Sortition in the Harvard CS department

Prof. Ariel Procaccia is giving a course at the Harvard Computer Science department titled “Optimized Democracy”. Procaccia has been writing about sortition for a while and sortition plays an important part in this course’s syllabus:

Students in the course explore the mathematical bedrock of democracy itself, and then build upon those foundations by applying computer science theory to some of the most vexing problems facing modern policymakers.

For instance, one topic covered in Procaccia’s course, sortition, dates back to the very first democracy in ancient Athens. Citizens seeking a position on the Athenian governing committee would self-select into a pool of candidates, intended to be generally representative of the city population, and then magistrates would be chosen by lot.

Fast forward 2,500 years and this method is still being used to form citizens’ assemblies. These assemblies have been used most prominently in Ireland to discuss constitutional reform and in France and the U.K. to debate environmental issues.

But there’s a problem.

“There is this tension between giving everyone a fair chance to participate in the citizens’ assembly, and this idea that we want to represent the population at large. The people who volunteer have a large self-selection bias,” he said. “That leads to very compelling algorithmic questions of how to balance these issues.”

12 Responses

  1. It’s hard to disagree with the last paragraph. As data scientists, both Yoram and Alex might have something to say on that


  2. It’s interesting that this course is being offered in Harvard’s Engineering and Computer Science department. I’ve mentioned citizens’ assemblies to a couple of engineer friends and they immediately brought up the problem that voluntary samples are statistically unrepresentative. To them it was just blindingly obvious, yet to people working on democracy studies you can devote 200 pages to explaining this (and they still don’t get the point). Both Yoram and Alex are engineers in their day job, so we really need to hear from them.


  3. *** Facts support Keith Sutherland concerns about representativity without conscription.
    *** The French “Convention Citoyenne pour le Climat” was formed by allotment from all citizens. But only 1/3 of the first allotted did accept. Thus the organizers used stratification, ensuring similarity with the general population along 6 parameters: gender (51 % women, 49 % men), age, school credentials, socio-professional class, kind of territory (from big towns to countryside) and geographical area (including overseas).
    *** A research team of Paris School of Economics (Adrien Fabre et al.) compared poll answers to questions between the 150 CCC members and a representative sample of 1003 French citizens.
    About the climate issue, there is strong similarity – thus we could say that for this issue the CCC is representative.
    But the study did show, likewise, that it is not true for all questions. Especially it is not true for some questions about general “values”. The general population, when questioned about values to be given to children, gives more weight than the CCC members to obedience (34 % vs 12 %) and work diligence (46 % vs 32 %). These answers are reactions to words, and, sure, we would need deeper studies to get the real ideologies. But anyway these differences did prove the CCC is not the mirror of the populus, is not an authentic minipopulus. And these discrepancies, or discrepancies about another points, could be important for other issues – outside the climate.
    We have here evidence that, without conscription, the minipopulus is not authentic. Discrepancies between minipopulus and referendum could come out, which would not be the result of best deliberation by the minipopulus, but from a lesser representativity.
    *** When Dahl proposed the mini-populus idea, he wrote:
    “The judgment of a minipopulus wouId “represent” the judgment of the demos. Its verdict would be the verdict of the demos itself, if the demos were able to advantage of the best available knowledge to decide what policies were most likely to achieve the ends it sought. The judgments of the minipopulus wouId thus derive their authority from the legitimacy of democracy.” This legitimacy discourse implies the minipopulus mirrors the populus, and that needs civic conscription.


  4. We’ve been over this idea of mandatory service before. Forcing unwilling citizens into a decision-making body is a very dangerous idea. The method of attaining low refusal rate is by making service sufficiently rewarding. At the same time it must be clear that accept the offered slots implies that high quality service is expected.


  5. Yes, we’ve been over this idea of mandatory service before. But I think the CCC example adds information, because I think the organizers worked as efficiently as possible to get representativity (I am not sure all experiments were as concerned by the point), and they failed.


    Fabre (Adrien) 2020
    “Comparaison entre les réponses des 150 et d’un échantillon représentatif”

    Fabre (Adrien) et al. 2020
    « Convention Citoyenne pour le Climat : Les citoyens de la Convention comparés à des échantillons représentatifs de la population française. Note de travail » halshs-02919695.


  6. Accuracy of the representativeness of a mini-public is not a simple cut and dry issue. For some relatively less critical policy decisions, a community might happily accept as legitimate a very rough approximation. they could happily embrace the decision because it was made by an impartial group of ordinary, but well-informed citizens, rather than by a self-serving or corrupt elite group. For high-stakes decisions, accuracy is more important, but there are ambiguities there too.

    Even a tiny “direct democracy” community will have some abstaining citizens. Some religions insist their members refrain from participation in any public policy making (election, lottery, referendum) as too worldly. In short, almost NO public policy making will ever be FULLY representative. With quasi-mandatory service some people compelled to attend will look at their phones, or nap, etc. This, however can be viewed as accurately representing the lack of interest of similar citizens in the general population. In reality, there is ALWAYS an element of self-selection involved in community life (and democracy). How much is too much to lose legitimacy is a hard question.

    My solution for high-stakes topics is to allow somewhat more self-selection distortion in the time consuming tasks of developing and refining proposals (with final refining still handled by a mini-public), and limit quasi-mandatory service to a very short duration jury-like task that listens to pro and con arguments and either accepts or rejects a policy. in the example of the French CCC, in my view, the final proposals need to go to a quasi-mandatory service, very large and more accurate mini-public for ratification (rather than to the elected government officials). With stratification, the CCC was accurate enough for the earlier pieces of the overall task.


  7. Andre,

    Thanks for the references. Yes – the CCC surveys are useful, although the findings are not surprising. In fact, it would have been rather surprising if there was a close alignment between the self-selected one third which accepted the offered slots and the rest of the population. Those findings however in no way indicate that mandatory service would be a good solution to the problem of providing good representation.

    I started writing a comment listing the many problems with the mandatory service idea, but there are so many of them that I think I will create a post about this issue instead. For now, it would suffice to say that, like stratification, mandatory service is a formalistic solution that does not address the root of the representation problem and at best creates a superficial image of good representation rather than actually attains good representation. However, unlike stratification, in the process of creating the false image of good representation mandatory service creates many new problems. Thus, unlike stratification which may be argued is better than nothing, mandatory service is a solution that is much worse than problem it purports to address.


  8. Terry,

    There is nothing wrong with people turning down offered slots as long as this choice really represents their informed and free choice to rely on the decisions made by others. In such a situation this is a perfectly legitimate choice. It is certainly a better and more legitimate choice than being forced to attend and then being inattentive and voting on a whim.

    The real problem is when people turn down offered slots – or accept them or forced into them but in fact are not attentive or active enough to represent themselves – because of constraints that are imposed on them in various ways. This could easily be the case even when they are formally present in some discussions and formally have up-or-down voting power. Thus, forcing people to attend adds nothing to the representativity of a decision making body and at the same time introduces a whole set of severe new problems. There is really nothing to say in favor of this idea.


  9. Yoram:> it would have been rather surprising if there was a close alignment between the self-selected one third which accepted the offered slots and the rest of the population . . . There is nothing wrong with people turning down offered slots as long as this choice really represents their informed and free choice.

    This avoids the crucial issue — 99.9% of citizens don’t even get the opportunity to turn down the invitation, so it’s essential that there is a close alignment between the sample and the rest of the population, otherwise the vast majority of the population are disenfranchised. How are you going to get round the entailments of your first claim? (that’s the first time that I’ve heard you acknowledge that it would be surprising if the voluntary sample was representative). As for the second claim, your focus on individual free choice has nothing to do with democratic equality. (But no doubt you will ignore my questions.)


    I’m puzzled by the notion that “approximate” representation will suffice for developing and refining proposals. This suggests that there is a “right” (or even True) outcome and that any errors introduced by approximation will be smoothed out by the final up-down vote of the jury. That’s a bit like saying it doesn’t matter who acts for the prosecution and defence in a law court as the jury will get it right in the end (anyone who has been watching the new Brian Cranston TV series will spot the flaw in that argument). There is a substantial body of evidence that demonstrates that imbalances in the perlocutionary force of speech acts and even the order in which people get to speak has a major impact on the outcome of deliberative exchanges. And pointing to imperfections in Athenian “representation” avoids the point that all citizens could (in principle) participate — Athens was a direct democracy, so of little relevance to large modern states. (And, if Mirko wins his arm wrestling match with Mogens, our cherished belief that the final decision was by a legislative jury will be wrong.)

    Andre was right to point to Dahl’s concerns — if the latter’s thoughts on sortition were more than a hastily-scribbled footnote we wouldn’t be having this extended debate.


  10. PS, Alex and myself argue that the only democratic solution to Dahl’s problem is for the advocates to be elected. Anticipating the reaction that this will make the alternatives akin to Tesco vs Sainsbury our claim is that, as well as its democratic value it will implement a Bayesian learning machine. This provides the ideal combination of Bayesian and deliberative reasoning (the verdict of the allotted jury). Who could possibly object to that?


  11. *** Habitual self-exclusion from mini-populus, or from vote, may come from personal parameters as strong egocentrism, or from religious affiliations leading to self-exclusion of the political body, as Bouricius reminds us. Well, we could say “they are not true citizens”. But it seems, in contemporary North-Atlantic polyarchies, it comes strongly from collective feelings of political alienation or collective feelings of political disability, especially strong in the lower classes.
    The effects of these feelings will be different along the situations. More British citizens did vote in the Brexit referendum than in parliamentary elections, and it is often thought it was a factor of the victory of Brexit.
    *** These facts would lead towards potential discrepancy between mini-populus and referendum, very hard problem.
    *** And these discrepancies will not be random. Often it will be possible to guess them.
    *** If I am not mistaken, Yoram Gat would exclude totally referenda from modern dêmokratia. Thus there would be no conspicuous problem, and any representativity concerns could be deemed metaphysical. It is consistent.
    *** But excluding totally referenda is a problem. Substituting partially, even totally sortition to parliamentary elections is not so much difficult, I believe, to get accepted, at least by French people. Cancelling referenda would be much more difficult, and would be rejected by many, including some of the more sensitive to the appeal of sortition.


  12. […] the murder of George Floyd, allotted police oversight commissions were discussed in California. A CS course at Harvard dealt with sortition and an algorithm for quota sampling from unrepresentative volunteers made it […]


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: