Participation Through Lottery

From "The Trouble With Elections: Everything We Thought We Knew About Democracy is Wrong," Chapter 10.3

How should survivors of a shipwrecked cruise liner set up a democracy?

When proposing lottery selection instead of either election, or mass-participation self-selection (as in a referendum), there are two common concerns. The first concern, that random selection could generate an unrepresentative group of people,¹ is not a serious issue, as I will discuss next. The second concern, that some people will agree to serve while others decline, so that those who end up actually serving will be unrepresentative, is a serious issue, which I will address subsequently.

Fear that an unrepresentative group will just happen to be drawn by chance, is a frequent worry when people first learn about sortition. What if a group sympathetic to neo-nazis happens to be drawn by lot? A short digression into sampling mathematics is necessary to appreciate the imagined and actual risks of random sampling. 

The highly developed techniques of scientific sampling derive from what is known as the law of large numbers. The term was introduced by the French mathematician Siméon-Denis Poisson in 1835, and it plays a major role in probability analysis. Poisson sought to model how predictable and stable probabilities of groups of random events, people, or flips of a coin, arise despite the absolute uncertainty about each individual event, person, or coin flip. If I want to determine if a coin is a fair coin, in other words there is an equal chance of landing heads or tails, just a few flips is not sufficient.  As I flip the coin over and over (a large number of times), even if it lands heads on the first several flips, if it is a fair coin, the total number of heads and tails will steadily get closer to  a 50/50 split.²  I don’t need to flip the coin a million times to decide.

It is obvious that a very small random sample of the population may by chance be extremely unrepresentative of the general population (though unlikely as unrepresentative as most elected legislatures). For a jury of twelve, the chance that a particular segment of society that comprises 30 percent of the whole population will happen to get a majority of seats in a random sample is 4 percent. But as the size of the jury grows, the odds of this happening drops precipitously. If the random sample has one hundred members instead of twelve, the chances of that 30 percent minority of the population getting a majority of seats drops to 1 chance in 110,684 (or 0.0009%). In other words, if a new jury of 100 were drawn every year, it would probably take many tens of thousands of years before even one jury had a majority of members from that 30 percent minority. If the jury had 250 members, the chances of that minority group getting a majority of seats drops to just 1 chance in 79,792,344,770. A crucially important fact is that the size of the random sample required to get a “very accurate” sample is barely impacted by the size of the population being drawn from. In other words, as I stated in an earlier chapter about Athenian democracy discussing scale, “A sample of 6,000 citizens (typical of the People’s Assembly) could accurately represent a population of 300,000,000 as well as 30,000.” What’s more, for practical purposes, a sample of 500 is also extremely likely to create a very accurate representation of a population of 30, 000 or 300,000,000. 

Looking at it another way: what are the odds that women, are wildly over- or under-represented in a body of 500? (We will estimate women at 50 percent of the population for simpler calculations, setting aside the complication of transgenderism, and the fact that women are somewhat more likely to outlive men, etc.). If the ideal is 250, we can expect that about 95 percent of the time, the number of women selected by lot will range between 273-227 , and that about 99.5 percent of the time, between 284-216. The chance of having a split that is worse than 300-200 is about 1 in 100,000. There is virtually no realistic chance that women would be as under-represented as they are in nearly all elected legislatures throughout the world today.³ If stratified sampling (discussed below) is used, it is simple to assure that the number of women would be exactly 250.

Non-participation and Democracy

So, a random sample of reasonable size will be quite representative of the population. But not all of those selected will agree to actually serve, such that the issue of self-selection bias resurfaces. If all segments of society were equally likely to decline, the random selection dynamic would still apply, and the representativeness would not be affected. However, it seems plausible that certain sorts of people would be less likely to agree to serve than other sorts of people, which could stymie the descriptive representation goal. For example, new mothers, or entrepreneurs launching a new business, or people with social anxiety would likely decline. But it is also likely that people with limited education, or who feel discriminated against as part of a marginalized group might decline. Depending on the issue being tackled, their absence could change the outcome of the panel’s deliberation, and destroy the legitimacy of the citizen’s assembly. 

Some theorists have called for quasi-mandatory service (akin to US jury-duty) to avoid these concerns altogether. Others view “forced service,” as a violation of freedom. It is also possible that those compelled to serve unwillingly may simply not pay attention or “do the work,” leaving the demographic or character they are hoped to be representative of, effectively unrepresented, despite their nominal inclusion. If what is called scientific, or stratified sampling is used (randomly selecting, but with quotas for a variety of demographic traits), among the pool of people willing to participate, this may assure representation that adequately approaches representativeness. But this can only have any hope of validity if there are no significant barriers to participation, and their non-participation is a genuine free choice. Alternatively, it can be argued that the token “forced” juror, who attends but ignores the deliberation, or the one who simply doesn’t show up, is in fact accurately representing, by their very refusal, the refusal to participate typical of such people in society.

A controversy in democratic theory as it relates to sortition, is how we should think of the impact of some people declining to participate in a lottery selected body. If there are obstacles to participation, such that some groups within society are systematically less likely to participate even if called, and the resulting body is un-representative of the population, the democratic legitimacy of the mini-public is forfeited. One question that I have not heard anyone fully address is, how should we think about a situation where the resulting body ticks off all the demographic categories (age, income, race, sex, education, etc.), but some unmeasured trait (or even unmeasurable) such as cognitive style is unrepresentative? Such a body would still be dramatically more representative than any elected chamber. Is that the best that can be hoped for democracy in the real world?

A thought exercise may be useful. Imagine a cruise liner shipwreck with 1,000 stranded survivors on an isolated island, who decide to live democratically. Initially they have the whole community gather to make decisions, though some passengers decline to attend, saying they trust whatever decisions the rest make. If only six hundred attend, is that a democracy? What about if only two hundred choose to participate? The group quickly agrees that it is impractical to have all members of their tiny democracy participate in all meetings to make all decisions, since labor is needed harvesting coconuts, fishing, and building shelters. They decide that a randomly selected group of one hundred, with regular rotation, is a good solution. If only sixty of the hundred chosen by lottery show up for the first meeting, should they round up the decliners and make them attend? What if they notice that a certain demographic group (perhaps young adults, or women) are particularly absent? Should they randomly select some replacement young adults and women (stratified sampling)? 

It seems to me there are three key factors. Firstly, are there obstacles that are causing certain types of people to decline (e.g. the young people are needed for climbing coconut trees to help the group), so that it isn’t really a free choice? Secondly, is the quality of the group decisions being hurt by the lack of diversity with loss of valuable information and perspectives, due to the absence of certain sorts of people? And thirdly, does their democracy retain or lose its legitimacy among the other 900 people if they learn that only sixty of the hundred drawn by the lottery actually participated?

In the next post I will examine how non-participation is actually being addressed using stratified sampling in citizens’ assemblies around the world today.

1. Indeed, at the US Constitutional Convention there was discussion of randomly drawing members of Congress to constitute an electoral college to select the president. The idea was that those drawn would retire to another room, and in order to limit deal making, not be allowed to leave until a president had been chosen. But due to a fundamental lack of understanding of probability mathematics (which was then barely being invented in Europe), some members were concerned that the random group might all be from one region of the new country, and the idea was dropped.

2. I should note that we would also need to alternate starting the flip with the tail or head side up, since a 2023 experiment with 350,757 coin flips using coins from 46 different nations and 48 different people flipping, found that due to precession in mid-flip, any coin actually has a 50.8% chance of landing with the same side up as it had when launched into its flip. So to get a true fair (50%) series of coin flips one would need to alternate which side was up at the start of each successive flip.

3. There are a handful of elected legislatures that have around 50% women, but these are rare exceptions.