Comparing Group Levels - Thinking Out Loud

In the spirit of the new year, I suppose it's about time to sum up some of the more rigorous approaches to group (level) formation. Here are the major disciplines I've been studying:

• Sociology via Keith Sawyer
• Economics via List & Pettit
• Biology via Okasha & Wilson

and

• Social Psychology via Haidt
• Science of Religion via Norenzayan & Atran
• Cultural Evolution via Turchin

Caveat:  This post is mostly a chance for me to work out-loud through the uncertain process of categorization.  My intent is to see if there is any way to piece together insights from the various traditions in order to

To start the thinking process, here's a cursory summary of the various approaches to multiple-group levels.

	Process (Interaction) Sociology	Structural Sociology	Rational Choice Economics	Multi-Level Selection Biology
Time scale	varies (duration determines group level)	medium (long enough for structure to solidify & produce readily observable feedback effects)	short (a couple of judgments)	generations
Measure	observation of communication artifacts	structure's level of control over individual	group agent  rationality & supercedence of indiv. rationality	fitness
Loose Philo tradition	communication theory & process philosophy	social constructivism	rational reductionism	positivism?
Determinant of Group level	duration of group or artefacts produced by interaction within group	causative power dynamics (with top-down bias)	supercedence of rationality to higher level (causative power dynamics)	covariant fitness

Based on this summary, how one deals with the nesting of groups seems like an interesting line of investigation.  Here's my first crack at categorizing nesting approaches.

GROUP NESTING METHODS

• By Group Size - While it might seem obvious that a higher level group necessarily has more members than a lower level one, this doesn't have to be the case.  Take for instance an authoritarian government.  It might control a very large population, but outside of a few sycophants there may be few agents who identify as part of the group. However, there may be many people who consider themselves part of the same nation's social institution, say education, a social movement or the nation's soccer fanatics.  These groups may be subservient to the larger group, but contain more members than the large group. Thus group size nesting is at odds with power relation nesting.
• By Control - Control is easiest to visualize and model when it acts on adjacent group levels.  However, in practice, feedback is likely to be complicated and occur between non-adjacent levels. Multi-level selection theory certainly accepts this postulate.  Utility of non-adjacent group interaction is determined by fitness covariance. This measurement oriented approach avoids some of the conundrums that comes with sociology's functional oriented approaches.
• Control by threshold level - if control from a group passes a certain threshold (probably low)  then a higher group is needed.  This type of control creates possible nesting problems.  A low level group has the theoretical possibility of controlling a high level group.  For instance, a particular team of soccer fanatics may theoretically control an authoritarian government even though they may be nested a couple of levels below the government (group of fans -> a team's fans -> soccer team -> sports body -> government)
• Control by net level of feedback - the level that has the most control supercedes the other.  This sets the stage for chicken & egg priority problems - whose feedback comes first.  I suspect this is part of the intractable micro vs. macro paradigm problem
• By Emergence - a higher level group is produced whenever emergent properties can't be be reduced down to the level of its agents. Complicated, non-adjacent nesting problems are possible. For instance, what happens when you have an emergent group entity functioning as part of another group of individual people?  Minimizing this would seem to require assuming groups rarely act as agents within the same level as individuals or pushing group effects back onto individuals.  The former, though plausible, seems rather arbitrary.  The latter takes you back to the reductionist problems of methodological individualism.

Perhaps a better way to categorize nesting methods is as follows:

• principle based methods - control, group size, other subjective lenses, & perhaps emergence
• covariance methods - fitness, & perhaps some narrow quantifiable subjective measures
• property methods - emergence 

BEGGING THE NESTING QUESTION

However this still begs the question of how to minimize overlapping within nested groups.  

Principle based methods seem to rely on the quality of used lenses.  Lens quality seems based on a hope that qualitative research can reveal accurate descriptors and researcher bias can be minimized by rigorous reflexive practice.  That type of stamp collecting has never interested me much. Bootstrapping doesn't wiggle my toes.

I like covariant methods.  They seem able to reveal the extent to which overlapping nests are or are not a factor.  They also seem better at minimizing subjective bias than principle methods . The timeframe of analysis seems problematic though.  How do you measure fitness for human groups that may produce counter-fitness results on a short time scale (say rich people having few kids), but increase fitness over long time scales (say rich people after a couple famines)?

Property methods seem intriguing.  However, emergent properties seem problematic to deduce.  List & Pettit have probably done the most work here (that I know of).  Their approach is based on deducing group-agent rationality. However, their rationality test seems overly cognitive.

Another way to minimize nesting overlaps is to only include certain types of groups.  List & Pettit have done this with their rational actor approach: only rational groups count.  Multi-level selection has indirectly done this due to the timescales inherit in its modelling.  (One gets around the timescale issue a bit by being able to look at whether genes present in some populations and not others correlate with historical fitness differences.)  Similarly, Sawyer suggest structural sociologists have created very narrow definitions for their group levels. However, by whittling down the focus on individual interaction they have challenges explaining the emergence of the structure they take for granted (Sawyer, 2013): Structure is what is socially stable.

Sawyer's sociological Emergence paradigm broadens standard sociological group levels.  He adds a couple of group levels based upon time stability.  This is done from a communicative tradition.  It includes the following levels:

1. Structure - written texts like laws, infrastructure
2. Stable emergents - group subcultures, group slang, collective memory, shared social practices
3. Ephemeral emergents - topic, relative roles, status assignments
4. Interaction - discourse patterns, collaboration
5. Individual - personality, cognitive processes, agency

Unfortunately he isn't explicit in mentioning the role of time.  Ephemeral's differ from stables only in terms of their survivability.  This backward looking approach isn't ideal.  What non-arbitrary and non-subjective measures could you use to accurately distinguish ephemerals from stables?  Subjective rates of change?  Arbitrary time scales that disregard context?  Subjective time scales that are sensitive to context? Having some characteristics of what likely correlates with survivability would be nice.

For instance, the science of religion via Atran has postulated a number of markers associated with stable group functioning.  If one trusts Norenzayan's conclusion that religion is an evolutionary adaptive group structure, then one can start to pick away at the markers correlated with adaptive groups at any level.  This mirrors List & Pettit's rational approach.  However, instead of just postulating two markers:

1. rationality via truth orientation & independent judgment, and 
2. group decision supravenience (i.e group control over the individual),

one would posit several markers such as:

1. freeloader detection & punishment
2. costly commitment displays
3. moral "Big Brother"
4. deference to group norms
5. sacred values & norms
6. assessment of individuals based upon their inferred intents
7. quasi-propositional truths (rationalization of cognitive dissonance)
8. etc.

The marker approach may represent another way of looking at emergents.  Instead of taking the standard approach by defining emergence as that which produces something novel from its constituents, the marker approach stipulates which emergent properties are acceptable (based upon their likeliness).  However, this greatly reduces scalability and limits the approach to a certain region of group levels.

TIME CRUSHES ALL

While I'm partial to the science of religion's marker approach, how do you measure any of that!  What seems to be coming out of this open-thinking process is the glossed over role of time.

From Sawyer (2013),

Most sociological discussion of emergence have focused on the broader macrostructures that emerge and how those emergent patterns constrain future interaction.  Yet these studies have not had much success in tracing the exact details of the moment-to-moment emergence processes whereby macrostructures are collectively created.  In contrast, the Interaction Paradigm has focused exactly on the moment-to-moment details of how ephemeral emergents result from interaction.  However, in shifting their focus to interactional process, they have tended to neglect the nature of what emerges and of what perdures across repeated encounters. (pp. 215)

Biology indirectly tackles the time issue with its adoption of a non-socially constructed fitness measure.  This enables covariant study. Instead of finding the historical record associated with a certain gene, you can just investigate populations that do and don't have it, looking at their fitness over time. Thus you get control groups.

FORMING A CONCLUSION

The following points seem to emerge from this discussion:

• narrow down the definition of "group" which is used to produce group-levels
• instead of using arbitrary definitions to limit "groups", use an objective measure
• pick a workable time frame for that measure's collection
• pick a measure that naturally facilitates finding control groups
• pick a measure whose collection is binary (like fitness)
• ensure the measure scales up and down
• I like science of religion markers

So, to me, it boils down to finding the most objective measure possible that catches the insights gleaned from the science of religion marker analyses.  Why?  Because I assume very wise people, especially those in sociology, have already thought hard about the standard ways of putting these points together.

OPTIONS

Options seem to be:

• create an aggregate measure of "adaptiveness" based upon previously observed markers
• follow a hybrid social interactionist path and, in terms of group levels, just talk about increasing, decreasing or very stable structures (on an arbitrarily long time)
• keep plugging away to find a novel measure construct.