Suppose you wanted to describe the political landscape efficiently. That is, you want to describe the range of existing political positions, and develop some way of evaluating which political positions are “similar.” (I’m American, so I’m going to think in terms of the US.) There are three ways you could go about this and they are radically different.
1. Politicians’ voting.
Because we have a two-party system, politicians (and some other political players, like Supreme Court justices) make decisions on a single, left-right axis. You can explain almost all the variation with one dimension: “More Republican” or “More Democrat.” Generally, if a politician supports one “Republican” position, he’ll support most of the other “Republican” positions. Studies have been done here (too lazy to look them up right now) supporting this phenomenon at all levels of politics.
2. Individuals’ opinions on poll questions.
Here, we do something different. We ask poll questions about a variety of issues and see what people think. Suddenly, it’s not one-dimensional any more. There are people who like low taxes and also like gay marriage. There are people who are anti-abortion and anti-war. The fact that individual opinions don’t line up along a left-right axis is what’s behind ideas like the Nolan Chart—although it doesn’t imply that the “real” way to explain opinions is necessarily two-dimensional. How many dimensions actually do efficiently explain all the different kinds of opinions? I don’t know, and that’s a statistical puzzle in itself, although I do know of techniques (like multi-dimensional scaling) that purport to estimate the “right” number of dimensions.
One thing we do know is that if we project all the opinions onto a set of dimensions that we like—for example, the old standby of liberal/conservative and authoritarian/libertarian—we can start to measure “political diversity.” People who describe themselves as Democrats, for example, span a much wider range of views than people who describe themselves as “Republicans.” Implicitly, we’re putting a Euclidean distance on a high-dimensional space, projected onto a few dimensions.
3. Logical and causal implications of policies.
Just as politicians’ voting doesn’t capture what regular people think, it could also be argued that people’s opinions on poll questions also don’t capture something important about politics. Namely, some policies logically imply one another. If you support raising spending on the war in Afghanistan for 2011, you must also support continuing the war in Afghanistan in 2011. If you support spending more on X, you must either support less spending on some other thing Y, or support raising taxes, or support running a higher deficit. Simply asking people about their opinions on a variety of questions doesn’t capture this structure.
Additionally, some policies have consequences for other policies. You cannot simultaneously be in favor of reducing carbon emissions, and be in favor of a set of policies that, on net, increase carbon emissions. Or, you can, but you’d have to be either ignorant or confused.
And some philosophical claims imply policies. If you believe “Congress should not do anything except the enumerated powers in the Constitution” then that implies you have to oppose all the things Congress does that are not enumerated powers.
In part 2, every policy position was just a coordinate in a high-dimensional vector space. Now, in part 3, it’s suddenly not so simple. You have a directed graph of implications. It’s very hard to get a handle on this graph. But just as this graph is more intractable, it’s also more informative. In the part 2 model, a person could easily support a variety of inconsistent positions, and, in fact, people do. In part 3, some of your policy choices are determined by your other policy choices—not just by correlation but by necessity.
What can we say about one policy node that implies a lot about other policy nodes? Well, it’s very influential. If you could examine a single person’s directed graph of the policy universe, with nodes colored red for “oppose” and green for “support,” then the choke-points, those nodes that imply the choice of color for lots of other nodes, are core beliefs. A pacifist’s graph, for example, is heavily influenced by the “War is wrong” node, because that logically implies his position on all specific wars.
Obviously this doesn’t have to be binary; you could have degrees of support and opposition, that imply updating the degrees of daughter nodes. Increasing your support of “End all wars” should increase your support of all nodes “End war X,” but increasing your support of “Go to war to defend allies” should decrease your support of “End war Y” if an ally was attacked in war Y.
This gives us a different way of defining which policies are similar to each other. Policies that are close “cousins” on a tree are similar; “End the war in Iraq,” and “End the Korean war” are close because they’re both implied by “End all wars”, but “End the war in Iraq” and “End the war in Afghanistan” are even closer because they’re both implied by “End the War on Terror” (which is implied by “End all wars.”)
Three kinds of similarity are not the same
Points that are close on the left-right spectrum of politicians’ decisions are not necessarily all that close in the higher-dimensional space of people’s opinions on poll questions. I also hypothesize that points that are closely correlated in people’s poll answers are not necessarily close in the part 3 sense of being close cousins on a tree of implications. That is, I would guess that people may tend to hold opinion B whenever they hold opinion A, even when A and B actually have nothing to do with each other.
How different is your debating partner?
I think part 3 is a good model for having political conversations—better than part 1 or part 2. How “different” a person’s politics feel from your own, once you’ve had a discussion with him, is not so much a matter of which party he votes for, or what his opinions are on a laundry list of issues. No: a person feels really “different” when you and he have opposite opinions on one of your influential nodes, something really far back on the tree. If one of your influential nodes is “Democracy with universal suffrage is the best form of government” and you’re talking to someone who says, “Hi, I actually support monarchy,” then that person is really different from you. A person who agrees with you up until the last branch on the tree is pretty similar to you, no matter how vehemently you disagree about that last branch, because you both can draw upon the same assumptions from farther up the tree.
The personal lesson is that it’s useful to be clear with yourself about levels of difference. People can spend a lot of time arguing or even hating people who are very similar, and completely forget that there’s a higher level on the tree. My politics are pretty different from Sarah Palin’s; but I’d be Sarah Palin’s ally against Louis XIV, I’d be Louis XIV’s ally against Aurangzeb, and I’d be Aurangzeb’s ally against a giant space squid intent on annihilating Earth—Aurangzeb may have wanted to get rid of all the Hindus in India, but we’ve got more in common with each other than with a planeticidal space squid. Squids aside, it’s rather ridiculous when people identify their “greatest enemy” or “greatest threat” as a person with only slightly different political views.
Three kinds of political similarity
Suppose you wanted to describe the political landscape efficiently. That is, you want to describe the range of existing political positions, and develop some way of evaluating which political positions are “similar.” (I’m American, so I’m going to think in terms of the US.) There are three ways you could go about this and they are radically different.
1. Politicians’ voting.
Because we have a two-party system, politicians (and some other political players, like Supreme Court justices) make decisions on a single, left-right axis. You can explain almost all the variation with one dimension: “More Republican” or “More Democrat.” Generally, if a politician supports one “Republican” position, he’ll support most of the other “Republican” positions. Studies have been done here (too lazy to look them up right now) supporting this phenomenon at all levels of politics.
2. Individuals’ opinions on poll questions.
Here, we do something different. We ask poll questions about a variety of issues and see what people think. Suddenly, it’s not one-dimensional any more. There are people who like low taxes and also like gay marriage. There are people who are anti-abortion and anti-war. The fact that individual opinions don’t line up along a left-right axis is what’s behind ideas like the Nolan Chart—although it doesn’t imply that the “real” way to explain opinions is necessarily two-dimensional. How many dimensions actually do efficiently explain all the different kinds of opinions? I don’t know, and that’s a statistical puzzle in itself, although I do know of techniques (like multi-dimensional scaling) that purport to estimate the “right” number of dimensions.
One thing we do know is that if we project all the opinions onto a set of dimensions that we like—for example, the old standby of liberal/conservative and authoritarian/libertarian—we can start to measure “political diversity.” People who describe themselves as Democrats, for example, span a much wider range of views than people who describe themselves as “Republicans.” Implicitly, we’re putting a Euclidean distance on a high-dimensional space, projected onto a few dimensions.
3. Logical and causal implications of policies.
Just as politicians’ voting doesn’t capture what regular people think, it could also be argued that people’s opinions on poll questions also don’t capture something important about politics. Namely, some policies logically imply one another. If you support raising spending on the war in Afghanistan for 2011, you must also support continuing the war in Afghanistan in 2011. If you support spending more on X, you must either support less spending on some other thing Y, or support raising taxes, or support running a higher deficit. Simply asking people about their opinions on a variety of questions doesn’t capture this structure.
Additionally, some policies have consequences for other policies. You cannot simultaneously be in favor of reducing carbon emissions, and be in favor of a set of policies that, on net, increase carbon emissions. Or, you can, but you’d have to be either ignorant or confused.
And some philosophical claims imply policies. If you believe “Congress should not do anything except the enumerated powers in the Constitution” then that implies you have to oppose all the things Congress does that are not enumerated powers.
In part 2, every policy position was just a coordinate in a high-dimensional vector space. Now, in part 3, it’s suddenly not so simple. You have a directed graph of implications. It’s very hard to get a handle on this graph. But just as this graph is more intractable, it’s also more informative. In the part 2 model, a person could easily support a variety of inconsistent positions, and, in fact, people do. In part 3, some of your policy choices are determined by your other policy choices—not just by correlation but by necessity.
What can we say about one policy node that implies a lot about other policy nodes? Well, it’s very influential. If you could examine a single person’s directed graph of the policy universe, with nodes colored red for “oppose” and green for “support,” then the choke-points, those nodes that imply the choice of color for lots of other nodes, are core beliefs. A pacifist’s graph, for example, is heavily influenced by the “War is wrong” node, because that logically implies his position on all specific wars.
Obviously this doesn’t have to be binary; you could have degrees of support and opposition, that imply updating the degrees of daughter nodes. Increasing your support of “End all wars” should increase your support of all nodes “End war X,” but increasing your support of “Go to war to defend allies” should decrease your support of “End war Y” if an ally was attacked in war Y.
This gives us a different way of defining which policies are similar to each other. Policies that are close “cousins” on a tree are similar; “End the war in Iraq,” and “End the Korean war” are close because they’re both implied by “End all wars”, but “End the war in Iraq” and “End the war in Afghanistan” are even closer because they’re both implied by “End the War on Terror” (which is implied by “End all wars.”)
Three kinds of similarity are not the same
Points that are close on the left-right spectrum of politicians’ decisions are not necessarily all that close in the higher-dimensional space of people’s opinions on poll questions. I also hypothesize that points that are closely correlated in people’s poll answers are not necessarily close in the part 3 sense of being close cousins on a tree of implications. That is, I would guess that people may tend to hold opinion B whenever they hold opinion A, even when A and B actually have nothing to do with each other.
How different is your debating partner?
I think part 3 is a good model for having political conversations—better than part 1 or part 2. How “different” a person’s politics feel from your own, once you’ve had a discussion with him, is not so much a matter of which party he votes for, or what his opinions are on a laundry list of issues. No: a person feels really “different” when you and he have opposite opinions on one of your influential nodes, something really far back on the tree. If one of your influential nodes is “Democracy with universal suffrage is the best form of government” and you’re talking to someone who says, “Hi, I actually support monarchy,” then that person is really different from you. A person who agrees with you up until the last branch on the tree is pretty similar to you, no matter how vehemently you disagree about that last branch, because you both can draw upon the same assumptions from farther up the tree.
The personal lesson is that it’s useful to be clear with yourself about levels of difference. People can spend a lot of time arguing or even hating people who are very similar, and completely forget that there’s a higher level on the tree. My politics are pretty different from Sarah Palin’s; but I’d be Sarah Palin’s ally against Louis XIV, I’d be Louis XIV’s ally against Aurangzeb, and I’d be Aurangzeb’s ally against a giant space squid intent on annihilating Earth—Aurangzeb may have wanted to get rid of all the Hindus in India, but we’ve got more in common with each other than with a planeticidal space squid. Squids aside, it’s rather ridiculous when people identify their “greatest enemy” or “greatest threat” as a person with only slightly different political views.