The challenge

2024 marks the 50th anniversary of the first known “minipublics”. Since then, the discussion about the “right” process for selecting citizens through random selection for a political office (aka sortition) is ongoing. Major progress has been made in the theoretical and practical field. We have seen a strong community of practitioners gather; an incredible amount of research being made. In some cases, we have even seen a full circle: For example, in Germany door to door recruitment was used in the very first Planning cells and is again gaining traction as a preferred method after four decades of ignorance.

Despite those progresses a core question remains: What is the best method to gather a representative sample of the citizenry to participate in a deliberative forum?

My hypothesis is following: There is none.

Instead, we need to embrace an intersubjective approach of sortition. An approach that puts at its core the realization that we do not have the technology nor the data to achieve an objectively clean sortition. More than that: Maybe we shouldn’t even try.

The argument of this essay goes as follows: What is sortition really; Why it is a socio-political process; Why we should embrace this and be modest about it; How we can work out a process to get a representative group people following an intersubjective approach.

What sortition is

To take decisions, Humans use a so-called selection procedure which can be defined as a process to go from many options to one, using a mechanism. There are four main categories of selection procedures: voting, market, examination (criteria), and random selection. How to distinguish these four procedures from another:

1. The vote uses the aggregation of individual opinions as a means of reducing the set of options. In a vote, preferences are translated into suffrages that are added with the help of an algorithm (simple majority, proportional representation, absolute or relative majority, consensus, etc.). This results in a reduced set of options.

2. In a market procedure, individual preferences are matched (reduced) using a price which is the balance between preferences (offer and demand). When offer and demand agree on a price, the reduction is done.

3. In an examination, the reduction is done by checking if a list of prefixed criteria is met. The most known of these criteria are rotation, birth (or nationality, inheritance), needs, merit, or time (a queue for example). Since the relevant criteria are defined by the actors themselves, and are combined, it is impossible to conduct a systematic enumeration here.

4. Random selection is different from the three previously described methods in that the reduction of the set of options bypasses the will of the actors: there is a moment of “uncertainty”. Random selection is a chance device (Alford 1958, 2) and reflects the “intentional choice to make the decision by a non-intentional mechanism” (Elster 1987, 108). The reduction is therefore based on the use of an aleatory moment.

Sortition is the use of random selection to select members of a decision-making body: Juries, Assemblies, Councils, etc.

Random selection can of course be used to reduce the set of options in other settings: In a lottery it usually determines who get a prize. In a coin flip it decides upon two options, for example who is going to kick the ball first in a football game.

Why is sortition a social and political process

Selection procedure, hence random selection and sortition are not artefacts isolated in space and time. They are embedded in a societal context. Each selection procedure not only is a mechanism but also an intellectual operation. When taking a decision people refer to “reasons why” they do it. For example, when selecting through an election they refer to the transparency, or the competence. When recurring to random selection they put in front reasons like equality of chance or protection against corruption. Market is often justified as being efficient or scalable or able to exist without a central coordination mechanism. Examination is justified by reasons like need or rationality. Random selection is often justified to be a way of allowing diversity, and to avoid vested interests or corruption. It means that any instance of sortition is based on the wish to achieve a particular goal.

Further, any selection mechanism is enshrined in a mental model of interpretation of which there seems to be 3: The first one is finalist meaning that the output is seen as the consequence of the will of a superior entity (god, the gods, destiny, etc.) the second one is deterministic meaning that the output is seen as the result of a causal chain that can be traced. The last one is probabilistic meaning that the output is seen as being based on a statistical probability of the event occurring given the input conditions. The third regime is the most shared nowadays.

“The names of all guild members over thirty years old were put in eight leather bags called borse. Every two months, these bags were taken from the church of Santa Croce, where they were ordinarily kept, and in a short ceremony drawn out at random. Only men who were not in debt, had not served a recent term, and had no relation to the names of men already drawn, would be considered eligible for office.”

Signoria of Florence, Wikipedia

In conclusion, we can say that sortition is the product of a design both in the tools that are used as well as in the justification that are given to its use. We can therefore embrace with confidence an intersubjective approach of sortition.

Why are current sortition processes unsatisfactory

Current processes of sortition are not satisfactory because they pretend to ignore the deeply relative nature of any selection process. Instead, they try to objectivize selection as if independent, objective criteria would exist to reach representativity. Let’s take 3 examples:

1. Age. In this case the goal is to mirror the age structure of the general population. But this omits the fact that young people are immediately rejected for being too young to participate (until 16, or 18 or any other age decided as being the right age). And the oldest people will simply not be able to participate because in high proportion they will be physically or mentally impaired (think 95+). Also, respecting the age pyramid may bring damaging biaises in countries with major imbalances like Western countries (a lot of old people) or countries like Morrocco with a very high number of young people (around 25%). This may be representative in the demographic dimension but completely unfair in terms of social justice or opinion biases in the group.

2. Gender: The classical dichotomy man/women is being questioned in many ways. More and more Human beings identify themselves on a more fluid scale of gender identities. How to treat this in a sortition process? Add one category (other)? But isn’t this a discrimination? So should we add 5 or 6 other options? And how to capture the % of the population identifying with those categories? And what if in a specific socio-cultural context this very option of going beyond two gender is seen as a bias by the majority of the population and leads potential participants to refuse to participate?

3. Attitudinal criteria: Many processes include some attitudinal dimension like a position on the topic at stake. For example, during the recruitment for the European Citizens’ panels, potential participants where asked about their view of the European Union from skepticism to support. This is a very interesting criteria as it allows to capture a diversity of views and can be related to a set of statistically representative data (Eurobarometer). But again: who is responding to Eurobarometer and can we be sure that there is not bias here. We have seen in the 2016 US American elections (and in many other contexts) that “representative” opinion polls can be very wrong in the way they reflect the actual views of the population. It means that the “universe” against which representativity can be measured is questionable. Moreover, attitudinal criteria are a rabbit hole: If we start adding one, why not add a second and a third and hundred more to better reflect the population?

So: Do we need to give up and flip back to election through vote as the best proxy for selecting a representative decision-making body?

The answer is no.

Intersubjective sortition

Whenever we want to use sortition, we will have to agree on the process that will make it representative in the eyes of the involved parties. This is what I call intersubjective. At each step of reduction, choices are being made. These choices are political, cultural, legal. They are different for each process and subjective. It means that there is a flaw in any sortition process because it relies on a process of design. That is true for any selection mechanism: deciding that women or children vote or do not vote is subjective.

“I am here as a soldier who has temporarily left the field of battle in order to explain - it seems strange it should have to be explained - what civil war is like when civil war is waged by women. I am not only here as a soldier temporarily absent from the field at battle; I am here - and that, I think, is the strangest part of my coming - I am here as a person who, according to the law courts of my country, it has been decided, is of no value to the community at all; and I am adjudged because of my life to be a dangerous person, under sentence of penal servitude in a convict prison”.

Emmeline Pankhurst. Freedom or death, speech delivered in Hartford, Connecticut on November 13 1913

Unlike the history of voting rights though, which has been a rather linear process of including more and more people, the discussion on sortition is more convoluted. Objectivity is even more relative. This is because of the justifications chosen to use random as a selection mechanism. Some arguments:

1. Equal chance: Such a justification would mean that we need to use pure random on the universe itself (everyone) to abide by the theorem of large numbers. This works only if the data on the population is complete from the outset and the number of participants high enough. In practice: impossible. And desirable? Do we want to create a database of all Human beings so complete that it allows to run that sortition? Maybe, maybe not.

2. Equal opportunity: Such a justification means to have a proactive process to overrepresenting or excluding some sectors of the population. But who is to decide? In the Panels of the Conference on the Future of Europe, it was decided that young people should represent 33% of the participants. This number is not statistically representative of the population as the 16-25 represent around 12% of the population and the 0-25 represent around 25%. The 33% came out of a negotiation process.

3. Inclusion: Such a justification means that we skew the universe and create a population by adding exclusion/inclusion criteria before applying any sortition. It can be age, but it could be sociodemographic criteria too: poorer people being overrepresented to counterbalance the power of rich people in society. Here again: The basis for the sortition is a political process.

As a conclusion, the best way to address this elephant in the room is to proactively propose an intersubjective approach to sortition: Sortition is a process of selection based on an intersubjective design: Stakeholders need to agree on the right set of justifications, criteria, and steps of the process for it to be legitimate. In this view, there is no intrinsic quality of the random selection but a procedural one.

Thinking in more practical terms, this process could be organized in six steps:

1. Go from the universe to the population: In this step, the goal is to define the possible options amongst the infinite set of options. For example, we can decide to focus only on living Human beings and exclude future generations and plants and animals. We can be more restrictive from the beginning and decide that only Human beings living in a specific country should be considered. Taking again the example above, we may decide to consider two genders, or 4, or more. At this stage, it doesn’t matter to know the proportion of each group in the population. So: We don’t need to know how many Human beings we have living in the country. We only need to decide if this category is part of the considered population. As an extreme example: To work on the topic of prostate cancer prevention policy, we may want to reduce the population for the sortition to male over 40.

2. Once the population is defined, we need to characterize it with the help of data: This is where we need to agree on a legitimate set of data as being good enough to represent the diversity of the population. In some context and for some data this may be easy. For example, the countries of the EU have good dataset on age. But, I once met an old man in Chile who was born in 1801 (so he told me at least). So, this may become tricky and the most important is to have agreement among stakeholders that the base data is legitimate.

3. Once the dataset is agreed upon (characterized population), it’s time to agree on the sample, meaning the subset of the population which is going to be the target for the group of participants. If my dataset says that I have 25% of my population aged between 15-25 and I have a group size of 100, my sample will have to contain 25 Human beings aged 15-25. At this stage, we may decide to skew the sample to reach certain goals. For example, we could agree to overrepresent the 15-25 age group to also consider the 0-14. But this will mean that we underrepresent other categories as we can’t go over 100%. The sample should be a number equal to the number of actual participants for each category (target).

4. The next step is to deploy the campaign of recruitment and reach out to real Human beings and profile them according to the criteria. This helps create the pool of candidates. Here again there will be a skew: Some target groups will be less responsive to the campaign: They will not open the door, will not answer the phone, will send their parents, sisters, sons. The random selection will be distorted, and it will be very difficult to know in which proportion. It is important to agree on the rules to handle such deviations (do we accept a person from the same household that allows to fulfill a slot in the pool?). It is also the major step in which the channels for the campaign need to be agreed upon: Random phone number generation? Letters? Door knocking? Social media? A mix? If a mix: What proportion of the pool should be recruited by each channel? The size of the pool should also be agreed upon: Do we want twice as much as candidates as participants to have a 1:1 reserve? Are we happy with 1,5 the size (meaning that we can accept 50% no show).

5. The next step is to select the actual participants from within the pool. This can be done by matching the targets of the sample with the candidates in the pool. This step is best done with software tools like panelot, StratifySelect, random.org, or equivalents. It can help create a main list and a reserve list.

6. As a last step there is a management process by which no show, and blatant deviations from the targets can be corrected as the process unrolls.

Conclusion

Sortition should be considered as a social and political tool and not as an objective artefact with intrinsic properties. Doing this, unlocks a design space for better concrete sortition processes, making them more transparent, legitimate and resilient.