A Test of "Collective Control" Theory

Continuing the discussion from [The more I read the more questions I get…

The following quote in that thread caught my attention because it suggests a way to test the “collective control” model of social behavior.

This sounds like it is a plausible quantitative prediction of the “collective control” theory of social organization proposed by Kent McClelland and described in his chapter in the Handbook. The prediction is that civility increases, and thus the need for police decreases, as the size of a social group – a collective – increases. So the collective control theory predicts that there will be a negative correlation between the size of a collective and the number of police per person needed to maintain civility.

I thought that this was very likely to be a correct prediction. I tested it by looking at the relationship between the size of cities and the ratio of police department size to population. I used a somewhat small sample of cities (12) with populations that ranged from 8.8 million (New York, NY) to 43,000 (Prescott, AZ). The ratio of police to population ranged from New York’s .006 (6 police/1000 people) to Prescott’s .001 (1 police/1000 people). The correlation between population and the ratio of police to population was .71, which is positive and statistically significant (p<.01, with apologies to Mike Acree;-).

So contrary to “collective control” theory, an increase in the number of people participating in collective control does not seem to increase the civility of the collective in the sense that it does not decrease the need for people to police the collective; indeed, the evidence so far is that an increase in the number of people participating in collective control seems to increase incivility, requiring more police per person in the population.

The results of this analysis are consistent with my observations of the behavior on two collectives that are familiar to me, CSGNet and this Discourse site. My informal observation has been that increases in the number of people participating on these collectives has not increased civility. Indeed, the level of incivility seems to have increased as the number of people participating in them has increased, which is the main reason I haven’t been posting much (or at all) lately.

This is great advice for anyone who is interested in learning, testing or applying PCT. PCT is unlike all other theories because it is a theory that is aimed at explaining a phenomenon that no other theory explains: the controlling that organisms do. In order to understand PCT you have to know what the phenomenon of controlling IS and how to recognize it in the behavior of living organisms. So, yes one has to be “anchored” to the phenomenon of controlling – which is seen as the behavior of organisms – in order to understand the theory that explains that phenomenon: PCT.

You have two variables here: the number of people who “participate in collective control of civility and comity” in the group, and the total number of people in the group. You have substituted the latter for the former. Classic straw man.

It doesn’t seem like a straw man of any sort to me, but whatever it was it was completely unintentional. When you said:

“As more of us participate in collective control of civility and comity here, no individual will have a ‘police’ role imposed on him or her”

I just assumed that you were talking about all those who come to participate in discussions on Discourse; the entire “collective”, I didn’t realize that “collective control” theory predicts that the need for policing is reduced only when the size of the collective is increased by people controlling for civility and comity.

But if that’s the prediction of collective control theory, how do you test it? How do you determine the degree to which the size of the collective is being increased by people controlling only for civility and comity? And what about the people entering the collective who are not controlling for civility and comity? It seems like the prediction about the need for policing the collective should also depend on the degree to which the size of the collective is being increased by people who are not controlling for civility and comity.

It seems like it would be very difficult if not impossible to get a measure of the proportion of those in a collective who are controlling for civility and comity relative to the proportion of those who are not. I would like to know how to measure those variables so that I can test the theory. Surely those variables must have been measured already since the theory makes predictions about their relationship to the need for policing.

I’m looking forward to finding out how to properly test your prediction that “as more of us participate in collective control of civility and comity here, no individual will have a ‘police’ role imposed on him or her”.

Best, Rick

Rick, did you seriously just take a comment out of context, misinterpret it, convert it into a hypothesis with no consensus from the generator of the statement, and then carried out your own ‘research’ on it to prove the unclaimed hypothesis as wrong?
Rick, please, science cannot work this way. I’m fascinated, and perturbed, at whatever goal you prioritise over and above others to justify this means.
I’m sure you can answer your own question - using the TCV - anyway - but I wouldn’t expect the necessary data to be collected and publicly available already!
It’s not to late to be collaborative…

I would like to retract the phrase about a straw man argument. It is an unwarranted assertion of motivation. I apologize.

The question would be better framed as what proportion of the population participates in collective control of civility and comity. We would have to specify those principles of communication in such a way that we could identify who is controlling them and who is not. Variable gain and potential conflict with control of other variables would fuzz this, of course, but for a rough cut you could pretend it was a binary choice.

Here’s an interesting corroborative case.

“He also credited the island community for coming together to make difficult decisions and rallying behind masks and vaccinations to help prevent the spread of the virus.”

There is genuine community here, despite the annual influx of city dwellers eroding that even as they seek it. (For everyone I know in the winter, there are 4 to 6 other people on Island in the summer, with their cars.)

That said, wouldn’t it be better to investigate the phenomena of collective control in simpler and more direct ways than population statistics involving variables that are difficult to specify and quantify?

I wasn’t trying to investigate a phenomenon of collective control; I was testing to see how well Kent’s model of collective control accounts for a phenomenon of collective control.

I’m pretty sure I got the prediction of the model right. My prediction was that the observed “virtual” reference state of a variable (degree of civility and comity) controlled relative to different reference levels by members of a collective will increase in stability (vary less) as the size of the collective increases.

I thought you had described a collective control phenomenon that could be used to test this prediction of the theory. The phenomenon is the amount of policing needed as the size of the collective increases. I thought you had predicted that as the size of the collective increases, the amount of policing needed would decrease.

This seemed like a reasonable prediction based on Kent’s collective control model, under the assumptions that a) everyone in the collective is controlling for civility and comity, though at different reference levels and b) that less policing would be needed if the stability of the “virtual” state of comity and civility in the collective increased.

I think it was reasonable to try to test this prediction using cities as the collectives and then using a) measures of the populations of those cities as measures of the size of each collective and b) the size of the police force in each city as measures of the stability of the virtual reference state of the virtual controlled variable: the civility and comity in the collective. I assumed that the bigger the police force, the greater the stability of the virtual reference for comity and civility and, thus, the less the need for policing.

But you say it’s not just an increase in the size of the collective that reduces the need for policing; it’s an increase in the number of people who are controlling for comity and civility. I think this could be added into a test of the model by adding crime rates to the prediction of policing level. The assumption would be that crime rate is inversely related to the virtual reference for civility and comity in a city. So using multiple regression you could “factor out” the level of the virtual reference for civility and comity across cities and see what the independent effect of city size is on the size of the police force.

I think I’ll give this a try. Yes, it involves the use of conventional statistics. But we could make the test more “PCT-like” by showing how the model behaves, in terms of the level and stability of the virtual reference for comity and civility, with changes in the total size and of the collective and the relative proportion of its members controlling at different levels for comity and civility.

At least it would be a start at a quantitative, model-based test of Kent’s “collective control” model of the behavior of collectives. Actually, there are at least three different ways organisms can interact in a collective – see the Social Control chapter (Ch. 7) of The Study of Living Control Systems. Kent’s is only one example of a PCT - based model of phenomena that involve the controlling done by groups of individuals – phenomena that could be referred to as “collective control”. So it’s somewhat misleading to call (or think of) Kent’s model as the model of collective control; it’s a model of collective control.

Best, Rick

This could be true if all members of the population are controlling the variables in question. Your prediction applies only to the subset who are.

It doesn’t matter, of course, whether you or I or Mr. Rogers thinks they are; what matters is whether they are.

I haven’t looked at the model of collective control published by Kent, since Kent and I worked together for a while developing the concept of collective control together with the mathematical analysis. So this comment refers only to my own current understanding of the concept. I assume Kent’s ideas are not too much different, but I don’t know whether anything he says contradicts what I say here.

  1. The core idea of collective control is that the control actions of several control loops controlling their own perceptions can create a situation in which some variable that is in principle identifiable using the Test for the Controlled Variable can be deduced as being controlled, with a determinable present value and a determinable reference value, with a certain loop gain. This variable is called a Collectively Controlled Environmental Variable (CCEV), and it is controlled as though an entity called a Giant Virtual Controller (GVC) is controlling a virtual perception to some determinable reference value.

  2. The Gain Rate of the GVC control loop is a function (with linear individual gain rates, the function is the sum) of the gains of the individuals controlling their own individual perceptions. (In Kent’s seminal demonstration experiment, there was only one dimension and two controllers, and therefore the two controllers necessarily controlled perceptions of the same environmental variable, but this is not true in general). The virtual reference value of the virtual perception of the CCEV is an averaging function of some kind among the different reference values for the perceptions controlled by the individual contributors to the GVC.

  3. Bill Powers’s concept of the “neural current” is the summed rate of firing of a “neural bundle” of individual neurons, each firing at a rate determined by its current and recent chemical environment and by the firings of other neurons that communicate through thousands of synapses, probably identical in no pair of neurons. The “neural current” is a virtual quantity, just like the virtual currents in the virtual control loop of a GVC. Below, I point out that averaging a stochastic variable over time or over enough such variables can allow an adequate approximation to a continuous variable, as Bill does for the “neural current” that he initially assumed might allow approximations to within 10% when modelling tracking studies. The experimental results using the virtual “neural current” variable throughout the brain part of a control loop have been much better than that.

  4. It is worth emphasizing that it is very likely that few pairs of “members” (contributors to) a GVC are controlling for the same perceptual variable at the same reference value as each other even when they use the same words to describe the perception they are controlling. What counts as “civility” in a Spanish face-to-face conversation would be “rudeness”, perhaps intolerable “aggression” in an American upper-class “civil” conversation. (I have no idea what Rick intends me to understand when I, or anyone else, reads “comity”, but I deduce from the usual semantic domain of prefix “com-” that it has something to do with togetherness).

  5. Collective control must be distinguished from collective side-effects. Climate change is due to the combined side-effects of all the perceptions controlled by all the people, animals, vegetation, sea creatures and microbiota in the world, very few of whom act to control their perception of any aspect of climate change. Climate change is a collective side-effect. that changes the local environment perceived by every living thing.

  6. Collective control must be distinguished from “collective effect” of either kind, direct (intended) effect or side-effect of action in controlling something. Collective effect can be observed directly, whereas collective control usually should be tested using the TCV.

  7. Collective effect, and therefore collective control, may be distinguished into two main types: continuous and stochastic. Kent’s 1993 demonstration was of the continuous kind. Collective effect of a “neural current” is stochastic, being based on the individual firings of nerves. All control that uses nerve firings rather than such other variables as hydraulic pressure, chemical concentration, and so forth, is at base stochastic, but by averaging over time, all stochastic effects can be treated approximately as though they were continuous, as is done in PCT analyses using neural currents.

  8. Whether stochastic effects are seen as continuous depends on the time-scale of perception of the effects of single events. For example, evolution is based on the birth and death of individual living control systems, but happens (usually) at a rate and scale that allows enough averaging to smear all these individual events into a quasi-continuous rate of species growth and decline.

These are just a few statements about how I understand collective control. I might say a lot more if we were to talk about interactions among different collective controllers. In fact, I do in the book I am writing: “Powers of Perceptual Control: An enquiry into language, culture, power, and politics.”


I didn’t, and still don’t, think I had to get hypotheses approved. And I did the research to test the hypothesis , not to prove it wrong. I was actually surprised by the result.

I think it can, should and does. I’m sorry I perturbed you.

Best, Rick

Yes, the apparent failure of the prediction (that the size of cities would be inversely related to # of police/capita) could be due to variations in the proportion of people in each city who are not controlling for civility and comity. I think my suggestion of using crime rate as a measure of those not controlling for civility or comity (or controlling for that variable at a very low reference level) might be a way of factoring those people out to get a better indication of how the size of the collective might influence the need for police.

I think all your suggestions about why I got the results I did (finding that police/capita increases with city size) are just fine. But I think that, rather than just pointing out that these concerns mean that the results cannot be taken as clear rejection of the model, it would be more productive to think of ways to test the model in a way that eliminates these concerns so that one can have more confidence that the model is a reasonable approximation to the mechanisms that produced the observed result.

This all started with your prediction that “as more of us participate in collective control of civility and comity here, no individual will have a ‘police’ role imposed on him or her.” I presumed that you made this prediction based on your understanding of Kent’s “collective control” model. I understand Kent’s model to make that prediction as well – at least in terms of the stability of the virtual controlled variable: the more agents in the collective who are controlling the same or a similar variable (relative to different references), the greater the stability of that variable.

I don’t recall seeing any empirical tests of Kent’s “collective control” model and your prediction suggested to me one possible way of testing it. Since you didn’t think much of my test I think it would help if you suggested a better way to test the model. Or, better yet, point me to the tests of the model that have been done that have led you to have confidence that the model correctly predicts that “as more of us participate in collective control of civility and comity here, no individual will have a ‘police’ role imposed on him or her”.

Best, Rick

Less perturbed now, still fascinated!

I appealed for participants in this discussion forum to engage actively in resisting disturbances to collectively controlled variables that I categorized as ‘civility’ and ‘comity’.

That they are disturbances to many participants is shown by the objections that participants have made over an extended period to what we have referred to with words like abrasive, ad hominem, insult, personal attack, snarky, dismissive, belligerent, and so on. These words, and the category words civility and comity, refer to perceptions about how people communicate. Over the past three decades we have seen a number of people control recurrent disturbances to these perceptions by leaving CSGnet and not returning. For this reason, and others, disturbances to control of ‘civility’ and ‘comity’ perceptions are also disturbances to control of perceptions involved in the institutionalization and success of PCT, which we have explicitly agreed are perceptions that we collectively control.

‘Comity’ is a less familiar word.

comity /kom"i tee/, n., pl. comities. 1. mutual courtesy; civility. 2. Also called comity of nations. courtesy between nations, as in respect shown by one country for the laws, judicial decisions, and institutions of another. [1535-45; < L comitas, equiv. to com(is) affable + -itas -ITY]
Usage of 'comity' in diplomacy and political negotiation associates perceptions of negotiating disagreements. That's what I had in mind.

The analogy to the social institution of a police department, and its paramilitary organization, is barely more than a metaphor. There are many, many variables involved in such institutions that are not relevant to this Discourse environment. That’s why I put the word “police” in scare quotes. The point I aimed to make was that we should have no analog of a police department here. I asked for each of us to participate in resisting disturbances of this sort.

When one person does it, it’s unpleasant business and raises questions of who are you to tell me what to do, and by what authority. When we all do, it an affirmation of this is how we conduct ourselves here, don’t be a jerk.

We could talk about the institution of police departments as (conflictually) collectively controlled perceptions and what variables they are instituted to control, but that is a different topic or set of topics.

Which is all well and good. But you also said that “as more people participate in collective control of comity and civility, no individual will have a police role imposed on him or her”. Whether you meant it to be or not, I saw this as a prediction of Kent’s collective model – the prediction being an inverse relationship between the size of a collective and the amount of policing in that collective – and could be tested by operationally defining the size of a collective as the size a city and operationally defining the amount of policing in those cities as the number of police per capita.

Whenever you test a model you also have to make some assumptions about the situation; in this case my main assumption was that everyone in a collective (city) is controlling for a different reference amount of civility and comity; some want a lot of it (those seen as civil and considerate) while others want very little of it (the uncivil and inconsiderate).

I can think of several legitimate objections to this proposed test; for example, maybe collective control theory doesn’t predict an inverse relationship between the size of a collective and the amount of policing in that collective; if it does make that prediction, then maybe city size and police per capita are poor measures of the size of the collective and/or amount of policing in it.

The test I Proposed was aimed at showing how one might go about doing a quantitative comparison of the predictions of the collective control model to actual data. As far as I know there have been no such tests, so right now collective control theory doesn’t seem to be an explanation of any data that needs to be explained.

I would much rather talk about how we could test a model of police departments as “(conflictually) collectively controlled perceptions” (if that’s what you think they are), rather than just talk about them. What excited me most when I discovered Powers’ application of control theory to the behavior of living things was not the theory per se but how well the theory, in the form of quantitative models, fit the data; and what those models then revealed about the nature of behavior. If we’re going to expand on the basic PCT model – which I think is what Kent’s collective control model is aimed at doing; expanding the basic model to account for the behavior of collectives of individuals – I think it’s important to show, quantitatively, why the expansion is needed and how well it accounts for the behavior it is designed to account for.

And why the parentheses around “conflictually”, by the way. As I understand it, conflict is the very basis of Kent’s model of collective control, which, I presume, is the model you are talking about when you talk about “collective control”. In Kent’s model, conflict between control systems controlling the same variable relative to different reference specifications results in that variable being kept a stable virtual reference state. I think Kent (or perhaps it was Martin) also showed that the stability of the virtually controlled variable increases with the number of systems in conflict (the size of the collective). That’s why I thought your comment was a prediction based on Kent’s “collective control” theory; more stable virtually controlled variables means a more stable society, which I believe would be one that would require less policing.

Best, Rick

Your understanding is incorrect. If you read Kent’s chapter in the Handbook, you see for example on p. 242:

The important point to remember here is that collective control can result in stabilization of some features of shared environments whether or not conflict is involved.
The environmental feedback paths between and among participants in a public are rather more involved than those in a tug of war. More fully on p. 271:
Collective control stabilizes the physical environment in which the collective control takes place. A stabilized environment facilitates the control of some perceptions while making other perceptions more difficult to control. This restriction in the range of perceptions easily controlled within a stabilized physical environments increases the probability that individuals in that environment will control similar kinds of perceptions and, because of the neural process of reorganization, will develop associative memories that make for similar kinds of meaningful connections between their perceptions. A shared cultural environment emerges from these similarities in meaningful connections between perceptions. These cultural forms, such as language and protocols, are stabilized by giant virtual controllers, which of course are a variation of collective control. Stabilized cultural forms can facilitate the coordination of people’s interactions, which helps to make their collective control activities more cooperative than conflictive. Effectively coordinated collective actions are necessary for maintaining physical environments in stable forms. In short, people stabilize both their physical and their cultural environments by means of collective control, and these stabilized physical and cultural environments allow for the meaningful social interactions by which people maintain stable social structures over time. These three explanatory mechanisms can account for stability in social structures of all types, from the smallest to the largest.

I agree that tests of the model are important, but first they must be tests of the model and not tests of something kinda sorta like the model.

But even more important is models that explain phenomena. A model of a phenomenon or of phenomena, as you know from long experience, is a test of the model. So instead of working to challenge the model of collective control (or something kinda sorta like the model) why not go for a twofer: model some phenomena that involve people who live together in a public, and you get a test of the model into the bargain. First step, as with individual control, is to identify collectively controlled variables. Kent has some examples. I gave you some language data from my home Island. Bill Labov pointed you to people who are modeling such data in terms of autonomous agents. But lots of Kent’s examples are likely to be more tractable. Some useful organizing principles have been identified that should help with identifying variables and designing an investigation, such as the protocols of getting help, giving help, teaching, and learning, and the subtypes such as trading, contracting, etc.

These various types of protocols share a common framework: to begin the interaction, one party disturbs a perception controlled by the other party in order to indicate that the first party seeks to control some perception with the assistance of the other. The second party then continues the interaction (or not), by acting so as to reduce the first party’s perceptual error (or not). The interaction ends with first party acknowledging the second party’s actions in a way that cancels the disturbance that initiated the interaction. Protocols are frequently more complicated than this basic outline, involving many back-and-forth and simultaneous moves, as well as attempts by one party or the other to repair any errors of communication that may occur. (p.246)

Hi Bruce

Perhaps conflict is not the basis of Kent’s “collective control” model but it is certainly a very important part of it. The basis of Kent’s collective control model is that “stability” arises when many people control the same environmental variable, whether all those people have the same reference for the variable or not. Kent says (p.241):


Clearly, the idea here is that the more people there are controlling a variable relative to the same reference the more stable that variable will be. This is what would happen sans conflict since everyone is assumed to have the same reference for the state of the variable, an assumption that becomes more and more implausible as the number of people involved increases. Nevertheless, Kent’s statement here is probably where I got the idea his model predicts increased stability of social variables with increased size of the collective that is controlling those variables.

Conflict comes in as an important part of the model because Kent notes that the great stability that results when multiple people are controlling the same variable relative to the same reference still occurs even when all these people have different references for the variable. Here’s what Kent says about it on p. 241:

So increased stability results from multiple controllers controlling a variable even when those controllers are in conflict! So conflict in a collective is not a good thing but it’s not a bad thing either!

All this is, indeed, true of what would happen to the stability of a variable if it were controlled by multiple controllers. It can be easily demonstrated by simulation. But this model is not introduced as the means of explaining some specific (and common) social phenomenon. Indeed, I can’t think of a social situation, other than a multi-sided tug of war, where multiple people control a common variable relative to the same or variable references.

And even more important than that, the model should have been developed to account for some actual social phenomenon in the first place. By I think my proposed test was based on an understanding of the model that is a lot more accurate than “kinda sorta”. My model is just PCT, after all.

Yes, and that’s the problem; I have no idea what actual social phenomenon the model explains.

That’s exactly what I was doing. And I think I did a reasonable. But if you didn’t like what I did, how about showing me how I should have done it.

I don’t want to search through the whole paper for them. PLease tell me what collectively controlled variables Kent identified and how the model maps to the control of those variables. That would help me understand the model a lot better than just a bunch of talk describing the model.

Yes, and thanks for the data! I didn’t go to the existing modeling but developed a PCT model of my own and describe the results of that effort on pp.108-110 of the Social Control chapter of The Study of Living Control Systems. I recommend the whole chapter – and book – to you. The best way I have found to understand PCT is to start with a description of the phenomenon you want to explain and then develop a PCT model to account for that phenomenon. Remember, phenomena phirst!!

Best, Rick

Hi Rick,

I think it is a nice idea to utilize statistical data to develop and test PCT models of collective control. However, before that can be done the concepts and models must be in order and it would be wise to utilize proper sociological literature for that. I found it really very, very strange to connect together “civility”, policemen and sizes of the cities! I believe there is very little connection between people’s control for civility and the amount of police corps. Instead I would offer one other possible direction of explanation from the classical sociological thinking which should be also testable, but of course not so simply.

Let’s first assume that all people control for many different perceptions. Because of our common biological and cultural backgrounds there is some similarity between different persons: we control for nutrition, warmth, medicine, moving between places, social belonging, interesting activity, self-respect etc. which are traditionally called needs.

Secondly, all control requires some external resources and their availability is not equally distributed. Traditionally here is talked about economical inequality. In global scale it has increased very highly and I would think that it tends to be higher in bigger cities than in smaller ones. There could be some statistical data available about it.

If there is strong inequality of resources between people, we could assume somewhat conflictual collective control among them. Those who have much resources tend to try to (and can, because of their resources) secure the continuation of their possibilities to control by utilizing legislation and police forces. Those who have little resources tend to try to control their perceptions using illegal resources, by committing crimes. This conflictual collective control stabilizes both the police ratio and crime ratio to a high level, higher in the bigger cities than smaller ones.

As for civility and policing in the discussion forum, I am afraid that those who control for high civility will rather delegate the control of rule obedience to some authoritative person who must do the dirty work. Thank you Bruce.

You are predicting a positive relationship between inequality in the collective and both the police and crime ratio in the collective. To make this prediction you had to assume that inequality increases conflictual collective control and that increased conflictive collective control stabilizes “police ratio and crime ratio to a high level”. Neither of these assumptions are part of Kent’s collective control model. So while I think your predictions are reasonable, they are not derived from the collective control model; that model says nothing about how the degree of conflictive collective control relates to the degree to which a collectively controlled variable is stabilized.

My prediction of a negative relationship between city size and police ratio was based on an explicit component of the collective control model, which is that the stability of a collectively controlled variable increases with the size of the collective controlling that variable, whether the collective control is conflictive or not.

The fact that the prediction of the model failed – I found that there is actually a positive relationship between city size and police ratio – could be because some assumption I made is wrong. The way to demonstrate that this is the case would be to show how the collective control model can account for the data with the appropriate assumptions.

Best, Rick

OK, collective control model (CCM) says that

[New trial]

OK, collective control model (CCM) says that

  1. a collectively controlled environmental variable (CCEV) stabilizes to a value which is some kind of (gain weighted) average of the individual reference states, and
  2. the gain of the collective control is (some kind of) sum of the individual gains and so the bigger the collective the stronger the stability of CCEV.


Your (Rick’s) application of CCM says that because civility is a CCEV, then there should be lower police ratio in bigger collectives, right? Could explicate the logic: what is the connection with the stability of civility and the police ratio? For me it is somewhat unclear how this follows from the CCM.

For me it seems that there is no connection with the stability of the civility and police ratio. Rather there could be a negative relationship between the value of civility and high police ratio. (Because low value of civility can mean a higher crime ratio. Coarsely the criminal law defines a certain level of civility - or morality - which is required from all citizens. If your action is below that then you break the law. If your action is above that then it is OK, but it does need to be polite or friendly or any way appraisable - high civility.)

T. Eetu