub-populations. We're seeing this in Bernie and Hillary's struggle to get the votes of different ethnic groups, different age groups, and different economic classes.

In a parliamentary democracy with many parties, like in many European countries, this plays out by having to build coalitions among parties. In the USA with only two parties, the coalitions are built at this stage. I don't think this is a bad thing, as I think the strength of America is our ability to (sometimes gingerly) bring together many different viewpoints, ethnic groups, and cultures, and get them to live together in reasonable harmony without frequent tribal and inter-group violence (so far). (Sorry, Black Lives Matter, I'm not trying to belittle your legitimate claims about inter-group violence in the US, only to point out that while horrible it has not reached the level of open warfare and we seem to be finally bringing it into the open enough to possibly make some positive progress.)

Second, well, but "it's an unprecedented upset." This one-step thinking also carries a lot of associational baggage: it means "momentum"! Look at the derivative! That implies big change. Well, perhaps, but one learns in science that projecting derivatives is a tricky and unstable business. (See Mark Twain's quote on the growth of the Mississippi Delta.) Also, the "upset" depends on the difference between a poll and an election. An election is the event: its result is what it is (modulo errors, cheating, hanging chads, etc.) The poll is a sample that is much more akin to a measurement in physics. This plays quite well with stuff I teach in my physics class about measurement.

A measurement in physics is also a sample: an attempt to determine the property of something by "tasting" it – taking a little bit in a way that you can analyze the sample and not change the object being measured. Consider a thermometer as an example. When I'm poaching a salmon for a dinner party, I put a thermometer in my salmon poacher to test the temperature and find out how hot the water is. My students often assume "a measurement is a measurement and gives a true value", but it doesn't work this way. A measurement is simply a conjoining of two physical systems. What makes it a measurement is a set of theoretical assumptions about the process of their interaction. In the thermometer case, we assume:

·            The zeroth law of thermodynamics: Energy will move between two objects in thermal contact in a direction to equalize their temperature (thermal energy density). So energy flows from a hot object into a cold until they are the same temperature. This says we expect our thermometer to extract energy from the water until it is the same temperature as the water.

·            The probe does not affect the state of the measured object significantly: The thermometer removes some energy from the water and so reduces its temperature. We assume that it only takes a little and that reduction can be neglected. If I used my big poacher thermometer in an espresso cup to see if it was too hot, the temperature the thermometer reads would not be the original temperature of the coffee but something partway between.

·            The probe has a linear response: We calibrate our thermometers by placing them in melting ice and putting a mark 0 ^oC and then in boiling water and placing a mark at 100 ^oC. The bimetal in the coil (or the liquid in the thermometer) expands as it gets hot and shifts the marker on the dial. We assume that halfway between those points is 50 ^oC and so on, but that isn't necessarily the case. It could expand more when it's colder and slow down when it gets hotter.

Thermometers are carefully analyzed and can be trusted when used appropriately. (A similar analysis holds for voltmeters and ammeters.) But the point is: When we make a measurement it depends on theoretical assumptions about how our system is working.

What does this have to do with polls? Well, a poll is a sample. A few voters are chosen to stand for the full population. The sample is too small to be chosen randomly: the error would be too large. So typically polls begin with a model of the electorates demographics: who does the voting population consist of and which of those are likely to actually vote in the election. These are often based on previous similar elections. But Michigan has not held a truly competitive Democratic primary in a long time. 2012 Obama was unopposed. In 2008, Michigan tried to slip forward in time so as to be more important, and the DNC stripped half their delegates. Many of the candidates (including Obama) refused to campaign. The two previous primaries were caucuses.

So it may be that there is a tidal wave of surprise support for Bernie. But it could also be that the Michigan polls were based on crappy models. A failure of polling yes, but not representing a shift in support. The way we will tell is if somewhat similar states such as Illinois and Ohio that have had more recent contested primaries, and where primaries are held next week, also show significant underpolling for Bernie or not. I am willing to wait and see.

Third, I'm afraid I'm seeing a lot of "Cinderella underdog" metaphors; the idea that somehow the election is like a basketball tournament. You just have to keep winning the popular vote. But because of the electoral college this is a terrible metaphor and leads us astray. As Democrats we want to win the presidency. To do so we need a path to 270 electoral votes and since those states are almost all (except, I think Nebraska and New Hampshire) winner take all, it takes a careful analysis of an electoral strategy; how an where to devote resources to get out the vote – and which populations to concentrate on. This is where the great detail we are getting in the Democratic primaries can help us. And it is why "national polls" of one candidate against the other are, especially this early in the game, essentially useless. Not only do these show dramatic swings as the candidates face off against each other, they don't take into account the actual election mechanism.

If neither candidate gets a majority of the delegates as a result of the primaries (there are all those "superdelegates" or SDs), here's what I hope would happen. The SDs would all throw away their current commitments and turn to the Quants – the quantitative analysts who would make models of the presidential election based on various models of the electorate and the details of the primary results in the various states. There would be a spread (spray) of results – similar to what you see for paths of a hurricane – because of different assumption plus random factors. The SDs would then use their personal knowledge of their own districts to evaluate those models and make their choices. That seems to me a good reason to have SDs.

Maybe I'm dreaming to hope that things would work out this way and choose the best choice for the fall election based on a detailed analysis of what we have learned from the primaries, but I'm a bit afraid that the SDs would look to support their personal interests rather than the interests of the party. I'm sure that wouldn't be true of my SDs – representatives whom I voted for and like very much. It's just all those other folks you voted for!

In any case, I will actively support whoever appears to have the best likelihood of winning the actual election, based on a careful analysis of our country's complex voting problems, not based on my agreement with their program (Bernie 98% to Hillary 94%), nor on my assessment of who is likely to be a more effective president in practice (Hillary 4: Bernie 1). I am very dismayed at the direction the Republican party has been trending over the past 35 years and it seems to be getting worse and worse. (Full disclosure: I voted for Republicans in New York State Senate elections in the 1960's but have never voted for a Republican presidential candidate.)

So to my Bernie-phile friends who say he can win, I say, OK, show me! I'm watching!