Wednesday, October 05, 2005

Fundamental Problem with Sociology

Let me point out a fundamental issue that we have with historical analysis. I'm not speaking of the tendency to either underrepresent historical forces or to behave in cultural imperialism by looking at other cultures and ascribing to them not just intrinsic but also deterministic accounts of "sadism", "masochism", "laziness", etc. as if culture was an entity that made every person behave the exact same rather than being more of an permeable web. Rather, the difficulty has to do with the markovian process of society.

"Markovian process" is most commonly applied to evolution. It describes a system where each phase is determined in part by the previous system. In evolution, a stage where single-cellular life is dominant does not turn into birds. Further, random chance plays a major role. The moths in England that became black-winged did so because of the coal burning in industrial England, something they had no role in. There are theories that amphibians became predominant thanks to a random event, perhaps a prolonged drought that would make access to land more strategic.

How does this apply to society? It is very easy, looking at history, to say that hierarchy, domination, etc. is permanent, indeed even human nature. The many problems with this are obvious: There's no intrinsic evidence (say, genetic) to establish it; we all know from personal experience and cultural analysis that human can live under a great amount of institutions and cultures; the institutions we have precisely ENCOURAGE destructive, anti-social, etc. behavior, roles, decisions, and outlooks; etc. Further, to establish the point, the social evidence is necessarily restrictive: we only can get insight into about 10,000 years of the 100,000 or s years of human history, so we have no idea the cultures and the institutions that have passed away. But let's say that all of our history has indeed been hierarchical. Must this be because of human nature?

Suppose that one hierarchical group, maybe thanks to local resources, or a particularly large group, or some utilization of hierarchy, managed to gain success. Others followed. Given that quite a bit of modern civilization developed in Sumer and the Middle East/Africa more generally, this conclusion doesn't seem so limited. Also note that technology can change things. For a long time, nomadic groups would devastate hierarchical and stable groups.

The point is that we can't be sure about what institutions are best, or can be supported by humans, or what not, given that not all institutions are going to have been given a fair shake, that our understanding of history and the evidence we have available is contoured by that very hierarchy that self-selects future societies, and that one could imagine many living arrangements that humans might be able to support.


Anonymous bwong said...

"Markovian process" is most commonly applied to evolution. It describes a system where each phase is determined in part by the previous system. "

Actually, Markov process applies to a system that evolves in DISCRETE time steps in such a way that its current state depends ONLY on the immediate previous state,--the state it was in one time step ago.

The system does not retain any memory of earlier history.

It definitely does NOT apply to historical or biological evolution.

I prefer to continue on znet, well, until the 1200 and 1500 rules get too irritating. I hope the high tech relay back and forth consultations are not subjected to these rules. :)

4:21 PM  
Blogger Frederic Christie said...

From what I've seen, a lot of good theoretical work uses Markov processes or chains to describe natural selection: Greg Graffin, Bad Religion's lead singer (and a Ph. D!) uses the term to describe both evolution and, to a more imprecise extent, culture.

History, of course, is a worse analogy, because a lot of decisions are made by history that is poorly remembered and millenia old.

I hope the point did remain clear: Within a reasonable range, chance deviations could easily have underestimated human potentialities and institutions that could have been chosen.

5:25 PM  
Anonymous bwong said...

I am aware of works similar to
the link you posted.

But they are not realsitic models and aren't intended to be.

According to Darwin evolution is the result of natural selection acting on mutations. This is generally correct according to most biologists.

But it seems obvious that there are also other factors at work. Statistical drift being one.

The Markov model is a "tinker toy" model designed to explore the role of stochastic processes in evolution.

This doesn't mean stochastic processes alone is sufficient in any realistic model of evolution. The model just isolates one mechanism to study.

There are different kinds of stochastic processes.

The Markov process is only one. It is very unrealistic for the reasons I mentioned. However it is mathematically the easiest to study. That's why it is picked.

"Within a reasonable range, chance deviations could easily have underestimated human potentialities and institutions that could have been chosen"

Even chance deviations have a structure(otherwise there would be no point in studying statistics)

My point is, allowing chance variations, the standard deviation of human ability to devise social structures is quite small.

Social structures don't arise out of thin air. They are ways to solve certain problems.They are a form of technology if you like.

It seems that when confronted by similar problems humans always devise similar solutions.

Of cause cross culturual comparison is meaningful only when the cultures being compared have broadly similar material conditions.
Otherwise we cannot presume that they face similar challenges.

Also the cultures being compared should have developed in relative isolation in order to rule out the possibility that they might have adopted technologies(and social forms) from each other.

5:31 PM  
Blogger Frederic Christie said...

Right. My point was to describe how one of the forces that form society, and species, are random chance forces (in the latter, things like weather events or climate patterns). Those random chance forces may make us think that there is an intrinsic move towards, say, complexity in biologies, when that was simply an accident of history. Some French philosophers such as Bergman made this mistake.

Chaos theory is telling us a lot about whether or not chance deviations have a structure, actually. But what statistics tell us is that the small is unpredictable but the aggregate is predictable, thanks to mathematical tricks. The problem is that you have to absolutely random samples, and that is almost impossible.

So one dice roll is 1-6, but if you take the sum roll of several thousand dice rolls you get a mathematical expectation of 3.5. But that is a separate mathematical entity from each individual dice roll.

Given all the infinitely complex variables in history, including all the non-human variables, it seems reasonable to me to recognize that there are all sorts of institutions that have not been tried yet but are perfectly feasible. Though I hate to make an error Hume would disapprove of, we can see this lesson taught in history: many times we believed democracy, or capitalism, or similar to be impossible.

I really think that, even if you just look at culture, you see an incredible variety of behavior. For example, the Nacirema.

Of course social structures don't arrive out of thin air. But that doesn't mean every possible social structure has been tried. Even if we use your technology analogy, which I find begs some questions, we haven't tried all possible tools and the tools we have built have been almost intrinsically designed for certain ends, such as hierarchy. Doesn't mean others can't be done.

I think that we have to separate out how most of the times we see similar problems and similar solutions we're also seeing similar institutions at work. But I don't think that's in fact right. All sorts of alternatives have been tried across all sorts of spectra.

You add in a similar difficulty, an almost intractable one.

5:47 PM  
Blogger Frederic Christie said...

Your argument that of course different cultures must be measured at similar times in development is slick, and has some truth, but still doesn't hold up the conclusion. Even comparing cultures across modern industrial societies with similar levels of prosperity, we see quite different cultures. Compare Japan and Sweden, say.

For example: Tribal superstition is, of course, fairly consistent. But the forms it takes are quite wide, thanks to all sorts of variables that may or may not be chance-related that we can't identify, and those forms can be quite relevant. It seems to me that Native American mythologies are far more likely to accept natural selection, especially given the genius they showed in breeding, than Christian ones, which posit a discrete starting point with everything coming into play as is.

12:43 AM  
Anonymous bwong said...

"Chaos theory is telling us a lot about whether or not chance deviations have a structure, actually."

Actually Chaos theory is about intractability of complex systems rather than chance deviations.

Even more interesting, chaos theory presumes determinism,--i.e chance is impossible.

Chaotic behaviour only occurs in "classical mechanics".The world is deterministic like a clockwork in this framework.

On the other hand, quantum mechanics incorporates chance and probability in a fundamental way. It can be shown that chaos is impossible for quantum systems.

This is actually a very deep paradox
in physics. It has all the flavour of dialectics: Chaos is a feature of determinism,--or so it seems based on our current understanding.

"But what statistics tell us is that the small is unpredictable but the aggregate is predictable, thanks to mathematical tricks."

Correct. Another instance of dialectics.

"The problem is that you have to absolutely random samples, and that is almost impossible."

Actually, a deeper question is what is randomness?

Probabilists have many definitions of "randomness" but they are all inadequate in some ways (There is a classic by Richard Von Mise for the philosphically inclined but mathematical unsophisticated which discussses some of the issues.But Von Mise has been critiqued as well)

In practise statisticians use some mathematical models to capture the notion of chance as best as they can by basically saying that
any reasonable definition of random behaviour should have such and such properties.

So they basically define "chance" operationally using a set of empirical criteria. But it is easy to see phiolsophically this is unsatisfactory.Moreover there are non chance phenomena that pass all the tests for randomness, the most famous being the decimal expansion of "pi". It is a random sequence according to all statistical tests, but of course it is not really "random".

But what is
"randomness" in sociology?

In social statistics it is not even possible to accertain whether what is commonly attributed to "chance fluctuations" conform to statistican's empirical criteria. The mathematical models almost always don't apply.

In social statistics, "random flutuations" is basically just a short hand for "caues unknown".

Then you may ask, since we don't know whether "randomness" in the social sciences truly corresponds to the mathematical notion,--in fact I would claim that we know they never correspond,--how can we even apply the mathematical model?

Indeed we can't if you are a philosopher.

My understanding is that the social scientists basically pretend as if the mathematical models apply and push the envelope until they get some really fuck up results and then they go back to revise their theories.

One reason that this procedure "works" to an extent is because the tools for measurement in the social sciences are too blunt.
Deviations from the mathematical models just don't get registared.

Hence in social statistics they have this wonderful term "robustness". That means you can go ahead to apply a lot of mathematical tools even when the hypotheses of the mathematics are violated!

10:47 AM  
Blogger Frederic Christie said...

"Actually Chaos theory is about intractability of complex systems rather than chance deviations."

Chaotic systems are defined as systems wherein small changes in the initial state have large changes at the end, no matter the degree to which you can enumerate their principles. The common example is the two bowls. One is turned up. You drop in a ball, it will go to the bottom. The other is upside down. You put a ball on top, it can go any direction. The latter variation is due in no small part to chance, despite the fact that you theoretically know every equation.

When combining chaos theory and quantum analysis, among other trends in science, we see that in fact science is telling us a lot about how even Laplacean omniscience may not be enough to understand what we want. Chomsky makes this very point to B.F. Skinner, rebutting Skinner's simplistic behaviorism/mechanistic philosophy.

(Interestingly enough, your point about dialectics is commonly made in the literature in almost those terms: chaos falls into order and falls into chaos again).

The point about the definition of randomness is, of course, well taken, but even if we avoid philosophical nitpicking, what everyone can recognize as "random" can be very hard to generate. Many studies scrupulously picked random samples only to find that they missed an important obfuscating variable: say, that they were calling people who, umm, had phones, which skewed the data. The reason why so many good researchers can turn out data that seems so diametrically opposed has to do not just with what they're surveying and their personal issues/predilections but also their methodological and theoretical assumptions, the process with which they draw their conclusions, etc. etc. Statistics are often implied to be panaceas by overzealous statistics teachers, but the amount of things they can reasonably measure is actually quite limited.

In line with your point about the proceeding from philosophically limited concepts of chance, randomness, etc., we have the problem that appears even in very good science, let alone often crappy social science, that the conclusion is basically predetermined given the assumptions. Gould's proposal about punctuated equilibrium could have come earlier if scientists hadn't just used the lack of an interspecies shift as proof that one was there. It took a young student to say, "Hold on, there's something wrong here. Evolution must occur in quick bursts. Only that explains what we're seeing in the fossil record".

Good social science work is always far more nuanced. I think if you look at the way I proceed in my blog I regularly discuss mitigating factors, indeed factors that are quite plausible and run the opposite direction, for sociological theories. When you see someone saying "Blacks in this country have a GDP larger than most nations in the world", you know they're playing sociological poker, not being real honest advocates. The way they look for the conclusion reveals the game (what is "black GDP?" GDP is something that accrues only to nations that we can define, moron).

To be fair, my understanding of robustness is a little less harsh than that. You do a lot of very complex comparison of bell curves and analyses of data and standard deviations and such and you find models that can tolerate, within a certain degree of accuracy, that there is a trend at work here that can be identified and sort of quantified within a margin of error. While it is technically true that robustness deals with the degree that the model can violate its own conclusions, that makes a lot of sense from a non-mathematical standpoint. You may notice that I regularly use "Even if" conditional statements. The sign of someone who has a good argument and cares about the issue and actually wants to enter into a dialogue with someone else is that they assume the world where their claims are far less compelling and where their opponents' are far more and proceed (somewhat like a negative proof, also often used in sociological work). When the problem comes is when we pretend that the latest poll is ironclad and indicates the undeviating opinion of the American people. If you actually look at polls cited by, say, conservative scholars, very interesting counter-interpretations almost immediately arise. I use TIME polls to prove that the majority were opposed to the war both on Constitutional and international law grounds, yet the original TIME issue talks about how there is a majority who SUPPORT the war. Add into uncertain models hacks and bad journalists who then selectively prove their theory through well-chosen interviews (do they run the interviews that disprove the point?) and who use Newspeak to prove 1 + 1 = 11 and these concerns become quite authentic.

11:40 AM  
Blogger Frederic Christie said...

I also have to add that even the best statistics mean nothing unless given analytical context.

A professor of mine last year did a study of what he called the "Cheney hypothesis": rightist governments deal with more terror than liberal ones. And his conclusion (drawn from a terror database that I would be very skeptical of, of course) was that indeed this is true. But he then said, "Hold on. Rightists often get temporary drops in terror by using preponderant force, but they permanently sabotage the peace that way. Of course extremists on both sides will try to stop those more likely to get lasting peace, but that's not relevant." And he also admitted upon my questioning that we have a problem of timeframe: If terror goes up in 2004, could it be residual effects of Clinton, Bush I or Reagan?

10:57 AM  

Post a Comment

<< Home