Impossible Political Systems: Further Adventures in Rawlsian Constitutional Design

I am somewhat amazed, now that I think of it, that no “serious” political philosopher I know of has ever proposed an electoral system like the one I proposed in this post last week, where individual voting power is inversely proportional to income. (By all means enlighten me if anyone has proposed something like it; I would be delighted to know. I’m pretty certain that among the many forgotten pamphleteers of the 19^th century someone must have come up with a similar idea but I don’t have the knowledge to locate these potentially existing antecedents). This probably means that it is a bad idea (and judging by the few reactions I got, most people think so); but if it is a bad idea, I would like to explore in more detail the reasons why it is bad, since it is not obvious to me that a system like that would not meet Rawls’ principles of justice.[1] (And I sort of would like to see a few more responses).

A recap: I suggested (more or less tongue in cheek) that Rawls’ difference principle could potentially be met by a political system where everyone has a vote, but the formal value of your vote declines the more you earn. There are a number of different ways of achieving this, but the most interesting (at least to me!) version of this system is the following.

We divide voters into n income (or wealth) equal classes (or quantiles). Voters in the first (poorest) class have median income y and a single vote each, whereas voters in the nth (richest) class have median income a_ny (that is, the median income of the richest class is a times the median income of the poorest class) and 1/a_n^x votes each, where x is a number between 0 and 1 that determines the extent of the “disenfranchisement” of high income voters. “Income” here is post-tax, post transfer income.[2] The value of a person’s vote thus depends on their income class; more specifically, it is inversely proportional to the ratio between the median income of their class and the median income of the poorest class. For n > 1, if x = 0 then every voter has one vote, and the system reduces to the normal “one person, one vote” system; if x = 1 then the extent of rich voter disenfranchisement is strictly proportional to the average income of their income class, so a voter in the k_th income class would have 1/a_k votes.

A numerical example may be useful. Imagine that this system had been in place in the USA in 2008, with x = 1 (so the value of the votes of the richer classes is strictly inversely proportional to their income), and n = 10 (so there are ten classes of voters). According to the Luxembourg Income Study, the income ratio between the bottom and the top decile of the income distribution in the USA in 2010 was 6.154 (so a₁₀=6.154). In this imaginary political system, in other words, the poorest decile of the income distribution would have had about six times the voting power of the highest decile. About 75% of these people voted democratic in the 2008 congressional elections, according to the American National Election Study, whereas only about 33% of the people in the highest income decile did so. A very quick and dirty simulation (see code and explanation here) suggests that if this system had been in place in 2008, the Democratic party would have won about 62% of the two-party vote (61% if we assume turnout rates would have stayed the same, with poorer voters voting at lower rates than richer voters), rather than the 54% that it actually won – an 8% difference, which one imagines would have been translated into somewhat different policies. A system like this would thus have amplified the influence of the bottom decile of the income distribution (and of the lower half of the income distribution generally), though of course parties would have behaved very differently in the new environment, so the example is merely illustrative. (A simulation for New Zealand is a bit harder to do given our different electoral system and my inability to use the NZ Election Study, but I’d love to see one).

Note that we could in principle consider systems where x > 1 or x < 0, though I doubt such regimes would pass Rawlsian muster. If x is much greater than 1, the votes of the richer classes are discounted very quickly: with a small number of classes (say n = 4) we then get a “dictatorship of the proletariat”; with a larger number of classes and a very large value for x, we get basically a simple dictatorship of the very poorest people in society. Similarly, if x > 0 and the number of classes is small, we get a régime censitaire, where the rich have more formal voting power than the poor (like the Roman republic of Cicero’s time); if x is much smaller than -1 and the number of classes is large we simply get a dictatorship of the richest people in society.

Note also that if n = 1 then the system reduces to the usual "one person, one vote" system; for n = 2 voters below the median income each get one vote, while voters above the median income each get  votes, where a₂ is the ratio of the average income of voters above the median income to the average income of voters below the median income; and so on. The smaller n, the more abrupt differences in voting power are (though they are ultimately less steep), whereas the larger n, the more gradual and steeper the differences in voting power. So, for n = 2, the superrich end up with the same voting power as the middle class, and the lower middle class ends up with the same voting power as the very poorest, though the poor and the lower middle class end up with more voting power than the upper middle class and the rich; for n = 100, the superrich end up with much less voting power than the middle class, but the middle class in turn ends up with less voting power than the very poorest, even though voters in adjacent income classes end up having similar voting power. If n is high but x is close to zero we have smooth differences in voting power but a small gradient, so that rich and poor end up having similar voting power per person but there are many small gradations. All of this would be easier to show in a simple interactive simulation (a Mathematica notebook, perhaps?) with a couple of sliders for n, x, and some choice of potential income distributions, but that is beyond my ability to do right now; for now, all I can offer is some R code here if people want to play with various choices of parameters.

So how should we choose the parameters n and x? Current voting systems in democracies at least pay lip service to the idea that each elector is formally equal, i.e., that n should be 1 and x should be 0, even if in practice the value of some voters’ votes is larger than the value of others. (In elections to the US senate, the million or so Montana citizens have about 37 times the voting power of the 37 million or so California citizens, a ratio that is much higher than the ones contemplated in the numerical example of this system above. this is an approximation - I should look up the actual numbers of voters, not just the populations - but it will do for a ballpark figure). But would a person choose these exact parameters for an electoral system from behind a veil of ignorance? Rawls himself notes that one must evaluate political institutions by their tendency to produce just outcomes (they are forms of “imperfect procedural justice,” like jury trials); and it is not clear that n = 1 and x = 0 yield the most just outcomes. And the advantages, from a Rawlsian point of view, of choosing larger values for both n and x seem considerable.

Most people (including Rawls) would say that the rich have more influence than the poor in politics, influence that is disproportionate to their numbers (though not everyone thinks this is a bad thing). The reasons are obvious. Standing for elections costs money, and the need for financing campaigns from moneyed private interests may push certain issues to the forefront of the public agenda, and make others invisible. The rich can lobby representatives more easily, and have more ability to coordinate and spread their ideas than the poorest. In the USA, Martin Gilens has argued  that the views of the poor have almost no influence on the actions of their representatives when these views diverge from those of the rich. To be sure, not all of these things are necessarily negative. The ability of the rich to lobby can be construed as an informational subsidy to legislators. If the rich are better informed than the poor, policy that is nonresponsive to the views of the poor might be of better quality by some measures. But it is difficult to deny that political inequalities exist; the question is whether they are arranged to the benefit of the worst off. If they are not arranged for the benefit of the worst off, then it is possible that changing the values of n and x, i.e., giving more formal influence to the poorest, would serve as appropriate compensation for their economic inequality.

The thing about a system where n > 1 and x > 0 is that the value of any one person’s vote changes as society becomes more or less equal. Regardless of how many classes we choose, and what value we give to x, the more income-equal the society, the more equal the value of the vote of rich and poor, and in a perfectly income-equal society everyone would have exactly one vote. By the same token, the higher the level of inequality, the higher the value of the votes of the poor relative to the votes of the rich, and as inequality increases, the more political power the poor gain. x thus represents a kind of “sensitivity parameter”: the higher its value, the more sensitive the political system will be to inequality.

Moreover, a system like this would bypass debates about which economic policies actually reduce inequality or produce the most benefits for the worst off, i.e., which policies would meet the “difference principle” (assuming, of course, that the difference principle is the right principle of justice for socioeconomic matters; and it may not be). It makes no assumptions about which kinds of economic policy actually do help the poor; if laissez faire improves their position, then the poor would be in the best position to approve of it; and if some other policy worsens their relative position, then the poor would get a right of “first refusal.” (As inequality increases in a society, the poorest would gain more and more formal political power). In the spirit of the difference principle, the rich are thus allowed to benefit from (economic) inequality so long as the poor approve (with x and n setting the “approval parameters” of the system). Thus, the higher the level of inequality, the more disenfranchised the rich become, and the greater the compensation to the poor in the form of political inequality (benefiting the worst off the most), which in turn might enable them to change those policies. Of course, if your political theory does not depend on Rawlsian assumptions, this point might leave you cold; but even utilitarians might see potential benefits here. (And your choice of x and n might say something about how much a person is willing to give up for the sake of economic equality).

Now, there are many potential problems here. The rich might underreport their income. (Though this should only be a serious problem if n is large). The poor might choose policies that are not in the interest of society as a whole. (But so can the rich; the question is whether, on average, granting more political power to the poor would result in more just decisions). They might redistribute property. (Which would result in their losing political power as economic inequality decreases). If we start tweaking n and x who knows where we might end up. (We could end up with political systems that grant more political power to the rich). Loss of formal political influence by the rich might have unanticipated consequences in the form of additional corruption and so on. Formal distinctions in voting power are an affront to the equality of citizens, and offend our sense of fairness. (True, though people take very large inequalities in elections to the US senate, for example, completely in stride. Also see next point). Perhaps the most important objection to a proposal like this, which Jay Ulfelder raised in conversation on G+ when I posted the original idea, is that politics is not  only about economic issues; it is also about many other issues, which we evaluate from the point of view of equal citizenship. Issues about religion, civil liberties, etc. should not be subject to the predominant influence of one social group; they concern all as equal citizens.

I grant that this is a powerful objection to a scheme like this. But here’s a refinement that bypasses or at least mitigates it. Imagine a bicameral legislature. The lower chamber is selected through an electoral system like the one described above, where x = 1, and n = 10, for example. We could call this the “Chamber of the Difference Principle.” The upper chamber, by contrast, is selected through an electoral system of universal equal suffrage (x = 0, n = 1), perhaps including some of the “random constituency” ideas I discussed in an earlier post to ensure the representation of suitably general interests. We could call this the “Chamber of Equal Citizenship.” Determining the exact relationship between these two chambers is beyond the scope of this post; but the (Rawlsian) idea would be that these two chambers would represent the two standpoints from which we evaluate social institutions, and work together to produce law and policy.

Now, I myself don’t know for sure whether to take this idea seriously. I lean toward thinking that a system like this is not only too shocking to our normal ideas of fair representation to be ever politically possible, but is likely to have some bad unanticipated public choice consequences. So I don’t know where I would set x and n, if not at 0 and 1, even if I’ve half convinced myself that higher values for both would be somewhat desirable. But I would like to know where others would place these values. Do you think that voting power should be inversely proportional to income? If you’ve read this far, it would be great if you could answer this poll.

---

[1] Rawls does say in section 36 of ToJ that the political constitution of the just society would honor “the precept of one elector one vote” as far as possible; but he could be wrong about that, even by his own lights; and anyway a Rawlsian could argue that departures from the one elector, one vote precept are justified in nonideal situations (as Rawls himself does). Furthermore, Rawls does express concern about maintaining the “fair value” of political liberty under conditions of economic inequality, a problem which this system would potentially eliminate.

[2] This is mathematically equivalent to a system in which the members of the richest income class each have one vote, and members of the poorest class each have a_n^x votes; it doesn’t matter which description we use, except that in the second x should perhaps be called an “empowerment” parameter (for the poor) rather than a disenfranchisement parameter (for the rich).

More Rawlsian Thought Experiments: An Inverse Income Voting System

(For some unknown reason, re-reading Rawls stimulates my weird idea generator. Another politically impossible proposal here, presented as a thought experiment)

[Update 25 August 2012: see the more detailed discussion of this proposal here]

Thinking about the difference principle today, it occurred to me that most of the discussion on the topic tends to be overly focused on the economic institutions that would ensure that the least advantaged group in society would fare best. Yet the difference principle is not restricted in its application to economic inequalities; in fact, the principle specifies that inequalities of authority and political power must also be justified to the worst off group in society: “[t]he second principle applies, in the first approximation, to the distribution of income and wealth and to the design of institutions that make use of differences in authority and responsibility” (§11, p. 53, emphasis mine). Is a standard democratic system – with its panoply of elections, constitutional protections, and so on –justified in those terms? Rawls seems to assume so, even though standard democratic institutions entail clear inequalities in authority and responsibility which are not obviously to the maximum benefit of the worst off.

Moreover, to the extent that Rawls discusses the connection between political and economic inequalities, he tends to think of the direction of causation as going from economics to politics. Unjustifiable political inequalities in authority and responsibility (as, for example, when the rich have undue influence in the political process) would be remedied, in his view, once objectionable economic inequalities are taken care of, which seems reasonable enough; after all, he notes, such inequalities are correlated with one another (§16, p. 83). With low levels of economic inequality, widely dispersed ownership of the means of production, and a healthy dose of public campaign finance, Rawls argues, whatever differences in authority and responsibility political institutions would still produce would be justifiable to the least advantaged (cf. §13, p. 71, §36, p. 198).

But why wait until such inequalities are fully remedied? Imagine that instead of the standard, one person, one vote system, we had a voting system where the poorest person with any income got 1 vote, the person with twice the income of the poorest person got ½ a vote, and a person with 20,000 times the income of the poorest person got 1/20,000^th of a vote. “Income” here includes any government transfers; people with no income would have 1 vote, just as the poorest people with any income. The specific value of the vote would be linked to the income records on file with the national tax agency, and no one could vote who was not linked in some way to the tax system. We might use broad categories instead of specific incomes – say, people making less than $10,000 a year get one vote, people making up to $20,000 half a vote, and so on; and of course we might decide that a different weighting of votes is required. [Update: it occurs to me that using the median income would be easier to set up. For example, voters below the median income each get one vote, voters between the median and twice the median get half, and so on.] Whatever the case, poorer people would have more formal influence than richer people.

In a fully income-equal society, everyone would have 1 vote; in an extremely unequal society, the very poorest would have the most political influence. High-flying hedge fund managers with extremely high incomes likely would have an infinitesimal vote, though they would of course still have many means of exercising influence. After all, elections cost money, and the rich are more able to run for election and lobby legislators. But the idea is that the least advantaged groups would have compensating advantages, as politicians would have to cater to the greatest mass of votes; yet these advantages would diminish as transfer payments increased, or society became more equal. Note also that the specific value of a person’s vote would fluctuate throughout their lifetime, being very high when the person is very young or very old, and very low during their peak earning years. (The system might work kind of like an automatic means test for politically distributed benefits, baked right into the political structure).

Would there be anything wrong with this system, from a Rawlsian point of view? Everyone can still vote and run for office, so the first principle of Rawls’ theory - the principle guaranteeing the equal basic liberties - is not overtly violated; and it seems to me that the inequalities of authority and responsibility produced by this system are more clearly justifiable to the worst off group in society than the inequalities produced by a standard democratic system. For one thing, it provides the least advantaged in society with the fair value of their political liberties in a direct, unmediated way that complicated systems of campaign finance cannot match. Moreover, a political system along these lines bypasses the theoretical discussion about which economic system would best produce outcomes in conformity with the difference principle; different proposals can be tried, and if they worsened the lot of the least advantaged, the poor would gradually acquire sufficient political clout to overturn them. (In theory, this is compatible with pure libertarian laissez faire, so long as such laissez faire actually does improve the condition of the least advantaged group).

Like any oddball proposal, this is very likely a bad idea. (I can imagine all kinds of bad incentives to underreport income, for example, and I’m not sure it would fit with the well-entrenched idea that having an equal voice is a mark of respect of equal citizenship.) But I’m curious: are there specifically Rawlsian grounds to reject this sort of system? (I'm sure there are other grounds). And what are the obvious problems I’ve missed in here? Does this lead to a sustainable politico-economic equilibrium, or simply to an intensification of class conflict?

[Update 2, 20 August 2012: fixed some minor typos. Also thought of some further refinements. Suppose we divide the income distribution into N equal quantiles. The voters in the lowest class have one vote, and the voters in the highest class have (1/(N^x)) fractional vote, where x is a parameter determining how extreme the disenfranchisement of the rich is. When x=0 everyone has one vote, as today; when x=1, the disenfranchisement is strictly proportional to class position, so the each person in the nth quantile has 1/n votes; but we could set x between 0 and 1, leading to a range of different weightings of formal influence for the poorest depending both on the number of "classes" and on the extent of disenfranchisement with income. With few classes and low x, the system is close to our own; with many classes and large x, there is a very steep disenfranchisement curve. What values of these parameters would be chosen behind the veil of ignorance?]

Musical Chairs, Veto Constituencies, Accountability Juries, and other Random Ideas: Further Thoughts on Randomizing Electoral Constituencies

(Attention conservation notice: some more thoughts on this proposal, which gained a modest amount of internet attention in the last couple of days – see Dylan Matthews here, Matt Yglesias here, and Evan Soltas here).

Having slept on it and seen a few responses, I am not yet convinced that the proposal I sketched in the post below would actually be a bad idea (though it probably is!). The basic idea is that the electoral constituency or electorate – the group of people who “selects” the legislator – should be separated from the accountability constituency or electorate – the group of people who “disciplines” the legislator¹ – in such a way that neither the legislator nor the electors would know in advance who the latter group would be. Such a system would (in theory) mimic some of the structural features of a Rawlsian “veil of ignorance,” so that legislators could not tailor their appeal to any particular group of people but are instead incentivized to speak and act in ways that can be publicly justified to any group in society. But there are a variety of ways of accomplishing this, as I hinted at in the original post. Some of these are very close to current practice; others are not so close. I want to consider here in detail a few of these variations:

1. “Musical chairs” representation. This is the simplest form of the proposal. (My original idea was in fact closer to option 2 below, “random veto constituencies,” though it seems to have been quickly simplified into this form, which is at any rate simpler and more elegant). Under musical chairs representation, non-incumbent candidates can run in any constituency they choose, but incumbent candidates running for re-election are randomly assigned a constituency shortly before the election date (say, a month or so earlier). The random constituency can be their original electoral constituency (let’s call this the “initial” constituency). We do not need to imagine that the incumbent draws any constituency with equal probability; a system where the incumbent has a higher probability of drawing his/her initial constituency could strengthen the incentive to do the kinds of constituent service that a lot of people seem to find valuable.
   
   More formally, let’s say there are N constituencies or districts in the system; come election time, incumbents can draw their own initial constituency with probability p, where p > 1/N, and any other constituency with probability (1-p)/(N-1). (More complicated assignments of weights are of course possible). When p = 1, the system works identically to current practice – the incumbent always draws his own constituency; when p = 1/N, the incumbent can draw any constituency with equal probability. We might thus think of p as a measure of “localism” in the system. The lower p, the more the re-election seeking politician will need to act in ways that can be publicly justified to any constituency in the country, but he will of course have less incentive to do constituency service or to represent his locality. p can also serve as a measure of incumbent advantage; the higher p, the higher the advantage, all other things equal.
   
   What might be the benefits of a system where p < 1? By facing the possibility of having to compete for re-election in a constituency other than the one where he has been originally elected, a politician would of course need to avoid acting in ways that can incur disapproval beyond his constituency. But he would still have incentives to represent his constituency, pace Dylan Matthews. For one thing, the chance of drawing his initial constituency is nonzero, and a reputation for serving the constituency one is representing would still be a valuable asset in a race in a different district. Indeed, they might still be able to too easily promise to extract rents on behalf of local interests, as Evan Soltas notes, though I imagine it would be harder to promise this credibly when they don’t know which district they would be representing in the next election. At any rate, promises of local rents aren’t the only criterion by which electorates judge the suitability of candidates, and this system would weaken the appeal of these promises. Moreover, while a district or electorate could insist on electing candidates with purely local appeal, these would, I suspect, tend to be one-term wonders, as would “extreme” candidates. Assuming the distribution of voter preferences is not itself polarized for other reasons, the system would tend to moderate polarization over time as candidates strategically target the “modal” constituency to a greater or lesser extent, depending on the strength of their convictions and their level of risk aversion.
   
   Elections might nevertheless become more competitive under this system, since p < 1 implies a lower level of incumbent advantage, which might be a good thing (more competitive elections tend to be associated with more public goods provision, though this varies depending on the kind of competition). More importantly, as Matt Yglesias notes, the forms of public justification themselves would shift. An incumbent US congressman who defended agricultural subsidies in Iowa during his term, for example, would need to do so in terms that would resist potential sceptical examination by an electorate in New York. And one should not discount the satisfaction electors would get from giving a sound thumping to a politician they find particularly obnoxious but would not, under current practices, normally be able to defeat. (Soltas’ objection that it “weakens accountability structures to have your performance judged by someone else than whoever installed you” is I think overstated; electors judging the performance of a non-local candidate can always opt for the local challenger, or judge the non-local candidate on the basis of his/her record on issues or national significance, as they often seem to do anyway, and as is perhaps appropriate for candidates to a national legislature). Nevertheless, it is probably safe to say that in the aggregate this system would not produce large shifts in policy from the status quo; after all, many mixed-member proportional systems (New Zealand!) try to achieve similar balances of “national” and “local” concerns, even if they use somewhat different mechanisms and have somewhat different effects on the forms of public justification.


2. Random veto constituencies. My original proposal tried to preserve local representation in a slightly different way. Perhaps the simplest form of it is the following: non-incumbent candidates are elected according to the vote totals in the constituency they are running, but incumbent candidates are subject to a veto in a randomly selected constituency (assigned shortly before the election). In essence, the challenger must win only in the local constituency, but the incumbent must win in both the local and the randomly selected constituency. If the incumbent loses in the randomly assigned constituency but wins in the local constituency, the challenger takes power; hence the idea of a veto. The system could make allowance for extremely popular incumbents, so that getting, say, 60% of the vote in their local constituency means that they only have to get 40% of the vote in their randomly-assigned constituency, and getting 80% of the vote in the local constituency means they only have to get 20% of the vote in the randomly assigned constituency. (Again, the chances of drawing a particular constituency could be suitably weighted to favour nearby constituencies, for example).
   
   It seems to me that this system would have most of the advantages of the previous one while preserving the possibility of strongly-rooted local representation. Races remain local, but politicians cannot act in ways that have a strong probability of being disapproved of elsewhere in the country; and incumbent advantage is strongly diminished. Nevertheless, it seems possible that a system of vetoes could create a lot of ill-will; I’m sure people in say, rural South Carolina, would not relish having their choices of representatives vetoed by people in San Francisco or vice-versa. So it does not seem politically sustainable in the long run, even if it would enable the schadenfreude of spectators to see politicians having to justify themselves to hostile electorates beyond their home turf (a benefit that is not to be underestimated; see Jeffrey Edward Green’s The Eyes of the People for a full academic defence).


3. Accountability juries. One problem with the proposals above is that they favour geographical units of representation at the expense of other possibilities. While randomization of such geographical units may have some benefits, certain interests are never going to be well represented on that basis. One could, however, modify the idea of veto constituencies so that the veto would be exercised by juries selected on the basis of some non-geographical criterion. For example, imagine that before each election, we drew five juries of 500 people each selected on the basis of income – the first jury being composed of 500 people randomly selected from the first income quintile, the second from the second, and so on. Each incumbent candidate is then randomly assigned, shortly before the election, to make his/her case before one of these juries; you could be assigned to speak before a jury of the poor, or before a jury of the rich, but you would not know which one in advance. We could have weeklong, televised trials, perhaps presided over by special investigating magistrates, in an echo of the “audits” Ancient Athens subjected its officials to at the end of their terms. The jury could then vote (perhaps with some suitable supermajority requirement) on whether or not the incumbent should be allowed to run for re-election (in that election; no permanent disqualification is envisioned). The possibilities here are, of course, much more various: how we would organize such juries, run the trials, and so on would be crucial questions. I suspect a system like this would shift policy more radically, but I like the idea of having representatives having to justify themselves to randomly selected constituencies that do not have a geographical basis.

Anyway, I’m sure there are huge problems with all of these proposals, which I don’t expect to be political reality any time soon; this is more a thought experiment than anything else. I would nevertheless be interested to hear what some of these problems are, or what alternatives people can come up with.

Rawlsian Legislatures: A Modest Proposal

(Attention conservation notice: various harebrained schemes I cooked up preparing for a seminar on Rawls that appear to structurally mimic ideas about a “veil of ignorance.” Of purely theoretical interest.)

[Update 13 August: see also my further thoughts on these proposals here].

John Rawls’ A Theory of Justice famously introduced the idea of an “original position,” a hypothetical situation in which citizens would come together behind a “veil of ignorance” to select principles of justice that can regulate their common life. There are different ways of understanding the OP, but one useful way – which Rawls himself favoured later in life – is to imagine that the “contracting parties” in the original position are not the members of society themselves, but rather their representatives. Each of these representatives – modelled as rational negotiators – is then supposed to bargain for the best possible “deal” acceptable to the citizens they represent on the terms of cooperation in society, but without knowing which specific set of citizens they represent. This is supposed to ensure that the negotiating parties will only agree on principles that would be acceptable to all citizens as “free and equal.”  Leif Wenar in the Stanford Encyclopedia of Philosophy describes the basic point well:

The original position is a thought experiment: an imaginary situation in which each real citizen has a representative, and all of these representatives come to an agreement on which principles of justice should order the political institutions of the real citizens. Were actual citizens to get together in real time to try to agree to principles of justice for their society the bargaining among them would be influenced by all sorts of factors irrelevant to justice, such as who could appear most threatening or who could hold out longest. The original position abstracts from all such irrelevant factors. In effect the original position is a situation in which each citizen is represented as only a free and equal citizen, as wanting only what free and equal citizens want, and as trying to agree to principles for the basic structure while situated fairly with respect to other citizens. For example citizens' basic equality is modeled in the original position by imagining that the parties who represent real citizens are symmetrically situated: no citizen's representative is able to threaten any other citizen's representative, or to hold out longer for a better deal.

The most striking feature of the original position is the veil of ignorance, which prevents other arbitrary facts about citizens from influencing the agreement among their representatives. As we have seen, Rawls holds that the fact that a citizen is for example of a certain race, class, and gender is no reason for social institutions to favor or disfavor him. Each party in the original position is therefore deprived of knowledge of the race, class, and gender of the real citizen they represent. In fact the veil of ignorance deprives the parties, Rawls says, of all facts about citizens that are irrelevant to the choice of principles of justice: not only their race, class, and gender but also their age, natural endowments, and more. Moreover the veil of ignorance also screens out specific information about the citizens' society so as to get a clearer view of the permanent features of a just social system.

In Rawls’ view, the veil of ignorance can also play a role in the selection and evaluation of constitutions and laws. While the representatives of the citizens in the OP are supposed to select principles of justice in complete ignorance of the citizens’ class, gender, plan of life, and even the general features of their society, the veil can be “lifted” gradually to allow the representatives to agree on how these principles apply to more concrete institutions. This is what Rawls calls the “four-stage sequence”:

After agreeing on the two principles and a principle of just savings, the parties then proceed further through the four-stage sequence, tailoring these general principles to the particular conditions of the society of the citizens they represent. The veil of ignorance that screens out information about society's general features is gradually thinned, and the parties use the new information to decide on progressively more determinate applications of the two principles.

At the second stage the parties are given more information about the society's political culture and economic development, and take on the task of crafting a constitution that realizes the two principles. At the third stage the parties learn still more about the details of the society, and agree to specific laws and policies that realize the two principles within the constitutional framework decided at the second stage. At the final stage the parties have full information about the society, and reason as judges and administrators to apply the previously-agreed laws and policies to particular cases. When the four stages are complete the principles of justice as fairness are fully articulated for the society's political life.

Re-reading Rawls recently while preparing to teach a class, it struck me that it would be possible to mimic some of the structural features of this interpretation of the veil of ignorance in actual legislatures.

The simplest way to do this, it seems to me, would be to divorce electoral constituencies from accountability constituencies. Suppose legislators are elected in a relatively large number of small single-member constituencies (I’m thinking of a small place like New Zealand, where electorates are small, but one could imagine more complex schemes elsewhere). They go to a Parliament or Congress and negotiate laws as best as they can. At the end of their term, however, they must justify themselves to a randomly allocated constituency (not necessarily the one in which they were elected), which decides whether or not they can run for re-election. (A variant: the accountability constituency [also?] has the power to impose a financial penalty on the legislator if it finds the justifications for its actions lacking). The trick here is that the constituency that can hold the legislator accountable is not known in advance, either to the electors or to the elected MPs. If Rawls is correct, this should encourage elected legislators to negotiate “fair” legislative proposals –legislative proposals that are broadly acceptable to all in society.

An example may help. Imagine the electors for Wellington Central elect Grant Robertson their MP. At the end of his term, the Electoral Commission randomly assigns him a different constituency. Say he draws Auckland central, for example. Robertson then has to go to Auckland Central to defend his record in parliament; let’s say he’s given one month to make his case. Auckland Central then holds an “up or down” vote deciding whether or not he can run in the next election. If he’s voted down, he cannot run in that electoral period (though he may run in later periods – no permanent disqualification is envisioned here); otherwise, he gets to run again, if he so wishes, in Wellington Central.

One can easily imagine all kinds of problems with this system. (Consider the possibilities for strategic voting; and I’m sure the pros could come up with all kinds of ways of gaming this system). But I’m having way too much fun thinking about it to worry about these inconveniences right now. For example, imagine accountability constituencies that are functional or income-based rather than geographical. Legislators could be elected in standard geographical constituencies, but then randomly assigned to income-defined constituencies to make their case for being allowed to run for re-election. We might imagine that large “juries” of people from specific income quantiles could be empanelled, and MPs randomly assigned, at the end of their terms, to make the case for their policies to one of these juries in week-long trials. The juries then decide whether or not the MP is to be allowed to run again. Or imagine we got rid entirely of electoral constituencies. Instead, people would vote on the abstract composition of the legislature (expressing their preferences not only about the party composition of the legislature, as in closed list PR systems, but perhaps also their preferences about the level of education or income legislators must have, what percentage of legislators must be women, of a particular race, etc.). Political parties are then tasked to fill a legislature with these characteristics, but the legislators must then, at the end of their term, justify themselves to a randomly assigned constituency, which has the power to impose fines (and perhaps to award prizes).

Ok, so what’s the benefit of this, you may ask? If Rawls is correct, the fact that legislators would not know in advance to whom in society they would be held accountable would mean that they would be inclined to act in ways that are “publically justifiable” to all, including the “least advantaged.” What do people think?