I devoted yesterday afternoon and most of today to (finally) producing the updated documentation for Version 3 of the ForagerNet3_Demography model, one of the agent-based models that I've been working with. You can read all about it on this page, and you can even download the raw code if you like. Files and citation information for the model are also available on OpenABM.org.

I used a version of this model (implemented in Repast J rather than Repast Simphony) in a recent paper published in Uncertainty and Sensitivity Analysis in Archaeological Computational Modeling (edited by Marieka Brouwer Burg, Hans Peeters, and William Lovis).

Over the summer I used to model to generate data for a paper on the minimum viable population (MVP) size of human groups. That's in the editing stages now -- hopefully I'll be able to get it done and submitted somewhere before the Christmas break.

​Onward.

I've been hard at work rolling a boulder up the Repast Simphony learning curve.  Computer modeling is a basic element of my three-headed "Shovels, Collections, and Code" research agenda. The other two are on track: I'm planning future survey/excavations at a natural levee system that appears to contain buried Archaic components, and I've started my collections work with an ambitious data-gathering effort oriented toward understanding the Kirk Horizon.  The computer modeling part of my work is an important part of building an interpretive framework that allows us to integrate the small-scale behaviors we can document at individual sites with the large-scale patterns we can describe through pan-regional collections work.  More on that later.

Call me crazy, but I find writing and debugging computer code to be relaxing.  It can be frustrating, of course, when you can't figure out the source of some problem or error, but overall the process of building and tuning a model is engaging and strangely soothing. The parts make sense, represent something, and work together. And there are rewards for elegant design. It's fun. During my dissertation work I sometimes had the luxury of taking full days (not 9-to-5 days, but 24 hour days) to focus on uninterrupted programming.  Those days are gone, but I still find myself enjoying the times when I can block out a few hours, close my door, and get into the code.

The rest of the world doesn't stop, however.  So I have a few things I wanted to briefly talk about today.
​

Amazing Tales of a Fringe Utopia in Northeast Illinois

A reader of this blog emailed me some links to material that, for all I know, is familiar ground to those who closely watch "fringe" theorists.  I had never heard of E. P. Grondine's manuscript He Walked Among Us, however, so I presume that others also have not.  The work (available in three parts here), details the history, philosophy, and inter-personal interactions of "fringe" figures (including David Hatcher Childress) in a small town in northeastern Illinois. Thus far I have only skimmed through parts of it (it will go on my summer reading list).  Check it out and see what you think. 

I don't know E. P. Grondine and am not yet very familiar with his work. He has commented on this blog at least once.
​

Plugging Memory Leaks

I've succeeded in getting one of my models (ForagerNet3_Demography_V3) operational in Repast Simphony. Repast Simphony operates a little bit differently than Repast J (the platform that I used to write the model), so I had to learn something about those differences, write some new code, and re-configure some other sections of the code.  Though I've still got some testing to do to make sure the model is behaving the same (i.e., it's doing the same thing it did before the conversion), everything seems to be working within the model itself.

The issue I'm trying to solve now is getting the model to run in batch mode.  Running a model in batch mode means that the computer performs a series of model runs (a "batch") automatically.  One of the great benefits of computational modeling is that you can do systematic experiments and determine cause and effect. You can hold everything about the model constant except for the value of single parameter, for example, and see how changes in that parameter affect the outcomes. You can run the model as many times as you want -- tens, hundreds, or thousands -- to flesh out those cause-effect relationships. Running the model in batch mode automates that process.  Ideally, I can start a batch running on Friday afternoon and return to my office on Monday morning with a large dataset ready to analyze.

There is apparently a built in batch configuration in Repast Simphony, but for some reason I haven't yet been able to get to it in the software. Maybe I need to reinstall.  For the time being, I've been using a simple little parameters file that just tells the software how many iterations of the model to perform and what random seed to use.  The model runs for the first 40-50 runs before throwing an "Out of Memory" error and locking everything up. It seems to run slower and slower with each iteration, which suggests to me there is a memory leak somewhere in my code. The model creates a bunch of objects (people, households, social links between people) during each run. Each of those uses memory. At the end of each run, all the object associated with that run should be tossed out to free up all the memory for the next run (when the model resets and starts fresh).  If some of the objects are "leaking through" and being retained in the memory, the progressive accumulation of those unused objects will eat more and more memory until there's none left.  I'm pretty sure I've got the model tossing all the people, households, and links (the three agent classes) out, so it may have something to do with copies of the spatial world and/or what's called the "context."  The structure of the "world" in Repast Simphony is different from that in Repast J, so I need to figure out how to make sure I'm getting rid of all the unused parts between each model run. That's my goal for today. Hopefully I can find and plug the memory leak and set my computer to work for me over the weekend.
​

The "Sword Report"

I still haven't been able to muster the combination of time and interest to read through J. Hutton Pulitzer's "sword report." What I know about the contents of the report I know from comments on Facebook pages (e.g., The Fraudulent Archaeology Wall of Shame and Fake Hercules Swords), this blog post from last week, and Jason Colavito's post about the report. My two main impressions are these (please correct me if I'm wrong):
​
The report contains no new information about the alleged "Roman sword from Nova Scotia." Pulitzer does not provide his XRF data that he claimed proved the sword was a "100% confirmed" Roman artifact. He spent months crowing about his XRF results, and, in fact, said he would release them "the next day" after Brosseau's results were aired on television. He has not done that. Why?  Remember when he said the exposed brass on the sword was actually gold? Whatever happened to that claim?

The report is an argument against Brosseau's interpretations, not her results.  Pulitzer seems to have abandoned the argument that his XRF results are correct and Brosseau's results are wrong, and is arguing that the metals identified by Brosseau are consistent with those produced by Romans.  He actually made this pivot some time ago (I wrote about it back in January). Sometime after Brosseau's results were aired, he decided that he had a better chance making a case for the antiquity of the sword based on Brosseau's results (which are well-explained and documented) than his own (let us never speak of those XRF results again?).  Apparently, the "case" for the Roman antiquity of brass with 35% zinc is based on the same sleight-of-hand he tried in January (see this post) with the added puffery of 70+ pages lifted more-or-less directly from this online study by David Dungworth.   Here is a direct quote from that study:

An Invitation for The Walking Dead to Enjoy Springtime in South Carolina

The Walking Dead is my favorite television show.  The program has had its ups and downs, but I think this season is pretty strong and I'm enjoying it. I'm a couple of episodes behind right now, so don't spoil it for me.

This week, Disney and Marvel warned Georgia Governor Nathan Deal that they will stop filming in the state if he signs a so-called "religious liberty" bill that many say will legalize anti-gay discrimination (here's the story in The Washington Post).  

I applaud Disney and Marvel for their stance, and hope that AMC follows suit and moves the filming of The Walking Dead out of northern Georgia. Given what's going on in North Carolina right now, the logical choice is to put the show in South Carolina.  There are signs this state is moving in the right direction (e.g., the removal of the Confederate flag last summer), and your business would be a nice encouragement. Having traveled up and down I-20 and 301 a few times now, I can tell you that you won't have any difficulty finding good locations for filming. You can enjoy the palmettos, azaleas, crepe myrtle, and Carolina wrens. We have some room at our house, so I can provide accommodations for at least two cast members (Glen and Maggie? Carol?).

Just think it over. You don't have to answer now. If you can't make it here this spring the flowers will still be blooming all year round. And there are butterflies.

After having set computer modeling aside (out of necessity) for the past year, I'm ready to start getting back to that component of my research.  I'm hoping to get things ramped back up this semester and be doing some fairly hardcore computational work by next semester.  I've got some technical and logistical issues to address, and I'll surely  have to spend some time getting back up to speed with Java and Repast.  But it shouldn't be too bad

One of the things I'd like to do when I get up and running again is take another look at how the children's labor might affect the dependency ratio and household-level calculations about family size, etc. (see this paper for my first attempt to address the issue). In the models I'm currently using (e.g., the ForagerNet series), I represent a child as either a "producer" or a "non-producer" by comparing the child's age to the value of a parameter that specifies the age at which children become producers. Clearly that's a significant simplification of reality: children do not magically become "producers" overnight on their 8th birthdays or their 12th birthdays, etc.

I was watching my two boys (ages 4.7 years and 2.2 years) shell beans at the kitchen counter over the weekend, and it got me wondering what kind (if any) quantitative data are available about age-based changes in the proficiency of children in doing various kinds of productive tasks. Watching my two kids, there was a huge difference in the proficiency (and interest level) of the older and younger boys.  The older one stuck with the task until we ran out of beans and, while not as fast as an adult, was really pretty good.  Despite an equal contribution of energy (at least in the beginning), the productivity of the younger one was much lower.  Part of that had to do with the desire to do a victory dance for each single bean that he managed to pull from a pod. 

What would the age-based "proficiency curve" of bean processing look like?  A linear progression? A rapid increase in proficiency between ages 2 and 6?  At what age do individuals reach "full proficiency" in processing beans?  

​Are there ethnographic data available that would allow me to understand how proficiency at processing various kinds of plant foods changes with age? How about tasks related to gathering?  Or planting?

There has been an increase in interest in the lives of children in hunting-gathering societies (see this paper by Nurit Bird-David), and I'm hoping that some quantitative data are available from recent ethnographic studies (e.g., see this 1994 paper by Blurton Jones et al., this 2004 paper by Raymond Hames and Patricia Draper, this 2009 volume edited by Barry Hewlett and Michael Lamb).  I'm less familiar with possible sources for quantitative data relevant to modeling age-based changes in children's work proficiency in agricultural societies.  Maybe there are government agencies or NGOs that monitor that sort of thing (the site of the International Labour Organization discusses children's domestic work, for example).

Anyway, this blog post is just a placeholder. I'm interested in tracking down some data if they're available (if you know of any, please let me know!).  

If there are not suitable data already out there, I might have to generate some of my own through some controlled experimentation.  Let me know if you like beans.  We may soon be producing a surplus at my house. And I may soon ask to borrow the participation of children of various ages.
​

As far as I know, humans are unique among animals in having an extended period between weaning and being able to subsist on their own.  We call this “childhood.”  The long period of post-weaning dependence provides our large brains with a lot of time to mature.  It also requires a lot of parental investment (in terms of time, energy, calories, etc.) and means that we would have to wait a long time between offspring if each one had to independent before the mother could have another.   We don’t do that, tending to have a shorter interval between subsequent births (the inter-birth interval, or IBI) than other great apes.  The long period of childhood dependence and the short IBI mean that, as a species, humans tend to have multiple, dependent offspring of different ages at the same time.  Speaking as a parent of multiple, dependent offspring of different ages, I can tell you that this is often no walk in the park.  This peculiar human strategy has a lot of costs.

Understanding when, how, and why this distinctly human reproductive strategy developed is a great evolutionary question.  Reducing the IBI increases the potential fertility of human populations, but also creates new demands on the energies of parents and families.  Human families today often offset those extra energy demands by getting help (evolutionary anthropologists call it “cooperative breeding”).  A new paper in the Journal of Human Evolution titled “When Mothers Need Others” by Karen Kramer and Erik Otárola-Castillo tries to further our understanding of where cooperative breeding comes from, using an “exploratory model” to try to understand the selective pressures associated with the evolution of human-like patterns of reproduction and child-rearing.  The goal of the paper “is to develop a model to predict those life history transitions where selective pressure would have been strongest for cooperative childrearing” (pg. 5). 

Kramer and Otárola-Castillo call their model the “Force of Dependence Model.”  Their model is a simple one, calculating “the net cost of offspring as a function of dispersal age, birth intervals and juvenile dependence” as a 3-dimensional surface (Supplementary Online Material  from Kramer and Otárola-Castillo 2015).  The authors use several different combinations of settings to represent a range of conditions from “ancestral” (juvenile independence at age 10, IBI of 6 years, and a dispersal age of 14) to “most derived” (juvenile independence at age 20, IBI of 3 years, and a dispersal age of 20).  Their graphs show that “net costs” within a domestic group (a mother and her offspring) are lower when offspring are spaced further apart and become independent at a younger age.  When offspring hang around past the age of juvenile independence, there is a net benefit to the domestic group as their productive capacities can be used to offset the drain of their younger siblings. The authors find that the strain points – where selective pressures for assistance would be greatest – occur in domestic groups with the most derived set of characteristics: late juvenile independence and a low IBI (lots of children who remain dependent for a long time).   

As I understand it, the “net cost” in this model more-or-less mirrors the dependency ratio (the ratio of consumers to producers) of a domestic group or family, something anthropologists have been interested in understanding for a long time.  The higher the ratio of consumers to producers, the higher the dependency ratio, and the higher the “cost” to each producer supporting the family.  The dependency ratio changes through the lifespan of a family in a patterned way: every domestic group that has children goes through a “pinch” period when the dependency ratio is highest, and the pinch period logically corresponds to the time when there are a lot of dependent offspring.  As I wrote in my 2013 paper in the Journal of Anthropological Archaeology (“Subsistence Economics, Family Size, and the Emergence of Social Complexity in Hunter-Gatherer Systems in Eastern North America,” available here):

“the duration and amplitude of the ‘pinch’ is affected by the rapidity of the addition of offspring and how quickly those offspring turn from consumers into producers.  The rapidity of addition of offspring will depend on factors such as fertility, infant and childhood mortality, and the number of wives. The productive potential of children will be affected by the presence and distribution of resources that can be procured by children and the foraging strategies that are employed to exploit those resources” (White 2013:128).

The main part of that paper used an agent-based model (ABM) to try to understand how the distribution of family size changes when the age at which children become producers (the “age of juvenile independence” in Kramer and Otárola-Castillo’s model) decreases and there is an incentive for polygynous marriage.  In addition to the ABM, I used a simple spreadsheet model to show how the dependency ratio changed through the course of the developmental cycle of an individual family in cases where the age at production was low (8 years old) and where it was high (14 years old).  In this simple model, I used an IBI of 3, a dispersal age of 16 for females and 20 for males, and a female reproductive period spanning ages 20-35 years (giving a total fertility of 6 offspring).

The figure below compares Kramer and Otárola-Castillo’s graphs from their cases with early and late juvenile independence (holding IBI at 3 and dispersal age at 14) with my data on changes in dependency ratio through the developmental cycle in cases with a single reproducing female and an age at production of 8 (top) and 14 (bottom).   My model data are the same as in my 2013 paper (Figure 5), but I have re-graphed them to make comparison with Kramer and Otárola-Castillo’s figure easier.  I have redrawn the graphs from Kramer and Otárola’s Figure 1 (third graphs from the left, top and bottom rows).   The dotted lines on the graphs of my results indicate a dependency ratio of 1.75, which is what I have generally used in my modeling efforts as a “typical” dependency ratio among hunter-gatherers (following Binford 2001:230).

My results showed the same pattern as Kramer and Otárola-Castillo’s:  the peak of the “pinch” comes earlier and is less severe when children become producers at an earlier age.  Even though our models have some differences (and some of the values of the parameters were different), the correspondence in results is notable. Compare, for example, when the amplitude of the “pinch” (peak dependency ratio in my results, greatest net cost in Kramer and Otárola-Castillo’s results; marked by stars) is greatest and the differences in amplitude between the early and late ages of juvenile contributions to subsistence.

The correspondence between my results and Kramer and Otárola-Castillo’s is unsurprising.  The idea that the dependency ratio of a domestic group changes through the course of its developmental cycle in a somewhat predictable way is not new (and the idea that the “pinch” comes when you have lots of little kids running around at the same time won’t come as a revelation to anyone who has multiple children).   This is a phenomenon that has been studied for decades (e.g., Chayanov 1966; Donham 1999; Fortes 1958; Goody 1958) and recognized as a key aspect of how hunter-gatherers organize themselves (Binford 2001:229).

So where does this kind of work put us in terms of understanding the evolution of human reproduction, society, and family life?  I think it primarily puts us in a spot where we’re asking some good questions.  Going back to the issue of the origins of monogamous pair-bonding (which I touched on briefly in this post about birth assistance and this post about australopithecine sexual dimorphism), having a two person (male-female) unit forming the core of a domestic group would have a mitigating effect on the strain caused by a decrease in IBI (i.e., you’d be adding another producer into the equation).  If the appearance of male-female pair-bonding was associated with a sexual division of labor (which is I think what most of us would hypothesize), males and females would presumably be focused on procuring somewhat different sets of subsistence resources.  Offspring could be largely “independent” with regards to some of those resources but not to others – think about the difference between collecting berries and running down large game.  A sexual division of labor and an environment where relatively young children could make some contribution to their own subsistence (even if that contribution does not include the full range of resources that are exploited) would go a long way toward easing the “pinch” that comes from having more children spaced closer together.

When does this happen in human evolution?  Of course that’s a tough thing to get at directly.  I think if you took a poll, the winner would probably be “around the time our genus emerges” or “with Homo erectus.”  An increase in total fertility (coincident with a lowering of the IBI) would help explain the population growth that must have been part of the dispersal of our species out of Africa prior to 1.8 million years ago.  It would also fit nicely with the evidence for an increased exploitation of animal resources around that same time.  Maybe Glynn Isaac was right all along to propose the emergence of human-like central place foraging with home bases and a sexual division of labor at the beginning of the Lower Paleolithic?

But what if monogamous pair-bonding and a sexual division of labor appeared much earlier – with australopithecines or even some pre-australopithecine like Ardipithecus?  If those things came along with bipedal locomotion, would a decreased IBI and increased fertility have followed automatically?  Maybe not.  Perhaps those earlier hominids just didn’t have the wherewithal to exploit their environments like later hominids did – perhaps the diversity of the resource base they could exploit wasn’t great enough to really leverage a sexual division of labor until animal products became readily attainable.  That may have required a suite of anatomical adaptations for daytime exhaustion hunting (loss of body hair, skin pigmentation, greater body size, stiffer foot) and cognitive/behavioral adaptations for making and using stone tools to process carcasses.  The date of the “earliest” proposed use of stone tools continues to be pushed  back (now it’s at 3.3. million years ago), but as far as I know the density of stone tools and butchered animal bones that appears at about 1.8 million years ago is unlike anything that precedes it.

More modeling work will be required to really understand how changes in the dependency ratio might have articulated with changes in reproductive, social, and technological behaviors deep in human prehistory.  In order to understand what changes in reproduction might have meant in terms of social interactions, however, we’ll need a different grade of model than that used by Kramer and Otárola-Castillo.  Of course I’m going to say that complex systems modeling is the way to go on this:  it will let us get past the limitations of deterministic inputs and help us understand how constraints, costs, and interactions would have played out within a society.   In order for “others” to help with raising and provisioning multiple dependents, those others had to have existed within these small-scale hominid societies and (again, speaking as someone involved in raising multiple small kids) there wouldn't have been some inexhaustible Plio-Pleistocene babysitting pool of “others” out there just waiting to step in and provide extra calories for a few years.  A different kind of modeling effort with broader scope will let us get at the group- and society-level contexts in which family-level changes in child-bearing and child-rearing would have played out. Stay tuned.

References

Binford, Lewis R.  2001. Constructing Frames of Reference: An Analytical Method for Archaeological Theory Building Using Hunter-Gatherer and Environmental Data Sets.  University of California Press, Berkeley.

Chayanov, A. V.  1966.  A. V. Chayanov on the Theory of Peasant Economy.  University of Wisconsin Press, Madison.

Donham, Donald L. 1999.  History, power, ideology: Central issues in Marxism and anthropology.  University of California Press, Berkeley.

Fortes, Meyer.  1958.  Introduction.  In The Developmental Cycle in Domestic Groups, edited by Jack Goody, pp. 1-14.  Cambridge Papers in Social Anthropology.  Cambridge University Press, London.

Goody, Jack.  1958.  The Fission of Domestic Groups among the LoDagaba.  In The Developmental Cycle in Domestic Groups, edited by Jack Goody, pp. 53-91.  Cambridge Papers in Social Anthropology.  Cambridge University Press, London.