You are on page 1of 9

Journal of Archaeological Science 120 (2020) 105168

Contents lists available at ScienceDirect

Journal of Archaeological Science


journal homepage: http://www.elsevier.com/locate/jas

The global ecology of human population density and interpreting changes


in paleo-population density
Jacob Freeman a, b, *, Erick Robinson a, Noelle G. Beckman b, c, Darcy Bird d, Jacopo A. Baggio e, f,
John M. Anderies g
a
Anthropology Program, Utah State University, Logan, UT, 84322, USA
b
Ecology Center, Utah State University, Logan, UT, 84322, USA
c
Department of Biology, Utah State University, Logan, UT, 84322, USA
d
Department of Anthropology, Washington State University, Pullman, WA, 99164, USA
e
School of Politics, Security and International Affairs, University of Central Florida, Orlando, FL, USA
f
Sustainable Coastal Systems Cluster, National Center for Integrated Coastal Research, University of Central Florida, Orlando, FL, USA
g
School of Sustainability, Arizona State University, Tempe, AZ, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Explaining variation in human population density constitutes a basic research problem in human ecology and
Human ecology archaeological science. To contribute to this basic research problem, we build a graphic model and conduct a
Radiocarbon global analysis of the effects of ecological variables, controlling for technological differences, on human popu­
Climate change
lation density. Our results indicate that human population density displays a consistent relationship with
Human population ecology
Macroecology
ecological variables across productive technologies. Human population density peaks in high productivity en­
vironments with moderate levels of species richness and moderate to low levels of pathogens among hunter-
gatherer, subsistence agricultural, and industrial societies. Population density is lower in low productivity en­
vironments as well as environments with high productivity, high species richness and high pathogen loads. The
productive technologies associated with agriculture and industrial production impact human population density
more than ecological variables, and shifts toward agriculture or industrial production may weaken the effect of
net primary productivity on population density. These results illustrate the nature of the relationship between
population density and ecological variables, which are partly driven by climate, and the results provide a basis
for hypotheses that researchers can use to analyze the potential effects of climate change on material record
estimates of human paleo-population density.

1. Introduction us to make a very practical point about the analysis of paleodemo­


graphic proxies: we can use the results of our study (and similar studies)
Explaining variation in the global distribution of human population to create more informed hypotheses about when climate variables
density has been a research question in archaeology and human ecology should and should not strongly correlate with estimated changes in
for decades (e.g., Tallavaara et al., 2018; Freeman, 2016; Freeman and paleo-population density. To this end, we use our results to propose
Anderies, 2015; Kelly, 2013; Hamilton et al., 2007; Luck, 2007a, b; hypotheses that researchers can use to guide their investigation of po­
Binford, 2001, 1983; Hassan, 1981). In this paper, we make two con­ tential relationships between paleoclimate and paleodemographic
tributions to this research question. First, to increase our understanding proxies. In the remainder of this paper, we propose a graphic model of
of the global scale variation in human population density, we evaluate a ecological constraints on human population density, evaluate the model,
model that describes the effects of productive technology and three key and develop a set of hypotheses for analyzing
ecological variables on human population density: the productivity of an paleoclimate-paleo-population density proxies.
ecosystem, the species richness of an ecosystem, and the pathogen load
of an ecosystem (from a human’s perspective). Second, our work allows

* Corresponding author. Anthropology Program, Utah State University, Logan, UT, 84322, USA.
E-mail address: jacob.freeman@usu.edu (J. Freeman).

https://doi.org/10.1016/j.jas.2020.105168
Received 12 August 2019; Received in revised form 22 April 2020; Accepted 25 May 2020
Available online 6 June 2020
0305-4403/© 2020 Elsevier Ltd. All rights reserved.
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

1.1. An ecological model of human population density species rich environments, a one unit increase in NPP results in a larger
increase in population density than in species poor environments. This
The inspiration for our paper comes from the explosion of research might occur because increases in NPP result in more energy available for
that uses radiocarbon ages to develop time-series of relative human human consumers, and higher species richness decrease the variability
population density (population size in a constant area) over the last 15 of NPP and important nutrients. Thus, increases in species richness
years (e.g., Robinson et al., 2019; Freeman et al., 2018a; Bevan et al., might amplify the positive effect of NPP on population density because
2017; Gayo et al., 2015; Kelly et al., 2013; Shennan et al., 2013), foragers in more species rich environments experience shortfalls of re­
referred to as the ‘dates as data’ approach. While this approach has been sources and nutrients less often. Understanding such interactions ben­
critiqued, our purpose is not to defend against or add to such critiques. efits from developing a more formal model of the effects of ecosystem
Rather, we want to address a separate research problem that this corpus attributes on human population densities.
of research raises. Suppose that the ‘dates as data’ approach provides a Second, throughout the Holocene human societies experienced
reliable proxy for changes in human population density. Would our drastic changes in technology. One of the most important examples of
understanding of the factors that affect human population density be such technological change occurred with the adoption of agriculture.
sufficient to interpret associations (or lack thereof) between paleo­ Conventional wisdom states that the adoption of agriculture increased
climate proxies and such datasets? We propose that the answer to this the carrying capacity of human societies (Freeman, 2016; Glassow,
question is: No (e.g., Robinson et al., 2019; Tallavaara et al., 2018; 1978). If the effect of technological change is large relative to changes in
Freeman et al., 2018b; Cromb�e and Robinson, 2014). As Tallavaara et al. ecosystem attributes, then these technological changes might be driving
(2018) argue, we cannot understand the impacts of paleoclimate change observed shifts in population density despite changes in climate that
on paleo-human population density without understanding how affect ecosystem attributes. Interpreting changes in the density of
ecological variables affect human population density in the first place. human populations over time requires understanding the relative effects
Holding technology and social organization equal, climate does not of technological change vs. ecological attributes on population density.
directly impact human populations. Rather, climate affects variables Finally, it is unclear whether population density displays similar
such as the net primary productivity, species richness, and the pathogen relationships with NPP, species richness, and pathogen load among
load of ecosystems (Kreft and Jetz, 2007). It is these attributes of eco­ agricultural and industrial societies as it does among hunter-gatherers.
systems that impact the population density of humans and mammals in Do ecosystem attributes constrain the population density of agricul­
general (Santini et al., 2018; Luck, 2007a). tural and industrial societies in the same way as they do human for­
Therefore, a key research problem in the analysis of paleoclimate and agers? If ecosystem attributes constrain human population densities in
paleodemography is to understand the chain of relationships between the same way across different productive technologies, this would sug­
climate and ecosystem attributes on the one hand and ecosystem attri­ gest that a general underlying process affects human population den­
butes and human population densities on the other hand. The relation­ sities. If the relationships between ecological attributes and population
ship between climate and ecological variables has been the subject of density vary systematically with technology, then this would suggest
research in ecology for decades and is fairly well developed (Kreft and that shifts in climate that impact ecosystem attributes would have
Jetz, 2007; Odum, 1997; Rosenzweig, 1968). However, the relationships different consequences for human populations, depending on their
between ecosystem attributes and human population density needs productive technology. Investigating the above potentialities can help
more description and explanation (Tallavaara et al., 2018). To help clarify when we should and should not expect associations between
build a knowledge base of the relationship between ecosystem attributes material record estimates of population density, paleoclimate proxies,
and human population densities, we build on previous research to pro­ and material traces of productive technology.
pose and evaluate a graphic model that may explain global scale vari­
ation in human population density. 1.2. A graphic model
Studying hunter-gatherers, Tallavaara et al. (2018, p. 1232) propose
a qualitative model to explain global variation in hunter-gatherer pop­ Fig. 1 builds on Tallavaara and colleagues’ qualitative model and
ulation density. They propose that: (1) The main ecosystem attribute provides a set of hypotheses to guide an empirical analysis of global
that affects population density is potential net primary productivity scale variation in hunter-gatherer, agricultural and industrial society
(NPP). NPP refers to the amount of biomass produced per unit space per population density. Fig. 1A describes the marginal benefits and costs of
unit time (Odum, 1997). This variable, thus, estimates the amount of higher NPP and higher population density, respectively, and Fig. 1B
potential energy available for secondary consumers in a given area. As summarizes the expectations for how population density varies with
NPP increases, the availability of food (energy) increases and, thus, so NPP, holding the effects of species richness and pathogen load equal, for
should population density. (2) Studies of biodiversity indicate that as the two different productive technologies.
species richness of ecosystems increases, so does the stability and mean The red curves on Fig. 1A describe the marginal benefits of increases
of ecosystem productivity (Wang et al., 2019; Mazancourt et al., 2013; in NPP for human consumers. These curves capture the idea that as NPP
Cardinale et al., 2012). Thus, more species rich ecosystems, at large increases, the availability of energy in an ecosystem that humans can
scales, should be more productive and less risky for human foragers consume increases, and thus the ecosystem can support more people per
because foragers can take advantage of species that serve similar needs unit area. The red curves also capture the idea that the marginal benefits
but have asynchronous responses to climate drivers. (3) Pathogen load of NPP display diminishing returns. This means that as energy becomes
should affect population density by increasing morbidity and mortality, more available, humans experience diminishing benefits. Benefits
which creates incentives for foragers to spread out to avoid the disease diminish because, even though more energy is available as NPP in­
transmission that causes such morbidity and mortality. creases, not enough time exists in the sleep/wake cycle or efficient
Tallavaara and colleagues’ analysis of approximately 300 hunter- enough infrastructure does not exist to exploit all of the extra available
gatherer societies broadly supports the above model. They conclude energy. As a starting hypothesis, we propose that the relationships dis­
that, despite much social and technological variation, hunter-gatherer played by the red curves holds across types of productive technology.
populations do not transcend the ecological constraints that affect the This may be the case as, ultimately, agricultural societies (subsistence
population densities of other mammals. While important work, the and industrial) depend on energy from the sun and water to grow crops,
above model needs more formalization and testing for three reasons. just as hunter-gatherers depend on these climate inputs for the growth of
First, Tallavaara and colleagues’ analysis indicates that ecosystem wild resources.
attributes interact in their effect on human population density (see, for Second, we propose that the marginal benefits of increases in NPP
example, Figure S1 Tallavaara et al., 2018). For example, in more vary with species richness. In low NPP environments, high species

2
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

hypothesis we expect this dynamic to occur across types of productive


technology. Indeed, species richness in agricultural societies mimics the
underlying species richness of ecosystems (Freeman, 2012) and could
lead to an exploration–exploitation tradeoff that generates steeper
diminishing returns.
Finally, the blue curves in Fig. 1A describe the marginal costs of
higher population density. The costs of higher population density
include increased disease transmission and potentially more conflict as
individuals encounter each other more often. These costs increase at an
accelerating rate as human population density increases. Note the two
blue curves. These represent different levels of pathogen load: A mod­
erate pathogen load curve and a high pathogen load curve. These two
curves capture the idea that the marginal costs of living at higher pop­
ulation density will increase faster when pathogens are more abundant
than when less abundant in ecosystems.
The key insight of the above model is that global scale variation in
human population density, holding productive technology equal, should
result from an interaction between NPP, species richness, and pathogen
load. As long as the red curves in Fig. 1A are higher than the blue curves,
population density increases as NPP increases, though at a diminishing
rate. Once a given red curve crosses a blue curve, population density
should decline as the costs of the next unit of increase in population
density outweigh the benefits of the next unit increase in NPP.
Fig. 1B summarizes a set of guiding empirical expectations that
follow from the model. The black dashed curves in Fig. 1B illustrate a
less productive technological base and the solid curves a more produc­
tive technological base. The more humped curves illustrates how pop­
ulation density should increase with NPP, holding species richness and
pathogen abundance equal at high levels. The flatter curves illustrates
how population density should increase with NPP holding species
richness and pathogen abundance equal at more moderate levels. The
key here is that as NPP increases from very low levels, population
density should, initially, be higher in species rich and pathogen abun­
dant environments than in moderate species richness and pathogen
abundance environments. This is because the distance between the high
diversity red curve and high pathogen blue curve (the net benefits of
higher NPP) is greater than the distance between the moderate diversity
red curve and moderate pathogen load blue curve in Fig. 1A.
However, as NPP continues to increase, the distance between the
Fig. 1. A-Graphic model of the marginal benefits of higher NPP and the mar­ moderate diversity red curve and moderate pathogen blue curve actu­
ginal costs of higher population density. The dashed red curve illustrates the
ally becomes greater than the distance between the high diversity red
marginal benefits of NPP holding species richness equal at a moderate level and
curve and high pathogen blue curve in Fig. 1A. This means that at a high
the solid red curve holding species richness equal at a high level. The dashed
blue curve illustrates the marginal costs of higher population density holding
enough NPP, population density in moderate richness and moderate
pathogen load equal at a moderate level and the solid blue curve at a high level pathogen abundance environments exceeds the population density of
of pathogens. For all values of NPP that a red curve is above a blue curve, an high species richness, high pathogen abundance environments. Indeed,
increase in NPP will result in an increase in population density. For all values of the high diversity red curve and high pathogen blue curve actually cross
NPP for which a blue curve is above a given red curve, increases in NPP will each other at high values of NPP, leading to a decline in population
result in decreases in population density. B-Expected relationships between density in such high NPP environments.
population density and NPP for two different combinations of species richness Note, the marginal benefits of NPP, alone, are always positive or
and pathogen load and for two different productive technologies. Dashed neutral in the model. The negative relationship between NPP and pop­
curves ¼ hunter-gatherers; Solid curves ¼ subsistence agriculturalists. (For ulation density displayed by the humped curves in Fig. 1B results from
interpretation of the references to color in this figure legend, the reader is
the interaction of NPP with species richness and pathogen load. The
referred to the Web version of this article.)
moderate species richness and pathogen load red and blue curves in
Fig. 1A never cross; thus, the moderate diversity and pathogen load
richness leads to more stable productivity and amplifies the benefits of curves in Fig. 1B never decline. In such environments, the benefits of
increases in NPP. However, high species richness may also result in more higher NPP always outweigh the costs of higher population density,
steep diminishing returns. The basic idea behind this postulate is that all though the rate of population density increase does decline.
individuals face an exploration-exploitation tradeoff. The more re­ In sum, the effect of NPP on human population density depends on
sources used, the more that time is spread among resources leading to context. At low levels of NPP, higher species richness decreases variance
less efficiency in the production of food. This amplifies the saturation in the supply of food, and this has benefits that far outweigh the costs of
effect of more energy (NPP) caused by not enough time or efficient aggregating into more dense populations. However, at high levels of
enough infrastructure to exploit all of the energy available in high NPP species richness, the benefits of more stable NPP do not outweigh the
settings. This is captured by the difference between the moderate di­ costs of aggregating into higher population densities in terms of disease
versity and high diversity red curves in Fig. 1A. Steeper diminishing transmission and fights. Thus, population density may decline due to the
returns occur in the high species richness (diversity) environment due to interaction between species richness, pathogen load and NPP. As a
the impact of allocating time across many resources. Again, as a starting starting point, we propose that this process holds across multiple types

3
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

of productive technology systems. The solid curves in Fig. 1B describe ethnographically recorded societies’ territory as the point to extract NPP
the proposed effect of productive technology in addition to ecological and species richness. We used a mean value for the modern nation states.
attributes. Here, the intercepts of the solid curves are higher than the Finally, we estimated pathogen load at a global scale by using esti­
dashed black curves, but the forms of the curves remain exactly the mates of pathogen diversity and the incidence of pathogens in modern
same. This quantifies the proposition that the dynamics proposed in populations. Pathogen load is a composite index of both pathogen
Fig. 1A operate across types of productive technologies. Shifts to agri­ abundance and ease of transmission. The pathogen load index is calcu­
cultural or industrial economies simply augment the energy available lated using the average z scores from the prevalence of ten different
along NPP gradients, leading to a higher population density at a given pathogens (spirochetes, leprosy, plague, schistosomes, leishmaniasis,
level of NPP, on average, for an industrial society vs. an agricultural trypanosomes, typhus, filariae, dengue, and malaria) (Tallavaara et al.,
society vs. a hunter-gatherer society. 2018; Low, 1990). The pathogen stress index was interpolated to
generate a global raster, again scaled to the resolution of the climate
2. Data and methods rasters used to project NPP and species richness (Tallavaara et al., 2017).
Fig. 2 compares the estimated NPP and species richness of each of the
To analyze the global distribution of human population density, we three productive technology groups in our data set. We constructed
use three global data sets on human population density. We selected Fig. 2 to inspect potential biases between productive technology groups
these data sets because (1) all three data sets attempt to provide a global in terms of the terrestrial ecosystems that they inhabit. The curves are
sample of hunter-gatherer, subsistence agriculturalist/pastoralist or in­ visual aids that help divide the graph into spaces where combinations of
dustrial societies, and (2) the data sets were collected independently and NPP and species richness are common (between the curves) and
without knowledge of our specific research question when assembled. extremely rare (outside the curves) on earth. The key point is that the
The first data set was published by Binford (2001) and contains infor­ productive technology types have distributions that overlap in the main
mation on the population density of 339 ethnographically recorded linear scatter, suggesting pretty equal representation of NPP-species
hunter-gatherer societies from around the world. The second data set richness environments on earth across the productive technology
includes 86 societies primarily dependent upon agriculture and pasto­ groups. The only minor exceptions occur in the upper left quadrant and
ralism for their subsistence needs, previously published to study the lower right quadrant of the graph. In the upper left quadrant, agricul­
latitude-domesticated plant species richness gradient (Freeman, 2012) tural and contemporary countries dominate these rare ecosystem types.
and population-territory scaling (Freeman, 2016). The third data set In the lower right quadrant, hunter-gatherers are more frequent in the
consists of population, land area, and energy consumption data more rare ecosystem combination of high productivity but low species
compiled from the World Bank (TWB, 2016) and International Energy richness.
Agency (IEA, 2016) since 1973 to study population size–energy scaling
(Freeman et al., 2018b). We simply calculated the mean population 2.1. Methods
density of each country since 1973. The data are available for 117
countries. Together, these three data sets provide information on the The expectations illustrated by Fig. 1B guide our analysis of global
global distribution of human population densities among human population densities. To evaluate these expectations, we conduct
hunter-gatherer, subsistence agricultural and industrial societies. a multiple regression analysis and specify interaction effects between
The resulting dataset consists of 529 societies (329 hunter-gatherer, the three ecological variables of NPP, species richness, and pathogen
86 subsistence agricultural and 114 modern countries). Ten hunter- load. A multiple regression recognizes that numerous explanatory var­
gatherer societies were removed due to our inability to accurately iables may impact a single outcome variable. In this case, the model
calculate ecological parameters as a consequence of the societies living sketched above indicates that NPP, species richness, pathogen stress,
on coasts or islands. We decided to remove rather than impute the
missing values as this does not change our results. The sample of hunter-
gatherer societies overlaps with Tallavaara et al. (2018). The
hunter-gatherer and subsistence agricultural societies territories are
defined by ethnolinguistic units and the territory of modern countries is
defined by modern boarders. This means that several countries have
very large territories. The consequences is that our calculation of
average NPP, species richness, and pathogen load (discussed below)
masks a high degree of environmental variation in several cases. This
fact actually decreases the power of the industrial data set to detect the
effects of ecological attributes on population density; nonetheless, we
still detect such effects as noted in the results below.
For comparability, to estimate NPP and species richness, we follow
the methods developed by Tallavaara and colleagues (Tallavaara et al.,
2018, 2017). We estimate NPP from the climatological parameters of
temperature and rainfall using the Miami model. Temperature and
rainfall data were downloaded from WorldClim.org (Hijmans et al.,
2005), and the underlying projection of the climate data, resolution and
extent of the climate rasters was used as a template to create species
richness rasters (Tallavaara et al., 2017, pp. 1–4). Species richness was
estimated by combining mammal, bird and vascular plant richness.
Specifically, bird and mammal richness data were obtained from Biod
Fig. 2. Scatter plot of NPP (x-axis) and species richness (y-axis); with all
iversityMapping.org website (Jenkins et al., 2013; Tallavaara et al.,
documented societies in this study coded as hunter-gatherer, subsistence agri­
2017). Native vascular plant richness was modeled from climate pa­ culturalists/pastoralist, and industrial societies. The figure creates a two
rameters following Kreft and Jetz (2007), and then interpolated using dimensional representation of the range of ecosystems that human societies
inverse weighted distance to the grid size of the data on bird and occupy on earth. The upper and lower curved boundary lines mark combina­
mammal richness (Tallavaara et al., 2017, 4–7). To link NPP and species tions of NPP and species richness that are rare among terrestrial ecosystems
richness data to each society, we used the center of each on earth.

4
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

and productive technology should all contribute to explaining variation density may vary by productive technology groups.
in the outcome variable of human population density. Interaction effects Our methodology compares the fits of the above regression models
refer to how two or more explanatory variables may work in concert, and selects the best fitting model as a representation of the data, in
rather than in an additive way, to explain variation in an outcome particular, whether population density varies significantly between
variable. For instance, if the relationships between NPP, species richness productive groups as we expect. Table 1 compares the fit of a general
and pathogen load were simply additive, then we would only postulate least squares regression model (Eq. (1) above, denoted glsm in the table)
one curve on Fig. 1B, and, by controlling for species richness and with a random intercept (Eq. (2), TimeRI) and random intercept and
pathogen abundance in a regression, the remaining variation in popu­ slope model (Eq. (3), TimeRS). The table provides fit statistics and uses
lation density would fit the postulated relationship between NPP and the log likelihood ratio (L.Ratio) to test the null hypotheses that TimeRI
population density well. However, the model described above proposes does not explain more variation in human population density than
that NPP, species richness, and pathogens interact. This means that model glsm, and that TimeRS does not explain more variation in human
changes in species richness and pathogens modify the relationship be­ population density than model TimeRI. In this case, we can reject the
tween NPP and population density (as one can see in the difference successive null hypotheses at p< 0:05, indicating that allowing the in­
between the moderate and high diversity and pathogen curves in tercepts to vary randomly as in Eq. (2) explains more variation in pop­
Fig. 1B). ulation density than Eq. (1). Further, Eq. (3), which allows both slopes
Here, we use general least squares and linear mixed effects multiple and intercepts to vary randomly explains more variation in human
regression models to assess the effects of ecological variables and pro­ population density than Eq. (2) (TimeRI). In short, human population
ductive technology on human population density at a global scale. We fit density varies between productive technology groups, and the slope of
linear mixed effects models because such models are appropriate in the main effect of NPP on population density varies between productive
cases where variation in an outcome variable may be caused by differ­ technology groups.
ences between groups. For example, how often parents read to their Hence, in our main results below, we report the results from Eq. (3).
children might affect a child’s reading performance in school, but per­ To evaluate the effect of the interaction of NPP, species richness, and
formance might also vary between schools due to different school en­ pathogens, we use effect plots and contour plots. These plots graph the
vironments. A mixed effects model allows one to model the variation relationship between NPP and population density holding species rich­
explained by parents reading to their children (the “within effect”) and ness and pathogens equal at particular values (as in Fig. 1B). This allows
variation explained by differences between school environments (the us to observe if species richness and pathogens modify the relationship
“between effect”). Here, we expect ecological variables to affect human between NPP and population density consistent with the expectations of
population density across all types of productive technology (the within the model. Note that in all regression models above, we used z-scores to
effect), but we also expect density to vary between groups of societies center all variables around their mean to avoid potential coefficient
with different productive technologies (the between effect of hunter- biases introduced by colinearity. We ran all regression models in R using
gatherer, agricultural, industrial technology groups). the nlme and effects packages. All data and scripts are available at: htt
We fit three main regression models. First, we fit a general least ps://github.com/people3k/JAS2020Ethnographic-Population-Density.
squares model fit by maximum likelihood that does not model differ­
ences in population density between productive technology groups. This 3. Results
is a “null” model that allows us to understand if allowing population
densities to vary between groups improves the performance (variation in To begin, we provide a summary of our results and then describe the
population density explained at a given level of model complexity) of a details of the outcomes of our analysis. Our results, in general, are
regression model. Specifically, consistent with the model presented in Fig. 1. (1) As NPP increases,
human population density increases; however, species richness and
lnD ¼ β1 þ α1 NPP þ α2 BIO þ α3 PAT þ α4 NPP : BIO : PAT þ ε (1)
pathogen load significantly modify the relationship between NPP and
where D is the population density of a society; NPP is net primary pro­ human population density. Specifically, there is an optimal population
ductivity; BIO is species richness; PAT is pathogen load; and NPP : BIO : density environment for humans, holding productive technology equal,
PAT denotes the interaction of these three variables. that is defined by high NPP, moderate species richness and moderate to
Second, we introduce a random intercept using a mixed effects low pathogen loads. (2) Differences in productive technology have a
regression model fit using maximum likelihood. much larger impact on potential population density than changes in
ecosystem attributes. (3) Productive technology significantly increases
lnDij ¼ β1 þ α1 NPPij þ α2 BIOij þ α3 PATij þ α4 NPPij : BIOij : PATij þ vi0 þ εij the population density of human societies and weakens the underlying
(2) positive, fixed effect of NPP on human population density.

where D is the population density of the ith society in productive tech­ 3.1. The effects of ecological attributes and productive technology on
nology group j; NPP is net primary productivity; BIO is species richness; population density
PAT is pathogen load and NPP : BIO : PAT denotes the interaction of
these three variables. The term vi0 is random variation of productive Table 2 illustrates the fixed coefficients (within effects) from the best
technology groups around the fixed intercept of β1 . This random vari­ fitting regression model in Table 1 (TimeRS, Eq. (3)) associated with
able captures the possibility that productive technologies may cause individual ecological attributes, the interaction of NPP, species richness
variation in human population density. The ε term is the residual error and pathogen load and the random effects associated with productive
not explained by the other variables in the regression model. technology groups. As predicted, NPP has a positive and significant fixed
Third, we introduce a random slope into Eq. (2) and fit a regression effect on human population density. This means, on average, across all
using maximum likelihood. productive technology groups a one unit increase in NPP results in a
0.809 unit increase in the log of human population density. Species
lnDij ¼ β1 þ α1 NPPij þ α2 BIOij þ α3 PATij þ α4 NPPij : BIOij PATij
(3) richness (BIO) and pathogen load (PAT) also have positive fixed effects,
þ vi0 þ vi1 NPPij þ εij
though these effects are not statistically significant. More importantly
given the hypothesis developed in conjunction with Fig. 1, NPP, species
where all variables are the same as in Eq. (2) above. However, in this
richness and pathogen load have a significant fixed interaction effect on
regression, we add the term vi1 NPPij . This term allows for the possibility
human population density. Species richness and pathogen load modify
that the slope of the relationship between NPP and the log of population
the relationship between NPP and population density as depicted in

5
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

Table 1
Table comparing the fits of alternative regression models. The mixed effects models that allow population density to vary randomly between productive technology
groups fit the data better than a general least squares model that does not allow population density to vary between technology groups.
Model df AIC BIC logLik Test L.Ratio p-value

glsm 6.00 2501.93 2527.56 1244.97


TimeRI 7.00 1973.19 2003.09 979.60 1 vs 2 530.74 <0.001
TimeRS 9.00 1964.47 2002.90 973.23 2 vs 3 12.73 <0.001

Fig. 3.
Table 2
Fig. 3 illustrates how species richness and pathogen load modify the
The fixed and random effects coefficients in the TimeRS regression model dis­
predicted effect of NPP on the log of population density. First, in very
played in Table 1 and Eq. (3). Standard errors reported in parenthesis and *** ¼
significant at the 99% level. low pathogen load environments (Panel A), groups living environments
with high species richness and low NPP have the lowest population
Fixed Coefficients
densities; however, as NPP increases there is a NPP threshold at 0.2
Predictors Coefficients (se) (about 650 g/m/yr) above which groups living in high species richness
NPP 0.809***
environments are predicted to have the highest population densities.
(0.18)
BIO 0.03 Above the 0.2 threshold, species richness amplifies the positive effect
(0.12) of NPP on population density (purple line has a steeper slope and is
PAT 0.164 above all other lines).
(0.103) Second, as pathogen load increases (panel A to B to C to D), the
NPP : BIO : PAT 0.224***
amplifying effect of species richness on the positive relationship be­
(0.072)
Constant 0.73 tween NPP and population density weakens. In fact, in high pathogen
Random Effects load environments (Panels C and D), low species richness and high NPP
θ2Intercept 2.52 environments have the highest predicted population densities. In both
θ2NPP 0.25 Panels C and D, below a threshold of 0.25 (1330 g/m/yr), high species
Model Fit richness increases population density at a given level of NPP, but above
N 529 this threshold, high species richness decreases population density at a
AIC 1964.47
given value of NPP. These results are consistent with our model and
Log Likelihood 973.23
hypothesis that high species richness diminishes the marginal benefits of
NPP and a greater pathogen load increases the marginal costs of high
population density in high NPP settings, actually resulting in a decline in
population density.
Fig. 4A provides an additional way to view the predicted effects of
NPP and species richness on population density by holding productive
technology equal. Two observations: First, the slope of the relationship
between NPP and population density decreases from hunter-gather to
agricultural to industrial societies. This means that a one unit increase in
NPP results in a less steep increase in population density among indus­
trial and agricultural societies than among hunter-gatherer societies.
The implication is that NPP has a positive fixed effect across all pro­
ductive technology groups, but industrial societies and agricultural so­
cieties, in a sense, are less constrained by NPP than hunter-gatherer
societies. Second, note that below a NPP threshold of 0.25 (1330 g/m/
yr), groups who live in high species richness environments fall above
each respective regression line. Conversely, above this threshold, groups
who live in high species richness environments fall below each respec­
tive regression line. This is consistent with our model and the results
presented in Fig. 3. The effect of NPP is modified by species richness and
pathogen stress, with high species richness contributing to declines in
Fig. 3. Three way effect plot of the interaction effects of NPP, species richness population density at high levels of NPP.
and pathogens on population density. The plot holds pathogen load equal in a Fig. 4B illustrates NPP on the x-axis, predicted population density on
series of panels (A,B,C,D) and allows species richness to vary within each panel the y-axis and contours that represent levels of pathogen load. Observe
while graphing the relationship between NPP and population density. Hori­ an interesting and complex pattern. Between standardized NPP values of
zontal lines indicate inflection points of NPP where each linear curve meets in a 1.2 and 0.9 (300 and 1650 g=m2 r), holding productive technology
given plot. All ecological variables are standardized. A-Low pathogen load, one
equal, groups who live in high pathogen load environments actually
standard deviation below the mean pathogen value; B-Mean pathogen load; C-
have higher population densities than groups who live in lower path­
one standard deviation above the mean pathogen load; D 1.5 standard de­
viations above the mean pathogen load. In each panel, purple curve ¼ 1.5 ogen load environments. One can see this pattern by noting that the red
standard deviations above mean species richness; blue curves ¼ 1 standard and dark orange data points fall above each respective regression line
deviation above mean species richness; green curves-mean species richness; and within the NPP window of 1.2 to 0.9. Consistent with Fig. 3, at mod­
red curves ¼ 1 standard deviation below mean species richness. (For inter­ erate levels of NPP, the benefits of higher species richness seem to
pretation of the references to color in this figure legend, the reader is referred to outweigh the negative effects of higher pathogen load. Conversely,
the Web version of this article.) above the NPP threshold of 0.9, higher pathogen load societies typically
fall below each respective regression line. Interestingly, below the NPP
threshold of 1.2 (very low productivity environments), societies
among agricultural and industrial societies with high pathogen loads

6
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

human population densities occur in setting with high NPP, moderate


levels of species richness and moderate to low pathogen loads. At lower
levels of NPP, higher species richness increases population density, and
at high levels of NPP, higher levels of species richness lead to lower
population densities. This may occur because of the combined effects of
steeper diminishing returns from more NPP in high species richness
environments and the increasing costs of higher population density due
to the abundance of pathogens in such environments.
However, not all of the patterns that we observed were consistent
with our model. Most importantly, the slope of the fixed effect of NPP on
population density decreases from hunter-gatherers to agriculturalists to
industrial societies. This suggests that while NPP does lead to higher
population density, regardless of productive technology, NPP con­
straints population density less among agricultural and industrial soci­
eties than among hunter-gatherers. This could make sense. Agriculture
is, after all, an attempt to augment ‘wild’ NPP by controlling the
dispersal of plants and animals and the flow of water and nutrients to
increase productivity. Industrial economies depend upon fossil plant
and animal remains that further augment the energy available for pro­
duction and transportation purposes. In short, technological evolution
may not fully emancipate humans from the constraints of current levels
of NPP, but technology and infrastructure do weaken the constraints of
NPP on population density.

4.1. Insights for paleoclimate–paleo-population analysis

The motivation for our work comes from recent research that com­
pares radiocarbon time-series across different climate, ecosystem, and
technological contexts. Much of this recent work has focused on how
prehistoric human populations were impacted by climate changes.
Climate changes do not directly impact human populations, but rather,
indirectly impact human populations through their effects on different
Fig. 4. Depiction of NPP on the x-axis, predicted population density from the ecosystem attributes. For research on human paleodemography across
TimeRS regression model (Eq. (3)) on the y-axis, and contours that represent different climatic contexts to advance, it is important to build middle
level of species richness in graph A and pathogen load in graph B. The color of range theory for how different ecosystem attributes affect human pop­
the points also indicates the level of species richness (A) or pathogen load (B)
ulations. We must be able to predict how paleoclimate changes might
for a particular society, and the regression lines are the best fit for each
have differently impacted the ecosystem attributes that then affected
respective productive technology group (hunter-gatherer, agricultural, indus­
trial). (For interpretation of the references to color in this figure legend, the
human populations throughout the Holocene. We can use the results of
reader is referred to the Web version of this article.) our study to create more informed hypotheses about when climate
variables should and should not associate with estimated changes in
paleo-population density. To illustrate, we provide two brief examples.
clearly fall below the regression line, while high pathogen load cases are
First, many archaeologists use large data sets of radiocarbon to
not seen among hunter-gatherers. The higher population densities
develop summed probability distributions or kernel density estimates of
generated by more productive technologies in very low productivity
paleo-population density over time. Once these proxies are constructed,
environments may also carry a pathogen penalty in such environments.
researchers often attempt to identify associations between paleoeco­
logical or paleoclimatological records and such paleo-population esti­
4. Discussion
mates. This type of research takes place without the context provided by
the finding of our paper: changes in productive technology have a larger
The goal of this paper has been to make two interrelated contribu­
effect than ecosystem attributes on population density and ecological
tions to the science of archaeology and human ecology. First, we have
factors that affect population density interact with each other. These
built and evaluated a graphic model that may explain global scale
results allow us to develop a set of hypotheses about the strength of
variation in human population densities. Following previous research,
association between paleoclimate/ecological and paleo-population
the model allows us to hypothesize that human population density
density fluctuation in any given region.
varies between three well defined productive technology groups
(hunter-gatherer, agricultural, and industrial) and, within each respec­
(1) If productive technology changes over the period of study in
tive group, net primary productivity (NPP), species richness and path­
question (e.g., the sequence displays the process of increasing use
ogen load interact to affect global scale variation in human population
of domesticated plants), then we should expect population den­
density.
sity to increase for most time-steps in a series from t to tþ1 to t þ
Our results partly support the model presented in Fig. 1. In partic­
i, just as a function of increasing productive capacity. This should
ular, hunter-gatherers have lower population densities than agricultural
weaken the cross association between fluctuation in paleo­
societies, and agricultural societies have lower population densities than
climatological (as well as paleoecological) variables and a given
industrial societies. Though there is some overlap in population den­
paleo-population estimate. In short, social-technological changes
sities, societies do form distinct productive technology groups. Within
that increase the carrying capacity of a system should result in
each productive technology group, higher NPP results in higher popu­
weak cross association between paleoclimate and paleo-
lation densities; however, species richness and pathogen load modify the
population time-series. Thus, weak associations that do not
relationship between NPP and human population density. The highest
meet an arbitrary level of significance should be expected–not

7
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

interpreted as a negative result–when a system is undergoing changes in human population densities requires understanding the in­
increased productive capacity. Such lack of association is a teractions between climate, productivity, technology and population
consequence of social-ecological system dynamics. density. As Tallavaara et al. (2018) argue, research into the relationships
(2) The association should be strongest between paleoclimate between ecological attributes, which are controlled to some degree by
records and paleo-population estimates when social- climate, and human population densities provides a critical frame of
technological systems are near equilibrium (change very lit­ reference for interpreting changes in human population densities over
tle over some interval of a time-series). In this case, one time. Our study illustrates a remarkably consistent relationship between
should choose carefully their paleoclimate proxies and take human population density and ecological attributes at a global scale.
into consideration that NPP is likely the main driver of pop­ Future work that uses radiocarbon time series to investigate the impacts
ulation density, but this relationship is modified in complex of paleoclimate change on Holocene human populations must consider
ways by species richness and pathogen stress. the relative impacts of ecosystem attributes and productive technology
in the specific populations under investigation. These considerations
In sum, to more productively investigate associations between will improve the ability of researchers to investigate how climate
paleoclimate and paleo-population time-series: first, split the paleo- changes impact human populations, and provide a foundation for
population time-series into social-technological phases of productivity comparative studies of human population ecology using
expansion, stasis and perhaps contraction. Next, expect the strongest paleo-population proxies.
associations between paleoclimate records and paleo-population records
during periods of relative social-technological stasis. Finally, consider Declaration of competing interest
using multiple paleoclimate/ecological proxies to attempt to capture
interactions between NPP, species richness, and pathogen load. The authors have no conflicts of interest to declare.
Obtaining these measures from paleoenvironmental records will require
statistical techniques and modeling, such as rarefaction analysis (Birks Acknowledgments
and Line, 1992; Odgaard, 1999) and biomization methods (Prentice
et al., 1996). This study was undertaken by PEOPLE 3000, a working group of the
Second, an emerging issue in the interpretation of radiocarbon time- Past Global Changes (PAGES) project, which in turn received support
series is the extent to which human population densities respond to from the Swiss Academy of Sciences and the Chinese Academy of Sci­
changes in climate over large geographic areas. For example, Robinson ences.The authors also wish to thank Miikka Tallavaara and two anon­
et al., 2019 have recently developed a method that reconstructs paleo­ ymous reviewers for helpful comments on an earlier draft of the paper.
demography from radiocarbon time series according to different pale­
oclimate zones developed from downscaled and transient paleoclimate Appendix A. Supplementary data
models over the whole western US. This method attempts to move
beyond approaches that reconstruct paleodemography according to Supplementary data to this article can be found online at https://doi.
arbitrary modern or cultural historical criteria, in favor of paleodemo­ org/10.1016/j.jas.2020.105168.
graphic reconstructions that consider the climatic contexts and changes
in which different prehistoric populations lived. This new method for References
reconstructing paleodemography according to transient paleoclimate
zones promises to build a more uniformly comparative approach for Barton, C.M., Tortosa, J.E.A., Garcia-Puchol, O., Riel-Salvatore, J.G., Gauthier, N.,
research on the impacts of climate change on human populations. Conesa, M.V., Bouchard, G.P., 2018. Risk and resilience in the late glacial: a case
study from the western Mediterranean. Quat. Sci. Rev. 184, 68–84.
However, this methodology still does not allow the ranking of climate
Bevan, A., Colledge, S., Fuller, D., Fyfe, R., Shennan, S., Stevens, C., 2017. Holocene
zones as more or less productive for human populations. We have a good fluctuations in human population demonstrate repeated links to food production and
sense of how climate impacts ecosystems, and our work builds the link climate. Proc. Natl. Acad. Sci. Unit. States Am. 201709190.
Binford, L.R., 1983. Pursuit of the Past: Decoding the Archaeological Record. Thames and
between ecosystems and human population densities. One can, thus, use
Hudson, New York, N.Y.
our results to rank climate zones in terms of their expected ecological Binford, L.R., 2001. Constructing Frames of Reference: an Analytical Method for
attributes and the distance of these attributes from the ecological space Archaeological Theory Building Using Hunter- Gatherer and Environmental Data
with the highest population densities observed in the ethnographic re­ Sets. University of California Press, Berkeley.
Birks, H.J.B., Line, J., 1992. The use of rarefaction analysis for estimating palynological
cord, regardless of social-technological organization. This would allow richness from quaternary pollen-analytical data. Holocene 2 (1), 1–10.
researchers to observe how climate-ecological space on a landscape Cardinale, B.J., Duffy, J.E., Gonzalez, A., Hooper, D.U., Perrings, C., Venail, P.,
fills-up with people. The zones modeled as most close to the highest Narwani, A., Mace, G.M., Tilman, D., Wardle, D.A., et al., 2012. Biodiversity loss and
its impact on humanity. Nature 486 (7401), 59.
population density ecosystems observed ethnographically should fill-up Cromb�e, P., Robinson, E., 2014. 14c dates as demographic proxies in Neolithisation
with populations first, leap frogging less suitable zones, and so on. models of northwestern Europe: a critical assessment using Belgium and northeast
France as a case-study. J. Archaeol. Sci. 52, 558–566.
Dalfes, H.N., Kukla, G., Weiss, H., 2013. Third Millennium BC Climate Change and Old
5. Conclusion World Collapse, vol. 49. Springer Science & Business Media.
Dean, J.S., 1994. The medieval warm period on the southern Colorado plateau. Climatic
Archaeologists and paleoecologists have long collaborated to inves­ Change 26 (2–3), 225–241.
Freeman, J., 2012b. Domesticated crop richness in human subsistence cultivation
tigate the effects of climate on human societies and populations (e.g., systems: a test of macroecological and economic determinants. Global Ecol.
Barton et al., 2018; Naudinot and Kelly, 2017; Kelly et al., 2013; Dalfes Biogeogr. 21 (4), 428–440.
et al., 2013; Van de Noort, 2011; Kerr et al., 2009; McIntosh et al., 2000; Freeman, J., 2016. The socioecology of territory size and a “work-around” hypothesis for
the adoption of farming. PloS One 11 (7), e0158743.
Dean, 1994). The ubiquity of radiocarbon data for Holocene archae­ Freeman, J., Anderies, J.M., 2015. The socioecology of hunter–gatherer territory size.
ology means that radiocarbon time series provide the most widespread J. Anthropol. Archaeol. 39, 110–123.
data for research on the impacts of paleoclimate changes on Holocene Freeman, J., Baggio, J.A., Robinson, E., Byers, D.A., Gayo, E., Finley, J.B., Meyer, J.A.,
Kelly, R.L., Anderies, J.M., 2018a. Synchronization of energy consumption by
human populations. However, one of the challenges associated with
human societies throughout the Holocene. Proc. Natl. Acad. Sci. Unit. States Am.
studying the effects of climate on human populations is that climate 115, 9962–9967.
indirectly affects human populations through the direct effects of tem­ Freeman, J., Byers, D.A., Robinson, E., Kelly, R.L., 2018b. Culture process and the
perature and water on the productivity of resources. Moreover, social interpretation of radiocarbon data. Radiocarbon 60 (2), 453–467.
Gayo, E.M., Latorre, C., Santoro, C.M., 2015. Timing of occupation and regional
institutions and technologies also change the potential density of re­ settlement patterns revealed by time-series analyses of an archaeological
sources in an environment. Hence, interpreting the effects of climate on radiocarbon database for the South-Central Andes (16–25 s). Quat. Int. 356, 4–14.

8
J. Freeman et al. Journal of Archaeological Science 120 (2020) 105168

Glassow, M.A., 1978. The concept of carrying capacity in the study of culture process. Archaeology. http://www.sciencedirect.com/science/article/pii/S1040618
Adv. Archaeol. Method Theor. 1, 31–48. 216001567.
Hamilton, M.J., Milne, B.T., Walker, R.S., Brown, J.H., 2007. Nonlinear scaling of space Odgaard, B.V., 1999. Fossil pollen as a record of past biodiversity. J. Biogeogr. 26 (1),
use in human hunter-gatherers. Proc. Natl. Acad. Sci. U.S.A. 104 (11), 4765–4769. 7–17.
Hassan, F.A., 1981. Demographic Archaeology. Academic Press, New York. Odum, E.P., 1997. Ecology: A Bridge between Science and Society. Sinauer Associates,
Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high Sunderland, MA.
resolution interpolated climate surfaces for global land areas. Int. J. Climatol.: J. Prentice, C., Guiot, J., Huntley, B., Jolly, D., Cheddadi, R., 1996. Reconstructing biomes
Royal Meteorol. Soc. 25 (15), 1965–1978. from palaeoecological data: a general method and its application to european pollen
IEA, 2016. World Energy Balance Data. International Energy Agency. https://www.iea. data at 0 and 6 ka. Clim. Dynam. 12 (3), 185–194.
org/Sankey/. Robinson, Erick, Nicholson, C, Kelly, R.L., 2019. The importance of spatial data to open-
Jenkins, C.N., Pimm, S.L., Joppa, L.N., 2013. Global patterns of terrestrial vertebrate access national archaeological databases and the development of paleodemography
diversity and conservation. Proc. Natl. Acad. Sci. Unit. States Am. 110 (28), research. Advances in Archaeological Practice 7, 395–408.
E2602–E2610. Robinson, E., Zahid, H.J., Codding, B.F., Haas, R., Kelly, R.L., 2019. Spatiotemporal
Kelly, R.L., 2013. The Lifeways of Hunter-gatherers: the Foraging Spectrum. Cambridge dynamics of prehistoric human population growth: radiocarbon ‘dates as data’ and
University Press, Cambridge. population ecology models. J. Archaeol. Sci. 101, 63–71.
Kelly, R.L., Surovell, T.A., Shuman, B.N., Smith, G.M., 2013. A continuous climatic Rosenzweig, M.L., 1968. Net primary productivity of terrestrial communities: prediction
impact on Holocene human population in the Rocky Mountains. Proc. Natl. Acad. from climatological data. Am. Nat. 102 (923), 67–74.
Sci. Unit. States Am. 110 (2), 443–447. Santini, L., Isaac, N.J., Maiorano, L., Ficetola, G.F., Huijbregts, M.A., Carbone, C.,
Kerr, T.R., Swindles, G.T., Plunkett, G., 2009. Making hay while the sun shines? Socio- Thuiller, W., 2018. Global drivers of population density in terrestrial vertebrates.
economic change, cereal production and climatic deterioration in Early Medieval Global Ecol. Biogeogr. 27 (8), 968–979.
Ireland. J. Archaeol. Sci. 36 (12), 2868–2874. Shennan, S., Downey, S.S., Timpson, A., Edinborough, K., Colledge, S., Kerig, T.,
Kreft, H., Jetz, W., 2007. Global patterns and determinants of vascular plant diversity. Manning, K., Thomas, M.G., 2013. Regional population collapse followed initial
Proc. Natl. Acad. Sci. Unit. States Am. 104 (14), 5925–5930. agriculture booms in mid-Holocene Europe. Nat. Commun. 4.
Low, B.S., 1990. Marriage systems and pathogen stress in human societies. Am. Zool. 30 Tallavaara, M., Eronen, J.T., Luoto, M., Dec. 2017. Supporting data and script for
(2), 325–340. ”Productivity, biodiversity, and pathogens influence the global hunter-gatherer
Luck, G.W., 2007a. The relationships between net primary productivity, human population density.” [Data set]. Zenodo. http://doi.org/10.5281/zenodo.1069786.
population density and species conservation. J. Biogeogr. 34 (2), 201–212. https://doi.org/10.5281/zenodo.1167852.
Luck, G.W., 2007b. A review of the relationships between human population density and Tallavaara, M., Eronen, J.T., Luoto, M., 2018. Productivity, biodiversity, and pathogens
biodiversity. Biol. Rev. 82 (4), 607–645. influence the global hunter-gatherer population density. Proc. Natl. Acad. Sci. Unit.
Mazancourt, C., Isbell, F., Larocque, A., Berendse, F., Luca, E., Grace, J.B., Haegeman, B., States Am. 115, 1232–1237.
Wayne Polley, H., Roscher, C., Schmid, B., Tilman, D., van Ruijven, J., Weigelt TWB, 2016. World Development Indicators. The World Bank. http://data.worldbank.org
Weigelt, A., Wilsey, B.J., Loreau, M., 2013. Predicting ecosystem stability from /data-catalog/world-development-indicators.
community composition and biodiversity. Ecol. Lett. 16 (5), 617–625. Van de Noort, R., 2011. Conceptualising climate change archaeology. Antiquity 85 (329),
McIntosh, R.J., Tainter, J.A., McIntosh, S.K., 2000. The Way the Wind Blows: Climate, 1039–1048.
History, and Human Action. Columbia University Press. Wang, Z., Chiarucci, A., Arratia, J.F., 2019. Integrative models explain the relationships
Naudinot, N., Kelly, R.L., 2017. Climate change and archaeology. Quat. Int. 428, 1–2 the between species richness and productivity in plant communities. Sci. Rep. 9 (1),
Frison Institute symposium: International perspectives on climate change and 1–17.

You might also like