NOTICE: this is the authors version of a work that was accepted for publication in

the Journal of Anthropological Archaeology. Changes resulting from the publishing

process, such as peer review, editing, corrections, structural formatting, and other
quality control mechanisms may not be reflected in this document. Changes may
have been made to this work since it was submitted for publication. A definitive
version was subsequently published in: THE JOURNAL OF ANTHRO-
POLOGICAL ARCHAEOLOGY [(2015)] DOI: 10.1016/j.jaa.2015.07.001

1
 Toward a Theory of Non-linear Transitions from
Foraging to Farming
Jacob Freeman, Matthew Peeples, John M. Anderies
Department of Sociology, Social Work and Anthropology, Utah State University
School of Human Evolution and Social Change, Arizona State University
School of Human Evolution and Social Change and School of Sustainability, Arizona
State University

Abstract
The evolution of agricultural economies requires two processes: 1) the do-
mestication of plants and 2) specialization in agricultural practices at the
expense of alternative subsistence pursuits. Yet, in the literature, domesti-
cation receives the lions share of attention while theories of specialization
lag behind. In this paper, we integrate ideas from human behavioral ecology
(HBE) with tools from dynamical systems theory to study the effects of eco-
logical inheritance on levels of investment in foraging and farming. Ecological
inheritance is an outcome of niche construction and our study provides a for-
mal link between foraging theory and niche construction. Our analysis of a
dynamic model of foraging and farming illustrates that the optimal alloca-
tion of effort to foraging and farming can lead to the emergence of multiple
stable states. The consequence of this is that low-level farming optimizes
subsistence (e.g., minimizing the effort required to meet a subsistence goal)
in a forager-resource system over a few years, but makes the whole system
vulnerable to punctuated change over decades due to rare events. We use the
insights of our model to propose a general ecological framework to explain
the evolution and diversity of transitions from foraging to farming.
Keywords: Agricultural change, Niche construction, Human behavioral
ecology, Foraging theory, Resilience

Preprint submitted to Elsevier July 19, 2015

 1 1. Introduction
2 The adoption of domesticated plants is one of the most studied prob-
3 lems in anthropology, particularly as an outcome of human-environment in-
4 teractions (e.g., Barker, 2006; Binford, 2001, 1968; Bowles and Choi, 2013;
5 Childe, 1928; Flannery, 1985, 1973; Hayden, 1990, 1998; Huckell, 1995; Ken-
6 nett and Winterhalder, 2006; MacNeish, 1992; Piperno and Pearsall, 1998;
7 Rindos, 1984, 1980; Rowley-Conwy and Layton, 2011; Smith, 1995, 2011a;
8 Wills, 1988; Winterhalder and Goland, 1997, 1993). The long standing jus-
9 tification for this intensity of research is that the adoption of domesticated
10 plants underlies major changes in the evolution of human societies, such
11 as the emergence of inequality and sociopolitical hierarchies (e.g., Harris,
12 1996; Hayden, 1990; MacNeish, 1992, p. 3; Smith, 1995, p. 3; Wills, 1988,
13 p. 1). However, the evolution of political hierarchy and inequality are not
14 correlated with the adoption of domesticated plants per se, but are often
15 coincident with specialization in the production of domesticated plants at
16 the expense of hunting and gathering (Bender, 1978; Flannery, 1968; Hay-
17 den, 1990; Hunter-Anderson, 1986; Nichols, 1987; Rindos, 1980). Thus, a
18 fundamental question is why, once domesticated plants became available,
19 did some people increase their time invested in farming at the expense of
20 hunting and gathering while others did not (Barlow, 2006; Freeman, 2012a;
21 Hunter-Anderson, 1986; Phillips Jr, 2009; Rindos, 1980; Smith, 2001; Vierra,
22 2005; Zvelebil, 1996)? In short, what dynamics explain the various levels
23 of investmentfrom zero to near the total input and output of a subsistence
24 systemin farming?
25 In this paper, we work within a human behavioral ecology (HBE) frame-
26 work to combine an optimal foraging model with dynamical systems theory
27 to study why individuals specialize (or not) in farming when wild resource
28 alternatives are available as well. Our work contributes to understanding the
29 coevolution of subsistence strategies and ecosystems. Specifically, we use the
30 insight gained from our dynamic model of optimal foraging and farming to
31 develop expectations for how the modification of ecosystems may result in
32 different trajectories of change in the evolution of farming strategies. Our
33 model, by its very structure, illustrates the importance of niche construction
34 and ecological inheritance (discussed below) for understanding non-linear
35 trajectories of change in human subsistence systems. We argue that HBE
36 provides an overarching framework to understand specialization in domesti-
37 cates at the expense of hunting and gathering. Within HBE, however, models

3
38 that incorporate the feedback dynamics caused by niche construction are im-
39 portant, though under used, tools to help understand non-linear trajectories
40 of evolution. Dynamical systems models are one tool for extending models
41 drawn from optimal foraging theory to understand such possibilities.

42 2. Ecological Inheritance, Foraging models, and Non-linear Trajec-

43 tories
44 A growing number of anthropologists are using niche construction theory
45 to help explain the domestication of plants (or lack thereof) from long-term
46 herbivore-plant interactions (Bleed and Matsui, 2010; Laland and OBrien,
47 2010; Rowley-Conwy and Layton, 2011; Smith, 2007, 2011b,a, 2015; Zeder,
48 2015). Here we take on a different, but equally important issue. Why, once
49 domesticated plants are available, do populations make different levels of
50 investment in these resources? For example, maize entered the southwest
51 US by 4200 BP (Merrill et al., 2009). Although no one denies that the
52 crop continued to evolve and become fine tuned to local environments, the
53 marginal change in the crops niche after 2000 BP was very small. Yet,
54 in west Texas, the level of investment in maize remained very low, while
55 in other parts of the southwest, such as present day Colorado, New Mexico,
56 and Arizona, populations became specialized in the production of maize (e.g.,
57 Hard et al., 1996). At a finer scale, the Fremont archaeological tradition of
58 modern day Utah and Colorado is well known for a high degree of variation in
59 levels of investment in maize, evidenced among sites and individuals (Simms,
60 2008), despite the fact that people had access to similar races and varieties of
61 maize. This pattern is known from other archaeological case studies around
62 the world as well, for example, central western Argentina (Gil et al., 2011).
63 Our supposition is that the evolution of disparate investments in farming is
64 a consequence of the different ways that foraging practices and cultivating
65 domesticated plants modify the composition and structure of ecosystems and,
66 thus, modify the net benefits of different behaviors for producing food over
67 time.
68 A key process in our supposition comes from niche construction theory:
69 ecological inheritance. Ecological inheritance is an outcome of niche con-
70 struction, which is the ability of organisms to shape environments to states
71 that are suited to their biology (and less frequently to states that are not)
72 (Odling-Smee and Laland, 2011, p. 221). Odling-Smee et al. (2013, p. 8)
73 define ecological inheritance as

4
 74 [t]he inheritance, via an external environment, of one or more
75 natural selection pressures previously modified by niche-constructing
76 organisms (Odling-Smee et al., 2003). Ecological inheritance typ-
77 ically depends on organisms bequeathing altered selective envi-
78 ronments to their descendants, but other organisms, including
79 unrelated conspecifics and members of other species that share
80 the same ecosystem may also be affected by this legacy. Where
81 an act of niche construction leads to a change in the species com-
82 position of the local ecological community, this too is regarded as
83 an aspect of the ecological inheritance.

84 Here we are concerned with the last aspect of ecological inheritance in

85 the above definition, changes in the composition of an ecological community,
86 particularly as a consequence of the optimal decisions of individuals. 1 We
87 assume, as is common in HBE, that selection favors the optimal solution
88 to a problem in any given context, whether the solution is arrived at via
89 some combination of genetic adaptation, physiological plasticity or cultur-
90 ally transmitted information (Brown and Richerson, 2014, p.108). To the
91 point, the ways that humans produce food, which is constructing their niche,
92 may affect the composition of an ecosystem; in turn, the changed composi-
93 tion of the ecosystem may change the behaviors that are optimaland, by
94 inference, fitness maximizingin the future. It is this feedback between the
95 optimal behavior now, an ecosystems structure, and optimal behaviors in
96 the future that needs investigation to better understand variation in peoples
97 investment in farming.

98 2.1. On the importance of modeling feedbacks

99 In HBE, many explanations for changes in human behavior are grounded
100 by optimal foraging models (OFM) (Bird and OConnell, 2006; Codding and
101 Bird, 2015; see also Winterhalder and Smith, 2000). Such models are a useful
102 starting point, but need extension to better explain levels of investment in
103 farming. To see why, please consider the following example.
104 Barlow (2006) proposes a OFM to describe investment in the production
105 of domesticates at the expense of hunting and gathering. With just a few
106 assumptions, this model provides a partial explanation for when foragers will

1
Other aspects of ecological inheritance are important and interesting; they are simply
beyond the scope of this paper.

5
107 choose to invest in farming at the expense of foraging. Most relevant here:
108 1) Individuals can invest their time in two mutually exclusive food producing
109 activities, cultivating maize (the dominant domesticated plant in prehistoric
110 North America) and foraging for wild foods. 2) individuals only have so
111 much time that they can use to produce food, thus ls + lc l. Where l is the
112 maximum amount of effort (in units of time) that an individual can devote
113 to producing food, and ls and lc are the effort spent foraging and cultivating,
114 respectively. (3) Individuals attempt to maximize their rate of energy gain
115 because this maximizes their fitness.
116 Given these assumptions, we can rewrite Barlows (2006) model in a com-
117 pact form. Individuals obtain an output of energy described by the following
118 equation:

A(ls , lc ) = As (ls ) + Ac (lc ) (1)

119 where A(ls , lc ) is the total energy gained from foraging and farming; As (ls )
120 is a function that describes the energy gained from foraging; and Ac (lc ) is
121 a function that describes the energy gained from cultivating. When indi-
122 viduals maximize output subject to the time allocation constraint above, it
123 is assumed that they fully utilize their labor. Thus, the in the labor
124 constraint becomes equality which leads to ls + lc = l or ls = l lc . Hence,
125 we can eliminate ls from equation (1) an rewrite equation (1) as

A(lc ) = As (l lc ) + Ac (lc ). (2)

126 Differentiating output with respect to lc yields the first order conditions
127 for a maximum, i.e.

A0 (lc ) = A0s + A0c = 0 (3)

128 which leads to a well-known condition in economics: at the optimal labor al-
129 location, the marginal productivity of labor in different occupations is equal,
130 (i.e. A0s = A0c ). From this basic result comes Barlows (2006) decision rule
131 for when to cultivate maize:

A0s < A0c . (4)

132 This rule stipulates that individuals allocate effort to the cultivation of maize
133 once the marginal gain from the next unit of effort so spent is greater than the
134 opportunity cost of that unit of effort spent foraging (Barlow, 2006, p. 96).

6
135 Thus, a decline in the marginal returns from foraging should spur investment
136 in cultivating maize and vice versa.
137 While a productive start, the model above needs further development.
138 The most relevant reason to our discussion is as follows. The model does not
139 explicitly describe why the marginal returns from either foraging or farm-
140 ing may increase or decrease. The causes of such changes could be either
141 endogenous or exogenous processes that change much more slowly than deci-
142 sions about how much time to invest in foraging and farming. For instance,
143 on the foraging side of the equation, we can imagine that climate change
144 may drive the productivity of important wild resources down, leading to the
145 adoption of more agriculture (if the climate change did not similarly affect
146 the marginal productivity of cultivating maize). Alternatively, population
147 growth could lead to the depletion of wild resources, which would decrease
148 the marginal rate of return from foraging relative to farming; this possibility
149 invokes ecological inheritance as a causal mechanism; the composition of the
150 ecosystem is changed by the depression of wild resources, which changes the
151 calculus about whether it is optimal to forage or farm (Laland and OBrien,
152 2010, p. 315).
153 A qualitative appeal to ecological inheritance as a causal mechanism
154 glosses over the possibility that ecological inheritance results in non-linear
155 trajectories of change in many systems (Odling-Smee et al., 2003, p. 20-
156 21). Equation (4) assumes that foraging and farming are mutually exclusive,
157 both as activities and in their effects on ecosystems. For an individual who
158 makes decisions over a time scale of months, this might be reasonable. How-
159 ever, over decades, the foraging and farming decisions of many individuals
160 may create interactions and interdependencies between foraging, farming and
161 ecosystems. For example, clearing land to plant crops over decades might
162 positively or negatively change the productivity of wild resources; in turn,
163 this interaction effect would determine how individuals should best allocate
164 their effort to foraging or farming. The interdependency of foraging and
165 farming decisions with the productivity of an ecosystem could result in strong
166 non-linearities and threshold effects in the dynamics of human-environment
167 interactions.
168 There are several potential reasons that the evolution of investments in
169 farming may exhibit such dynamics. However, our concern here is the poten-
170 tial effect of a specific type of non-linear feature in social-ecological systems
171 the emergence of multiple stable states and thresholdson the evolution of
172 subsistence strategies (e.g., Anderies, 2006, 1998; Clark, 1976; Flannery, 1985;

7
173 Freeman and Anderies, 2012; Lade et al., 2013). Multiple stable states refers
174 to the idea that there are discrete ways that an ecological or social-ecological
175 systems could be organized under the same environmental conditions (e.g.
176 hunting and gathering or agriculture) (Anderies et al., 2002; Janssen et al.,
177 2003; Lever et al., 2014; Scheffer, 2009; Staver et al., 2011). The emergence of
178 multiple stable states is important because their presence in a system could
179 lead to transitions from foraging to farming that are punctuated, difficult to
180 anticipate, and very costly to reverse. Thus, the emergence of multiple sta-
181 ble states under some regimes of ecological inheritance and not under others
182 is likely a factor that determines why systems of low-level food production
183 follow different trajectories of change (e.g., smooth and continuous versus
184 punctuated).

185 2.2. Multiple stable states

186 Figure 1 is a heuristic for understanding how systems with multiple stable
187 states differ from those with a single state or equilibrium. An equilibrium
188 for a dynamical system is the state to which the system tends over long
189 time periods (e.g., a pendulum, after given an initial push will eventually
190 end up still, and hanging down). The value of this state can change as the
191 parameters that define a system change (e.g., the length of the pendulum
192 or the gravitational constant). The curves in figures 1a-c show how the
193 equilibrium of a system may change as an ecological parameter changes,
194 and the blue dots represent the equilibrium solution for a dynamical system
195 for a given value of the ecological parameter (e.g. solar flux, soil fertility,
196 temperature, etc.).
197 In systems with a single possible equilibrium, there is a continuum of
198 states that the system can occupy (Figure 1a and 1b). The consequence is
199 that evolutionary change is continuous and reversible. Alternatively, in a
200 system with multiple stable states (Figure 1c), there are multiple equillibria
201 that a system may occupy (i.e., multiple potential solutions to a system of
202 equations for a given set of parameters; two blue dots for the same parameter
203 value). The consequence is that evolutionary change is sometimes continuous
204 and reversible, but may also be punctuated, difficult to reverse, and difficult
205 to anticipate. To illustrate these differences between systems, please consider
206 the following two examples.
207 First, assume that the choice an individual has to make in Barlows (2006)
208 model is whether to allocate time to cultivating maize or harvesting mesquite
209 beans (a seed that falls from a mesquite tree). Abstracting away from the

8
210 issue of processing time for the moment, the marginal rate of energy gained
211 from mesquite is determined by the density of mesquite trees. This is because
212 the density of trees determines how long people must search for beans. By
213 analogy, the density of mesquite trees is the ecological parameter in Figure 1a
214 or b (the x-axis) and time devoted to mesquite collection is the state variable
215 (y-axis). This implicitly assumes that the density of mesquite changes much
216 more slowly (e.g., years) than decisions about the effort to invest in collecting
217 beans (e.g., days-to-months). The black curves, again by analogy, represent
218 the equilibrium time invested in collecting mesquite at each given density
219 of mesquite trees. As the density of mesquite declines, selection pressures
220 should favor decreasing labor, at equilibrium, in the collection of mesquite
221 (the blue dot moves down an to the right in 1a or 1b). Given the assump-
222 tions of Barlows model model, a decline in effort devoted to mesquite at
223 equilibrium necessarily implies an increase in the production of maize. In
224 sum, there is a continuum of unique labor allocations to the collection of
225 mesquite determined by the density of mesquite; the change in investment is
226 continuous and reversible, and, thus, labor allocation to farming also changes
227 along a continuum.
228 Second, lets continue the mesquite and maize theme to provide a con-
229 crete illustration of the dynamics summarized in Figure 1c. Flannery (1985)
230 proposes that there are tradeoffs between clearing mesquite forests to plant
231 maize and collecting mesquite beans. Flannery (1985, p. 505) states,
232 With maize crossing the the 250 kg/ha threshold, a new loop
233 in the system was established; mesquite trees were cut down to
234 make way for maize, and although preceramic peoples continued
235 to collect mesquite pods, they were harvesting from ever dwin-
236 dling Prosopis groves. Instead of propagating mesquite seeds by
237 their collecting and threshing behavior, they were propagating
238 maize kernels by eliminating mature mesquite.
239 In this example, the ecological parameter is the productivity of maize con-
240 trolled by the selective breeding practices of early Mesoamerican forager-
241 farmers. The equilibrium state is the amount of area covered by mesquite,
242 which determines the abundance of mesquite beans.
243 Now we can analogize Flannerys proposition to Figure 1c. An increase in
244 the average productivity of maize (x-axis) causes a decrease in the level of ef-
245 fort devoted to the collection of mesquite at equilibrium (the blue dot moves
246 down and to the right). This occurs because individuals choose to plant more

9
 (a)

k
{
System state

Ecological parameter
(b)

k
{
System state

Ecological parameter
(c)

k
{
System state

k
{

250 kg/ha
Ecological parameter
10
Figure 1: Figures designed after Scheffer (2009). The black curves represent all of the
possible equilibrium states of a system for a given ecological parameter. The blue dots
represent a unique equilibrium solution. The arrows indicate the direction that the system
will move over time if it is not in equilibrium.
247 maize and cut down mesquite as the productivity of maize increases. In this
248 case, the 250 kg/ha threshold is marked by the thick red line. Once maize
249 productivity hits this threshold, farmers rapidly shift toward a low effort in-
250 vested in collecting mesquite at equilibrium (all of the points on the lower
251 right thick curve to the right of the shaded area). In Flannerys proposition,
252 the reason the shift between equillibria occurs is because, once maize pro-
253 ductiviy hits 250 kg/ha, deforestation and seed consumption decrease the
254 rate of mesquite reproduction past a critical threshold. The implication is
255 that farmers move from a low level of effort to a high level of effort devoted
256 to the production of maize and skip over intermediate levels of investment
257 (the dashed curve). In this case, the shift from low-level farming to intensive
258 farming is punctuated and costly to reverse. However there is more to the
259 story.
260 Figure 1c helps illustrate that Flannerys proposition is missing a crucial
261 element. When the ecological parameter (here the productivity of maize)
262 falls within the shaded zone, the system is vulnerable to a variance induced
263 transition. This is because deforestation and consumption decisions are not
264 solely controlled by the productivity of maize (the ecological parameter).
265 There are shocks that constantly perturb an ecological system generated
266 by dynamics such as El Ni no-La Ni na weather cycles and pest cycles. For
267 example, a disease outbreak could shock a mesquite forest and cause a decline
268 in mesquite forest cover. In the shaded area of Figure 1c, this shock would
269 stimulate individuals to compensate by growing more maize despite the fact
270 that productivity is, on average, below the 250 threshold. This behavior, in
271 turn, would drive the rate of mesquite reproduction too low for the forest
272 to recover. Forager-farmers would then end up in an environment with a
273 low density of mesquite trees, and their high investment in the production of
274 maize would keep the forest from recovering. In this situation, the transition
275 is not only punctuated and difficult to reverse, but also difficult to anticipate
276 based on forager-farmers prior knowledge of the system.
277 In sum, the coupled processes of niche construction and ecological in-
278 heritance can lead to multiple stable states and critical thresholds. The
279 consequence of multiple stable states is that two mechanisms may generate
280 a punctuated shift in the allocation of effort to farming: 1) crossing a pa-
281 rameter threshold (e.g., the 250 kg/ha line) through a slow process internal
282 to a subsistence system or 2) crossing a state variable threshold (the dashed
283 black curve) via a fast external shock like a La Ni na episode. This second
284 mechanism is captured in the example above when forager-farmers, combined

11
285 with a pest outbreak, deplete the density of mesquite just enough (past the
286 dashed curve in the shaded area) that they cause a regime shift. In short, the
287 emergence of multiple stable statesdue to niche construction and ecological
288 inheritancemay effect the trajectory of evolution that a system follows.

289 2.3. Questions

290 Two questions follow from our discussion: 1) Under what ecological and
291 social conditions might the optimal allocation of effort to foraging and food
292 production cause multiple stable states to emerge? And, if multiple sta-
293 ble states emerge, 2) what variables could create different trajectories that
294 the evolution of food production follows? To contribute to answering these
295 questions we develop and analyze a dynamic model in the spirit of Barlows
296 OFM and Flannerys insights regarding the interdependency of humans and
297 ecosystems. Our questions hone in on the consequences of niche construction
298 and ecological inheritance for the evolution of various levels of investment in
299 farming.

300 3. A Dynamic Model of Time Allocation to Foraging and Farming

301 Our goal is to construct a model with the simplest possible assumptions
302 to capture the biological realism of niche construction and ecological inheri-
303 tance. In particular, we want to capture, in a minimal way, the interdepen-
304 dencies of foraging, farming, and the productivity of an ecosystem. As we
305 note below, our stylized model has the analytical power to explore alternative
306 scenarios of niche construction and ecological inheritance. A more complex
307 model would make this more difficult because complexity necessitates a sac-
308 rifice of analytical power. For our argument here, analytical power is more
309 important than model complexity, and, thus, we have chosen to keep the
310 model as simple as possible.
311 Our stylized model has two components: forest reproduction and growth
312 and time allocation to foraging and/or farming. We couple these two compo-
313 nents to develop a dynamic model of time allocation to foraging and farming
314 in a forest ecosystem under three basic assumptions.

315 1. The individuals in our model are households. The household is a basal
316 cooperative economic unit in many forager-farmer and forager soci-
317 eties. In particular, a male and female pair or pairs form the core
318 of a group who performs complimentary subsistence activities. For

12
319 example, Holmberg (1950, p. 124) states the nuclear family...is the
320 fundamental economic unit among the Siriono. He goes on: Hunt-
321 ing is largely a family affair, as are fishing, collecting and agriculture
322 (Holmberg, 1950, p. 124). These types of observations are common
323 from diverse places such as the American southwest, as among the
324 Yavapai (Gifford, 1936) to the the islands of Indonesia, as among the
325 Nuaulu (Ellen, 1978).
326 2. We assume that households may choose to produce two staple resources:
327 seeds produced by trees (e.g., nuts like acorns or the mesquite beans
328 mentioned above) and/or a domesticated plant, which, for concreteness
329 we call maize. It is common in the ethnographic record for seeds or
330 hearts produced from trees to provide a staple of forager and forager-
331 farmer diets. Examples include pi nyon pine nuts in the Great Basin
332 of North American (Steward, 1938), palm hearts in South America
333 and the Pacific (e.g., Holmberg, 1950; Ellen, 1978; Roscoe, 2002), and
334 acorns in California (Kroeber, 1953). The keys here are 1) decisions
335 are interdependent because clearing trees for agriculture effects the pro-
336 duction of seeds, and 2) there is a decision to make: How much effort
337 is best allocated to the production of each to maximize fitness?
338 3. Territory is open access. Open access means that there are no rules
339 that limit who, when, and where trees may be cleared to plant and
340 who, when, and where seeds may be collected. Our implicit rules are
341 simply you harvest, you own and you clear, you own. This means
342 that once a resource is harvested, or a plot of land is cleared, the re-
343 source is used by the household that did the harvesting or clearing. For
344 example, among Shoshoni foragers in the Great Basin of North Amer-
345 ica, Steward (1938, p. 253) tells us that once work had been done
346 upon the products of natural resources they became the property of
347 the person or family doing the work. Again, among the Siriono Holm-
348 berg (1950, p. 44) states that property is limited to the immediate
349 possession, by a family, of a garden plot, by virtue of having cleared
350 and planted it.

351 3.1. Forest reproduction and growth

352 We define the density of seeds and trees at time t with a set of ordinary
353 differential equations:

13
 State Variables Interpretations
s(t) The density of seeds (biomass/area) at time t
p(t) The density of trees (biomass/area) at time t
Parameters Interpretations
rj The rate of resource regeneration where j {p, s}
Kp Carrying capacity of trees
ds natural seed depletion rate
N Density of households
dc Rate of tree clearance
c Land cleared of trees (under cultivation)
K Minimum daily energy return rate
hi Per capita harvest of resource i in s,c
Ai The productivity of labor in resource i in s, c
Decay in returns to labor due to search time
li Labor directed toward resource i in s, c
l Total available per capita labor

Table 1: Model state variables and parameters

s = rs p ds s (5)
p = rp s(1 p/Kp ). (6)

354 This set of equations captures a minimal set of processes to generate a bio-
355 logically defensible story about the reproduction and growth of trees. The
356 change in seed biomass (s) is the intrinsic production of seeds (rs ) multiplied
357 the density of trees in a forest (p) less the rate of seeds depleted by animals
358 and decay (ds s). In turn, the change in tree biomass per unit area (p) is the
359 logistic growth function. The intrinsic growth of trees increases in proportion
360 to the density of seeds in the forest (rp s), and the growth limiting effects of
361 intraspecific competition (Kp ) constrains the growth of trees. Here Kp sets
362 the area that households can use at time t. This means that when we speak
363 of the number of households at time t (N (t)), we are referring to a density.
364 N (t) is best thought of as the ecological footprint of households, given their
365 technology; when N (t) = 0, humans have no footprint in the forest.

14
366 3.2. Harvest and time allocation
367 Now that we have defined the ecosystem, we posit a set of rules that define
368 an evolutionarily sensible strategy for how households allocate their effort to
369 the production of seeds and/or maize. To determine the best allocation of
370 effort to foraging and farming, we must first define the harvest of seeds and
371 maize.
372 We make three additional assumptions to define the per capita harvest
373 of households. First, seed output displays diminishing returns. Diminishing
374 returns occur because the more seeds that one gathers, the farther one must
375 travel to gather more seeds (e.g., Thomas, 1971, 21-33; Zeenah, 2002). Thus,
376 as seeds are depleted, an ever higher proportion of time is devoted to search-
377 ing, which decreases the marginal productivity of effort. Second, we assume
378 that agricultural technology exhibits constant returns to scale and that large
379 scale cooperation is not necessary (like maintaining large irrigation ditches)
380 to produce maize. All else equal, this means that one extra day of labor
381 invested in the cultivation of maize results in a one unit increase in yield.
382 The consequence of this assumption is that decisions about how much time
383 to invest in cultivation at the level of each individual household adds up rea-
384 sonably well to the aggregate level of cultivation by all households. Finally,
385 we assume that households minimize their time devoted to the production
386 of sufficient food to meet their demands (which is equivalent to maximizing
387 the rate at which they gain energy).
388 Given these assumptions, we define hc (t) and hs (t) as the per capita
389 harvest from cultivation and seed gathering, where: hs (t) = As (s)ls and
390 hc (t) = Ac (p)lc . In these equations ls and lc are the time devoted to gathering
391 seeds and cultivating maize. As and Ac are functions that represent the
392 productivity of effort spent foraging and farming, and 0 < < 1 captures
393 the decreasing marginal productivity of seed gathering. We assume that
394 As (s) = s because, for a given level of effort, lower seed density implies
395 a lower number of seeds gathered, and Ac (p) = p, which implies that soil
396 fertility scales linearly with tree density. This soil fertility assumption follows
397 from the idea that lower tree density is associated with shorter fallow periods
398 and, thus, shorter times for trees to regenerate soil nutrients. The effect of
399 these last two assumptions is that we can replace As (s) and Ac (p) with s and
400 p, respectively. Thus, hs (t) = sls and hc (t) = plc . The per capita harvest
401 of seeds and maize depends on the abundance of seeds (s) and trees (p)
402 (which we defined in equations (5) & (6)) multiplied by the amount of time a
403 household invests in foraging (ls ) or farming (lc ), respectively. These simple

15
404 harvest functions allow us to capture, in a minimal way, the interdependency
405 of foraging and farmingor at least the harvest of seeds and cultivation of
406 maize.
407 We determine how much effort should be allocated to foraging and farm-
408 ing by solving a constrained optimization problem:

minimize lc + ls (7)
subject to: lc + ls l (8)
Ac lc + As ls K 0 (9)

409 where l is the total time available, and K is the minimum caloric goal per
410 person per day to meet biological, reproductive and social needs. The con-
411 strained optimization problem simply says: given that you have l total units
412 of labor available per time period, produce K for that time period with as
413 little of your time budget allocated to foraging and farming as possible.
414 Box 1 presents the solution to the constrained optimization problem
415 households face. The solution defines a set of decision rules for allocating
416 effort to foraging and farming. Our decision set is a heuristic that tells a
417 household, at each time step, to check the optimal allocation of effort and
418 then allocate effort optimally if doing so does not violate the effort or min-
419 imum calorie constraints noted above. The heuristic allows a household to
420 converge on the best mix of foraging (ls ) and farming (lc ) that meets K in
421 the shortest amount of time whenever possible.
422

423 [BOX 1 HERE]

424

425 For example, Figure 2 illustrates five potential cases or ways that a house-
426 hold might allocate labor at any given time t, given the productivity of the
427 ecosystem at that time. The value ls is the point at which the marginal
428 productively from cultivating is greater than the marginal productivity from
429 gathering seeds. In case 1 a household can meet their energy constraint,
430 K at a point that is less than ls ; thus, the household is a hunter-gatherer
431 household. In case 2, it is optimal for a household to gatherer seeds until ls
432 and then cultivate maize until their energy constraint is met, K. Thus, the
433 household is a low-level food producer. In the model, a household could move
434 continuously between cases until they converge to some kind of an attractor.

16
435 The main point is that the optimal allocation of labor changes over time as
436 the productivity of the ecosystem changes. We define the interdependency
437 of foraging, farming, and the ecosystem below.

K
1
4
Total Output

3 5

ls l l ls
Total Labor Total Labor
Figure 2: Graphic depction of the optimization problem faced by households. When
the marginal benefit of cultivation (slope of blue line) exceeds the marginal benefit of
gathering seeds (slope of tangent to red curve) swith to cultivtaion. This occurs at ls ,
equation (16), Box 1. 1: Highly productive wild resource, never cultivate, set ls = ls ,
subsistence constraint K easility met. 2: Set ls = ls and lc = lc . It pays to cultivate, and
the subsistence constraint K is met. 3: Set ls = ls and lc = l ls . It pays to cultivate,
but the subsistence constraint K is not met. Cases 4 and 5: It never pays to cultivate
because l < ls . The subsistence constraint K may (case 4) or may not be met (case 5)
depending on the productivity of the resource.

438 3.3. Coupling the ecosystem with harvest and time allocation
439 The production of food creates two niche constructing activities in our
440 model (i.e., two ways households modify the ecosystem to suit their biolog-
441 ical and social needs for energy): gathering seeds hs (t) and the harvest of
442 maize hc (t). The effect of gathering seeds on the dynamics of the forest is
443 obvious, the more seeds that individuals gather, the lower the reproductive
444 rate of trees in the forest. The total harvest of seeds in the model is N hs (t);
445 the number of households multiplied by each households harvest of seeds.
446 Similarly, given our production function of hc (t) = plc , to cultivate maize
447 a household must choose how many trees to clear, which partly determines
448 the reproductive rate of the forest, as well as the fertility of soil. Thus,

17
449 households make a clearing decision: If the amount of cleared land is insuf-
450 ficient for the desired production of maize, then land must be cleared. We
451 define c(t) as a households rate of land clearing at time t. Total tree clear-
452 ing, N c(t) is the number of households multiplied by each households rate
453 of clearing. Finally, we assume that land is cleared at a rate proportional
454 to the difference between the desired cleared area and the available cleared
455 area: c(t) = dc (Acd /Aca ). Where dc is a measure of how effectively a house-
456 hold clears trees per unit effort and how responsive they are to deviations
457 of Acd /Aca , the ratio of area that needs cleared to the area already cleared,
458 away from Kp .
459 We can now couple the behavior of households with the growth and re-
460 production of the forest:

s = rs p ds s N hs (10)
p = rp s(1 p/Kp ) N c. (11)

461 Given seed and tree densities (s(t) and p(t)) at time t, households make
462 a decision about how much effort to allocate to foraging (ls ) and farming
463 (lc ), which determines hs (t) and c(t). In turn, hs (t) and c(t) determine the
464 future values of s(t) and p(t) through equations (10) and (11) and, thus, the
465 best allocation of effort to foraging and farming in the future. In sum, the
466 fitness of a households decisions now depends, in part, on the forest that it
467 inherits.

468 4. The Evolution and Resilience of Low-level Food Production

469 We begin this section with a summary of the main results gleaned from
470 an analysis of our model. Next, we illustrate the dynamics of the model that
471 led us to each of the three results.
472 Our analysis indicates that 1) as our model households attempt to behave
473 optimally, their niche construction and its legacy, ecological inheritance, gen-
474 erate multiple stable states: a productive state and an impoverished state.
475 Once multiple stable states emerge, households face the risk that they may
476 get pushed into the impoverished state and experience negative impacts on
477 their fitness. This risk is a feature of the ecosystem inherited from the pre-
478 vious harvest activity of households. 2) As households cultivate more maize,
479 the resilience of the productive state decreases, and the risk of the system

18
480 flipping into the impoverished state increases. Thus, in the short-run (a few
481 years), it is optimal to allocate effort to the cultivation of maize as a strategy
482 to mitigate diminishing returns from foraging; however, in the longer-run
483 (decades), the system is more vulnerable to a punctuated change that is dif-
484 ficult to anticipate. 3) It is less costly for households to reverse between
485 foraging and farming when cultivation is 1-40% of the diet. However, once
486 the 40% threshold is approached and crossed, it is likely difficult to go back
487 to low-level food production.
488

489 [BOX 2 HERE]

490 4.1. The emergence of multiple stable states

491 Figure 3 is a set of phase plane plots that illustrates the inter-relationships
492 between seed and tree density over time for three different values of popu-
493 lation (N ): N = 0, N = 1.1, and N = 1.85. The green and orange curves
494 on each of the graphs are the isoclines. The isoclines are all of the points at
495 which seed density or tree density are not changing (i.e., s = 0 or p = 0).
496 Where these curves intersect, the system is in equilibrium because s = 0 and
497 p = 0. An equlibrium may be stable, also known as an attractor, or unstable,
498 also known as a repellor. An attractor is a collection of equillibria to which
499 a system is likely to return over the long-term if perturbed. A repellor is a
500 collection of equillibria that a system is likely to move away from over the
501 long-term, if perturbed.
502 Figure 3a has one attractor where the isoclines intersect (s = 1, p = 1)
503 and, thus, one stable state. The stable state is pristine because, at N = 0,
504 humans do not have an ecological footprint in the system. The stable state
505 is globally resilient. Resilience refers to how much environmental change a
506 system can absorb and still maintain its feedback structure (Holling, 1973).
507 For example, the black arrowed curves on Figure 3 illustrate how a system
508 evolves over time for a given set of initial conditions. To illustrate, please
509 follow the curve labeled I1 on Figure 3a. This system starts at a seed
510 density of s = 0.25, and a tree density of p = 0.3. Over time, the trees
511 produce seeds, which sprout and grow into trees and the forest recovers
512 toward the pristine state (s = 1, p = 1). In the long-run, the stable state in
513 this environment is globally resilient because no matter where we start the
514 system, it will always move toward s = 1, p = 1.
515 Please hold-on to the idea that resilience is a long run property of an at-
516 tractor. In real systems, perturbations constantly occur, for example, a pest

19
 1 (a) (d)
Seed density (s)

0.8

0.6

0.4

I1
0.2

0
0 0.2 0.4 0.6 0.8 1

Tree density (p)

1 (b) (e)
Seed density(s)

0.8

0.6

0.4

I2
0.2
I3

0
0 0.2 0.4 0.6 0.8 1

Tree density(p)
1 (c) (f)
Seed density (s)

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1

Tree density (p)

Figure 3: Phase plots: All parameters are held constant at default values except population
(N ). a-N = 0; b-N = 1.1; c-N = 1.85. The 20circles highlight stable attractors. The square
in graph b highlights a repellor. Each phase plot is paired with a stability landscape (a-d;
b-e; c-f).
517 outbreak in a forest that pushes tree and seed biomass out of equilibrium.
518 The black arrowed curves on Figure 3 help us imagine the consequences
519 of perturbations. In the case of the curve labeled I1 , we can imagine that
520 a pest outbreak killed-off a bunch of trees in one year, thus setting the initial
521 conditions for the next year at s = 0.25, p = 0.3. In the short-run, the pest
522 outbreak depressed the biomass of the forest, but, at longer time scales (over
523 a decade), the forest moves toward the pristine attractor. In a real system,
524 biomass may never actually get to the pristine state because perturbations
525 are constant. However, the system is globally resilient because, over time, it
526 continues to move back toward our pristine stable state. Thus, the black
527 arrowed curves help us understand the sets of initial conditions for which the
528 forest will tend to evolve toward a particular attractor.
529 In Figure 3b, we add some humans to the ecological system (N = 1.1).
530 Now, the subsistence decisions of individual households modify the ecosys-
531 tem through the harvest of seeds and/or investment in producing maize. In
532 turn, the modified ecosystem changes the net benefits of investing (or not)
533 in farming for all households. There are two consequences.
534 First, the pristine attractor is no longer pristine; the attractor moves
535 down and to the left. This attractor moves down and to the left because the
536 subsistence decisions of individual household scale-up to change the com-
537 position of the ecosystem. At this productive attractor, it is optimal for
538 households to expend 12 % of their total available effort (i.e., 12 % of their
539 subsistence time-budget) to gather seeds; and this allocation of effort allows
540 households to achieve their daily caloric goal (K) in a minimum amount of
541 time. In the productive state, households do not work very long and can
542 invest a lot of time in other activities that enhance fitness.
543 Second, when N = 1.1, the system displays multiple stable states. These
544 multiple stable states emerge because the niche constructing activities of
545 households changes the long-term composition of the ecosystem. In this
546 modified ecosystem the harvest behavior of households (collecting seeds and
547 clearing trees to grow maize) introduces the risk that a perturbation may
548 drive the system into an impoverished attractor. The impoverished attrac-
549 tor on Figure 3b is separated from the productive attractor by a repellor
550 (the blue square). In this multiple stable state environment, now if a pest
551 outbreak reduces tree and seed biomass to the point labeled I3 , the sys-
552 tem will move toward the impoverished attractor over the long-run. Such a
553 flip is not simply caused by the perturbation (pest outbreak), however, but
554 is a result of individual households modifying the ecosystem and the modi-

21
555 fied ecosystem, in tandem with the perturbation, feeding back to change the
556 optimal allocation of effort to collecting seeds and cultivating maize.
557 To explicate the dynamics in this environment a bit more, please consider
558 the following what, if scenario. Again, if a pest outbreak were to occur and
559 decrease the biomass of the forest to the initial condition labeled I2 , we
560 would observe the following dynamics. Individual households would adjust
561 their allocation of effort. At I2 the mean household would expend 62 % of
562 their time budget collecting seeds and 4 % clearing forest, planting and har-
563 vesting maize. Over time (absent any perturbations), seed biomass initially
564 declines, and the mean household realizes that it is optimal to cultivate more
565 because the marginal returns from foraging are so low. The time devoted to
566 cultivation peaks at 47 % and gathering seeds bottoms out at 14 %. However
567 tree biomass increases over time because forest reproduction is higher than
568 the rate of clearing to produce maize. Due to this, the abundance of seeds
569 begins to increase again as well, and the marginal gain from gathering seeds
570 increases. As the system moves toward the productive attractor, the mean
571 household in the model decreases its allocation of labor to maize and in-
572 vests more effort in collecting seeds until the productive attractor is reached.
573 The insight here is that it is optimal for households to produce maize to
574 compensate for short-term declines in the productivity of seeds caused by a
575 perturbation. However, this very act creates the vulnerability that the com-
576 bination of growing more maize to compensate for a decline in the marginal
577 return of seeds and a external perturbation will decrease the biomass of the
578 forest past a state variable threshold.
579 If the biomass of the forest decreases past such a threshold (all sets of
580 (s,p) left of the dashed blue line), a feedback loop between seed gathering and
581 clearing is initiated. Lower tree density leads to fewer seeds which means it
582 is optimal for households to clear more land to plant maize. The short term
583 decision to clear land leads to lower tree density which reduces seed density
584 over the longer term and the system moves toward the impoverished attrac-
585 tor. One can imagine how this feedback cycle could lead to the destruction
586 of the forest.
587 However, at N = 1.1 the impoverished state is only impoverished in terms
588 of the ability of households to collect seeds. Each household can still obtain
589 K; they simply are doing it by working a lot harder than they do in the
590 productive state. At the impoverished attractor, households spend 71 % of
591 their subsistence budget cultivating maize and 29 % gathering seeds.
592 In short, when N = 1.1, the system displays multiple stable states, which

22
593 emerge because households modify the long-term composition and produc-
594 tivity of the ecosystem. In the productive state, households are foragers who
595 need not work very long. In the impoverished state, households are commit-
596 ted farmers and work long hours. Note: if the system were hit by a large
597 enough perturbation, we would observe a regime shift from foraging at the
598 productive attractor to fully committed farming at the impoverished attrac-
599 tor! There is more to this story, however. The impoverished attractor is not
600 very resilient to environmental change.
601 Figure 3c illustrates how sensitive the impoverished attractor is to in-
602 creases in the ecological footprint of households. In this setting, we have
603 increased N to 1.85 and the system bifurcates (i.e., crosses a parameter
604 threshold). The only attractor is the impoverished attractor, very near s = 0,
605 p = 0. The destruction of the forest occurs, in part, because of the feedback
606 noted above where clearing decreases tree and seed biomass, which makes
607 cultivating a more optimal strategy so people clear more forest. The other
608 assumption that leads to the deconstruction of the forest is that soil fertility
609 is proportional to tree biomass. As the biomass of trees declines, the yield
610 of maize per unit area declines. The consequence: people need to clear ever
611 more area to meet their desire for maize, but the fertility of the soil is just
612 too low to ever get enough food to meet K. Hence, once people flip to the
613 impoverished attractor, any small increase in population or decline in the
614 productivity of trees (rs & rp ) will cause large declines in the productivity of
615 maize and decimate the productivity of the ecosystem.
616 In sum, Figure 3 illustrates that the niche construction activities of house-
617 holds changes the composition and productivity of the ecosystem, which, in
618 turn, affects the optimal decisions of household. The consequence of this is
619 changes in population density do not scale linearly or even continuously with
620 changes in investment in farming. Rather, at a critical threshold of popu-
621 lation density, about N = 1.1 above, multiple stable states emerge and all
622 households in the system may experience a regime shift from obtaining all
623 of their diet from seeds (foraging) to obtaining almost all of their diet from
624 maize (farming). The system has the potential to display non-linear transi-
625 tions from a state in which foraging activities predominate how people use
626 their time to a state in which farming activities predominate in how people
627 use their time to produce food.

23
628 4.2. The long-run effect of low-level farming
629 The emergence of multiple stable states has a subtle but profound con-
630 sequence. It is often optimal for a household to engage in low-level farming
631 (i.e., allocate a small proportion of time to growing maize) at the produc-
632 tive attractor; however, when households adopt low-level farming, the system
633 looses resilience and becomes more vulnerable to a flip from the productive
634 to the impoverished state.
635 Figure 4 is a bifurcation diagram analogous to Figure 1c. This diagram
636 is a tool for illustrating how, holding other parameters equal, N controls
637 the structure of the system. On the x-axis we have plotted N , and on the
638 y-axis we have plotted the percent of diet obtained from collecting seeds
639 (hs /K) when the system is at equilibrium (i.e., s = 0 and p = 0). The
640 thick black curves are all of the productive and impoverished attractors that
641 would emerge for given values of N . The dashed curved is the collection of
642 repellors that emerge for given values of N and separates the productive and
643 impoverished curves. The shaded area defines all of the values of N for which
644 the system has three equllibria (two attractors and a repellor).
645 Figure 4 demonstrates that low-level food production is often optimal
646 over the short term, but also creates conditions that may lead to the rapid
647 and difficult to reverse process of adopting agriculture at the expense of
648 hunting and gathering over the longer run. Reading from left-to-right, as N
649 increases, households in the productive state (the top left branch) remain
650 pure foragers until a threshold of about N = 1.25. At N = 1.25 households
651 begin to cultivate maize as a minor component of their diet at the productive
652 attractor, and the proportion of the diet from foraging begins to decline.
653 Once households begin to optimize by cultivating maize, small increases in N
654 lead to fast decreases in the proportion of the diet that comes from foraging
655 (the productive harvest branch bends down and to the right). However,
656 foragers can only replace a maximum of about 40 % of their diet with maize
657 before the system bifurcates (crosses a parameter threshold), and the only
658 attractor is the impoverished state.
659 It is intuitive that households can replace declining returns from seeds
660 with maize with little disruption to their foraging system over the short run
661 (Barlow, 2006). However, cultivating maize in response to declining returns
662 from foraging (caused here by increases in N relative to rp and rs ) means
663 that the system is more likely to flip over the long-run (i.e. cross a state
664 variable threshold). Given enough time, a big enough shock is sure to occur
665 in most environments. To elaborate: we can estimate the resilience of the

24
 1.00 productive harvest branch

6
Proportion of diet from seeds (hs /K)

0.75
6

The probability of a flip to the alternate at-

0.50 tractor (or regime shift) increases as the dis-
tance between the stable productive branch
and unstable branch decreases (double arrow ?
lines)
?

unstable harvest branch

0.25

0.00 impoverished harvest branch

0.0 0.5 1.0 1.5 2.0

Population density, N (households/area)

Figure 4: Summary of the relationship between N and the proportion of the diet obtained
from seeds at equilibrium. The upper solid line represents all of the productive attractors;
the dashed middle line, all repellors (unstable branch), and the lower solid line all of the
impoverished attractors. Shaded area defines the parameter space in which a perturbation
can cause the system to flip between attractors.

666 productive attractor by estimating the size of the attractor. We estimate size
667 by calculating the Euclidean distance between the productive attractor and
668 the unstable harvest branch (the repellor) in Figure 4 (Anderies et al., 2002).
669 As the size of the productive attractor declines, the probability that a per-
670 turbation induces a regime shift increases. The point here is that a one unit
671 decline in the proportion of diet from foraging, at the productive attractor,

25
672 induces a decline in the size of the productive attractor that is greater than
673 one. In short, increases in population density, even very small increases,
674 increase the susceptibility of the system to a regime change caused by a
675 perturbation, such as a drought. A N goes up, a perturbation of a constant
676 size is more likely to induce a regime shift from the productive to the im-
677 poverished attractor because households supplement foraging with farming,
678 which transforms the composition and productivity of the ecosystem.
679 In sum, the optimal allocation of effort to foraging and farming can make
680 a subsistence system that includes low-level farming vulnerable to a flip
681 caused, in part, by the very rules that allow individuals optimize their pro-
682 duction of food. If the system flips into the impoverished state, the ecosystem
683 is vulnerable to degradation. The implication is that if a system flips, house-
684 holds either abandon the area or engage in classic Boserupian responses.
685 Boserupian responses are strategies that invest labor in new niche construc-
686 tion activities and increase the productivity of land per unit area, as well
687 as decrease the need to further harvest trees (thereby improving soil fer-
688 tility). Hence, at the productive attractor, it is likely not very costly to
689 increase or decrease effort in farming as perturbations (droughts, etc.) dic-
690 tate year-to-year. However, if environmental variation initiates a flip into
691 the impoverished state, households have to leave or lock-in to a run away
692 niche construction strategy that mitigates the loss of the forest and declines
693 in soil fertility. A central issue is whether households might ascertain from
694 their knowledge of the environment that the risk of getting pushed into the
695 impoverished state is a novel feature of their social and ecological world. If
696 so, they might be able to adapt to longer-term variation and modify how
697 they allocate effort in the short term; if not: look out!

698 5. So What?
699 Our work is motivated by a basic question. Why, once domesticated
700 plants become available do some people increase their time invested in farm-
701 ing at the expense of hunting and gathering while others do not? One way to
702 answer this question is to use the tools of foraging theory and ask: When does
703 an individual maximize their fitness by investing in farming at the expense
704 of foraging, and this has been a productive line of research (e.g., Barlow,
705 2002; Keegan, 1986; Kennett and Winterhalder, 2006; Layton et al., 1991;
706 Tucker, 2001). However, the foraging models that anthropologists typically
707 use can benefit from further development (e.g., Winterhalder and Goland,

26
708 1997; Winterhalder et al., 1988). Foraging models, often, do not account
709 for the feedbacks caused by niche construction and, thus, the process of
710 ecological inheritance. Ecological inheritance is known to cause non-linear
711 trajectories of evolution. An important possibility is that ecological inher-
712 itance may result in multiple stable states, or ways that humans could be
713 organized to obtain food in a given environment. Multiple stable states can
714 lead to transitions that are punctuated, difficult to anticipate, and very costly
715 to reverse.
716 Therefore, we have combined an optimization model with a dynamical
717 systems model to study the effects of niche construction and ecological in-
718 heritance. We asked two specific questions. Under what ecological and social
719 conditions might the optimal allocation of effort to foraging and low-level
720 farming cause multiple stable states to emerge, and if they do, what vari-
721 ables should create different trajectories that the evolution of investment in
722 farming follows, especially the rate of specialization in farming? The analy-
723 sis of our model provides an answer to the first question, and creates a more
724 informed position to begin to answer the second question taking into consid-
725 eration how niche construction modifies ecosystems over time, and thus the
726 optimal way to allocate effort.
727 The model generates multiple stable states when the ecological footprint
728 of households increases beyond a critical threshold (Figures 3b & 4). The
729 emergence of multiple stable states has four implications.
730 1. When a household behaves optimally in the short-run, this act exposes
731 all of the households to the risk of flipping from a productive state to
732 an impoverished state over the longer run.
733 2. The risk of the system flipping emerges when households are foragers,
734 which opens up the possibility that households may suddenly transition
735 from full-time foraging to full time farming, in terms of their allocation
736 of labor and dependence on a domesticate for calories. However, the full
737 time farming at our impoverished attractor is very sensitive to increases
738 in population density or declines in the productivity of wild resources.
739 Any slight change will cause the system to move toward a completely
740 denuded state.
741 3. The adoption of low-level food production in the productive state makes
742 the whole system more vulnerable to perturbations. The implication
743 is that low-level food production is costly to maintain. This is because
744 the optimal allocation of effort to collecting seeds and cultivation max-

27
745 imizes fitness in the short-run. But, a low-level of farming changes the
746 composition of the ecosystem over the longer-run. Thus, the risk of flip-
747 ping to the impoverished state, and the fitness consequences that come
748 along with such a flip, are inherited through the modified ecosystem.
749 4. The dynamics of our model provide a theoretical basis for defining
750 low-level farming with domesticates as a space defined by a diet be-
751 tween 1 and 40% dependence upon domesticated plants. Note: The
752 40% threshold is not a universal. The exact value would vary from
753 system to system, depending on the specifics of particular ecosystems.
754 The key is the qualitative insight that there is a parameter threshold,
755 and it will likely fall somewhere between 1 and 40% in many particular
756 systemsunless one changes how foraging and farming activities modify
757 the ecosystem (a future direction for research noted below). Inciden-
758 tally, there is empirical evidence that major changes in the way house-
759 holds organize labor occurs cross-culturally at a threshold of about 35%
760 dependence on domesticated plants (Freeman, 2012a,b).
761 The issue now is translating, in a sense, from the insights afforded by
762 the model to evaluating variation in levels of investment in farming, either
763 archaeologically or ethnographically, to see how far we can generalize. A
764 test of any argument built from our model would require a comparison with
765 alternative models (Freeman and Anderies, 2015), and this is beyond the
766 scope of our current paper. However, we would like to end our discussion
767 with an example to help illustrate how insights drawn from our model can
768 be used to guide archaeological research.

769 5.1. A case study

770 The Colorado Plateau is an arid region with elevations ranging between
771 1,500 and 3,000 meters. The region is drained by the Colorado River and its
772 tributaries, and is also dissected by numerous small perennial and seasonal
773 drainages. Many of these drainages form deeply incised canyons bordered
774 by high upland mesa tops. The upland mesa tops are, in most locations,
775 dominated by pi nyon and juniper trees. Over the course of prehistory, these
776 highland mesa tops became the primary settings occupied by prehistoric
777 farmers across much of the Plateau.
778 The Colorado Plateau spans two archaeological cultures; the Ancestral
779 Puebloan and the Fremont. Both Ancestral Puebloan and Fremont popu-
780 lations engaged in farming. One major distinction, however, is that while

28
781 populations in the Ancestral Puebloan region were committed agricultural-
782 ists by about A.D. 700 or earlier in some places, populations in the Fremont
783 region relied on hunting and gathering, in many locations, throughout pre-
784 history (Barlow, 2002; Madsen and Simms, 1998; Simms, 2008). Thus, while
785 the transition to agricultural dependence appears to have been relatively
786 complete in the Ancestral Puebloan region, the occupation of the Fremont
787 region represents more of a mosaic with different strategies being used in
788 different places over the long term. Our concern here is not with the initial
789 arrival of maize but with the potential for various levels of investment in
790 agriculture and the development of new strategies, especially extensive up-
791 land cultivation. The transition to upland cultivation occurred on several
792 parts of the Colorado Plateau, such as: the Mesa Verde/Northern San Juan
793 region, Cedar Mesa, portions of the Mogollon Rim, and areas along the Fre-
794 mont and Green Rivers. This transition was variably timed but primarily
795 occurred between about A.D. 200 and A.D. 750 in the four corners (Mabry,
796 2005, p. 128-129; Matson, 1991) and as late as A.D. 1000 further north in
797 portions of the Fremont region (Dodd, 1982; Talbot and Wilde, 1989).
798 Based on paleobotanical and settlement data, many researchers have sug-
799 gested that the prehistoric agricultural systems in the pi nyon-juniper wood-
800 land mesa tops along the Colorado Plateau involved a type of shifting culti-
801 vation. This method of cultivation involved the clearing of small field areas,
802 likely with fire, followed by planting in the cleared areas. This was followed
803 by a long fallow period when the woodland plots were allowed to regrow and
804 soil nutrients regenerate (e.g., Kohler, 1992a,b; Kohler and Matthews, 1988;
805 Matson, 1991; Matson et al., 1988; Peeples et al., 2006; Stiger, 1979; Sullivan,
806 1982; Varien, 1999). Local transitions from to the extensive cultivation of up-
807 land mesa tops are interesting because explanations for this change are not
808 straight forward. Extensive shifting cultivation is less efficient than many
809 other types of agriculture and foraging (Barlow, 2002, p. 80), it is risky
810 (Mabry, 2005, p. 135-136), and it may not be locally sustainable (Kohler,
811 1992a,b; Kohler and Matthews, 1988). So why the increased investment in
812 agriculture, especially the shift to extensive shifting cultivation?
813 In our model, once foragers begin to cultivate maize, even just a little
814 (e.g., 5 % of the diet), the consequence of this behavior is to make the whole
815 system at ever greater risk of a flip from the productive to the impoverished
816 state. The flip occurs when the rate of forest regeneration divided by the
817 rate of forest consumption crosses either a parameter threshold (e.g., caused
818 by increases in population density) or a state variable threshold. In this

29
819 second case, holding all other parameters equal, a shock pushes the system
820 into the impoverished attractor. This occurs because it appears optimal
821 for individuals to cultivate more in the short-run due to a decline in the
822 productivity of seeds, but doing so transforms the structure of the ecosystem
823 too much and everyone ends up working harder with a significant proportion
824 of maize in their diet. These dynamics suggest two real world variables that
825 should control the transition from a low investment in maize farming to
826 a higher investment: population density and the size of shocks that hit a
827 ecosystem (e.g., droughts).
828 The dynamics of the model described here allow us to develop at set of
829 expectations associated to changing levels of agricultural investment in re-
830 sponse to niche constructing activities and the emergence of multiple stable
831 states. First, we expect that the rate of change in the shift from the ini-
832 tial adoption of maize to a higher investment on the Colorado Plateau to
833 vary with both population density and the coefficient of variation in rainfall.
834 Holding rainfall (a proxy for productivity in this case) equal, an increase in
835 population density will make a flip more likely. Further, our model suggests
836 that this increase should be associated with a threshold-crossing and rapid
837 transition rather than a gradational change. Holding population density con-
838 stant, we also expect a higher coefficient of variation in rainfall to increase
839 the rapidity of a transition. Please note at this point that the households in
840 our model are stupid. By this we mean that they do not modify their niche
841 construction activities, even if the flip from the productive to the impover-
842 ished attractor. We do not expect real humans to act this way; we expect
843 that once a regime change begins to occur in the real world, individuals en-
844 gage in new forms of niche construction to manage a forest and reduce the
845 extreme uncertainty that would come with a total denuding of trees. This
846 may happen in some instances, but what we are likely to see is a rapid shift
847 from a high forest density equilibrium with little commitment to maize to
848 low forest density equilibrium with a high commitment to maize and much
849 more active management of an ecosystem.
850 In order to assess our expectations in an archaeological context, we would
851 first need to establish temporal control over assemblages on the various por-
852 tions of the Colorado Plateau with varying levels of population and variation
853 in rainfall. Evidence consistent with ideas developed from our model would
854 include: 1) Estimates of commitment to maize and estimates of forest struc-
855 ture change in tandem over time. 2) The rate of change in economic strategies
856 increases where population densities were initially higher when maize was in-

30
857 troduced, and where rainfall was more variable, more intensively shocking the
858 integrated forest and subsistence system. Testing such a model would involve
859 assessing agricultural commitment (using evidence such as the form and size
860 of maize grinding tools, maize ubiquities, etc.) in relation to viable estimates
861 of population densities (based on site/feature counts and/or absolute date
862 frequencies), as well as paleoclimatic and archaeological evidence for changes
863 in the structure of forests and ubiquities of wild resources along with changes
864 in the coefficient of variation in rainfall through time. For example, an ar-
865 chaeological question that follows from our line of reasoning is: In contexts
866 where we can observe synchronous changes in agricultural investment and
867 wild food availability, is the span of time between the earliest arrival of do-
868 mesticates and higher levels of agricultural investment shorter for regions
869 that had greater population density and/or variation in rainfall? If the an-
870 swer to this question is yes, this would be consistent with the assumptions
871 that underly our model.

872 6. Future Directions for Research & Conclusion

873 The model that we have constructed is a first attempt to more rigorously
874 analyze how niche construction can lead to the coevolution of the composition
875 of an ecosystem and the fitness of economic decision making. Much more
876 work is needed. Here we suggest a couple of possible directions for future
877 research.

878 1. In a recent paper in PNAS, Bowles (2011) asks why people ever increase
879 their commitment to farming (at least based on seed agriculture)? His
880 study of ethnographic societies documents that foraging for wild seeds
881 is more productive, in terms of energy return per person hour, than
882 cultivating domesticated seeds, at least over the time scales observed
883 by ethnographers. Thus, increased investment in farming is not in-
884 tuitive. One possible answer to this riddle is that the evolution of
885 property rights provides an opportunity for individuals to invest more
886 time in farming because 1) it has a higher potential to produce a sur-
887 plus than foraging for seeds and 2) property rights protect the fruits
888 of individuals labor from expropriation and, thus, provide an incen-
889 tive for individuals to work harder (Bowles and Choi, 2013). In short,
890 property rights may be necessary for increased investment in farming to
891 evolve. However, the comparative accounting framework developed by

31
892 Bowles (2011) does not account for the feedbacks between foraging and
893 farming decisions and the productivity of an ecosystem caused by how
894 people construct their niche over longer-time scales. Our analysis illus-
895 trates how a subsistence system might flip from one where more time
896 is allocated to foraging to one where more time is allocated to farming
897 (Figures 3b and 4). This is an important dimension of the problem. It
898 may be the case that human subsistence systems display well defined
899 clusters of subsistence strategies that are reflective of various attrac-
900 tors that are an outcome of niche construction activities in different
901 ecosystems (Ullah et al., 2015). More work is needed to explore these
902 possibilities, however, which leads us to point two.
903 2. Our model supposes that niche construction causes a particular direc-
904 tional effect on the productivity of wild resources and domesticates
905 (controlled by soil fertility). The directional effect is a double negative,
906 harvesting seeds and clearing the forest depletes the future produc-
907 tivity of trees and the soil. It is this dynamic that causes an inverse
908 relationship between the degree of dependence on domesticates and the
909 resilience of the integrated subsistence and ecosystems. The composi-
910 tion and productivity of an ecosystem need not change in this double
911 negative manor. The vary same behaviors that we posit in our model,
912 depending on the ecosystem, may lead to four possibilities in terms
913 of effects on the future productivity of seeds and farming: + +, + ,
914 +, and . An import question to ask is what changes to the
915 assumptions of the model developed above are necessary to generate
916 large basins of attraction for low level farming (i.e., makes low-level
917 farming resilient)? Using this strategy, the challenge would be to build
918 reasonable assumptions about the harvest behavior of forager-farmers
919 that generate the four possible scenarios listed above. This strategy al-
920 lows for the possibility of a more precise description of what happens in
921 each case (+ + vs. + ) in terms of the trajectories of change. As we
922 change the assumptions about how niche construction affects the pro-
923 ductivity of an ecosystem and the productivity feeds back to change the
924 fitness of different behaviors, the systems might generate very different
925 features from those observed in our paper, like oscillations or chaotic
926 attractors, as opposed to stable equillibria.
927 3. Another direction for research is to better theorize about how niche
928 construction activities at local scales may move across levels of orga-
929 nization and create cascades of change across large regions. Human

32
930 societies create social connections with each other that may link their
931 niche constructing activities across scales. To our knowledge, this has
932 not been a major topic of research with regards to the spread of agri-
933 culture.

934 In the end, we have shown that multiple stable states occur when we
935 include the process of ecological inheritance (caused by niche construction)
936 in a optimization model drawn from foraging theory/microeconomics. This
937 model illustrates that the optimality of individual foraging decisions over
938 shorter time scales and the resilience of a system over longer time scales are
939 both important for understanding the evolution of various levels of invest-
940 ment in farming. There has been a recent debate over whether HBE provides
941 an overarching framework for studying the adoption of domesticated plants
942 (Gremillion et al., 2014; Smith, 2015, 2011a; Zeder, 2015, 2012, 2006; see
943 also Flannery, 1985 & Winterhalder and Goland, 1997 for an earlier version
944 of the argument) and the argument may equally apply to our topic. We
945 argue that HBE can provide a useful overarching framework to understand
946 specialization in domesticates at the expense of hunting and gathering. Mi-
947 croeconomic models within the HBE framework, however, need extension
948 to include the feedback dynamics caused by niche construction, which may
949 cause non-intuitive trajectories of evolution. Dynamical systems models are
950 one tool for improving our understanding of such possibilities.

951 7. Acknowledgments
952 We gratefully acknowledge the support of the NSF Dynamic Coupled Natural
953 and Human Systems Program (#CNH-1113991). We thank Micheal OBrian,
954 Jim Roscoe, and an anonymous reviewer for insightful comments that helped
955 improve our manuscript. All mistakes are, of course, our own.

956 References
957 Anderies, J. M., 1998. Culture and human agro-ecosystem dynamics: The
958 Tsembaga of New Guinea. Journal of Theoretical Biology 192 (4), 515530.

959 Anderies, J. M., 2006. Robustness, institutions, and large-scale change in

960 social-ecological systems: the Hohokam of the Phoenix Basin. Journal of
961 Institutional Economics 2 (2), 133155.

33
962 Anderies, J. M., Janssen, M., Walker, B., 2002. Grazing management, re-
963 silience, and the dynamics of a fire driven rangeland system. Ecosystems
964 5, 2344.
965 Barker, G., 2006. The agricultural revolution in prehistory :why did foragers
966 become farmers? Oxford University Press, Oxford; New York.
967 Barlow, R. K., 2002. Predicting Maize Agriculture among the Fremont: An
968 economic comparison of farming and foraging in the American Southwest.
969 American Antiquity 67 (1), 6588.
970 Barlow, R. K., 2006. A formal Model for Predicting Agriculture among the
971 Fremont. In: Kennett, D. J., Winterhalder, B. (Eds.), Behavioral Ecology
972 and the Transition to Agriculture. University of California Press, Berkeley,
973 CA, pp. 87102.
974 Bender, B., 1978. Gatherer-hunter to farmer: A social perspective. World
975 archaeology 10 (2), 204222.
976 Binford, L. R., 1968. Post-Pleistocene Adaptations. In: Binford, S. R., Bin-
977 ford, L. R. (Eds.), New Perspectives in Archaeology. Aldine, Chicago, pp.
978 313341.
979 Binford, L. R., 2001. Constructing frames of reference: An analytical method
980 for archaeological theory building using hunter- gatherer and environmen-
981 tal data sets. University of California Press, Berkeley.
982 Bird, D. W., OConnell, J. F., 2006. Behavioral ecology and archaeology.
983 Journal of Archaeological Research 14 (2), 143188.
984 Bleed, P., Matsui, A., 2010. Why didnt agriculture develop in Japan? A
985 consideration of Jomon ecological style, niche construction, and the origins
986 of domestication. Journal of Archaeological Method and Theory 17 (4),
987 356370.
988 Bowles, S., 2011. Cultivation of cereals by the first farmers was not more
989 productive than foraging. Proceedings of the National Academy of Sciences
990 108 (12), 47604765.
991 Bowles, S., Choi, J.-K., 2013. Coevolution of farming and private property
992 during the early Holocene. Proceedings of the National Academy of Sci-
993 ences 110 (22), 88308835.

34
 994 Brown, G. R., Richerson, P. J., 2014. Applying evolutionary theory to human
995 behaviour: past differences and current debates. Journal of Bioeconomics
996 16 (2), 105128.

997 Childe, V. G., 1928. The most ancient East:The oriental prelude to European
998 prehistory. K. Paul, Trench, Trubner & Co. ltd, London.

999 Clark, C. W., 1976. Mathematical bioeconomics: The optimal management

1000 of renewable resources. Wiley, New York.

1001 Codding, B. F., Bird, D. W., 2015. Behavioral ecology and the future of
1002 archaeological science. Journal of Archaeological Science 56, 920.

1003 Dodd, W. A., J., 1982. Final Year Excavations at the Evans Mound Site.
1004 University of Utah Anthropological Papers 106. University of Utah Press,
1005 Salt Lake City.

1006 Ellen, R. F., 1978. Nuaulu Settelment and Ecology. Martinus Nijhoff, The
1007 Hague.

1008 Flannery, K. V., 1968. Archaeological Systems Theory and Early Mesoamer-
1009 ica. In: Meggers, B. J. (Ed.), Anthropological Archaeology in the Ameri-
1010 cas. Anthropological Society of Washington, Washington D.C., pp. 6787.

1011 Flannery, K. V., 1973. The origins of agriculture. Annual Review of Anthro-
1012 pology 2, 271310.

1013 Flannery, K. V., 1985. Guila Naquitz: Archaic foraging and early agriculture
1014 in Oaxaca, Mexico. Academic Press, Orlando.

1015 Freeman, J., 2012a. Alternative adaptive regimes for integrating foraging and
1016 farming activities. Journal of Archaeological Science 39 (9), 3008 3017.

1017 Freeman, J., 2012b. Crop richness in human subsistence agricultural systems:
1018 a test of macroecological and economic determanents. Global Ecolology
1019 and Biogeography 21, 428440.

1020 Freeman, J., Anderies, J. M., 2012. Intensification, Tipping Points and Social
1021 Change in a Coupled Forager-Natural Resource System. Human Nature 23,
1022 419446.

35
1023 Freeman, J., Anderies, J. M., 2015. A comparative ethnoarchaeological anal-
1024 ysis of corporate territorial ownership. Journal of Archaeological Science
1025 54, 135147.

1026 Gifford, E. W., 1936. Northeastern and Western Yavapai. University of Cal-
1027 ifornia publications in American Archaeology and Ethnology 34, (4).

1028 Gil, A. F., Neme, G. A., Tykot, R. H., 2011. Stable isotopes and human
1029 diet in central western Argentina. Journal of Archaeological Science 38,
1030 13951404.

1031 Gremillion, K. J., Barton, L., Piperno, D. R., 2014. Particularism and the
1032 retreat from theory in the archaeology of agricultural origins. Proceedings
1033 of the National Academy of Sciences 111 (17), 61716177.

1034 Hard, R. J., Mauldin, R. P., Raymond, G. R., 1996. Mano size, stable carbon
1035 isotope ratios, and macrobotanical remains as multiple lines of evidence of
1036 maize dependence in the American Southwest. Journal of Archaeological
1037 Method and Theory 3 (3), 253318.

1038 Harris, D. R., 1996. The Origins and Spread of Agriculture and Pastoralism
1039 in Eurasia. UCL Press, London.

1040 Hayden, B., 1990. Nimrods, Piscators, Pluckers, and Planters - The emer-
1041 gence of food-production. Journal of Anthropological Archaeology 9 (1),
1042 3169.

1043 Hayden, B., 1998. Practical and prestige technologies: The evolution of mate-
1044 rial systems (Design theory). Journal of Archaeological Method and Theory
1045 5 (1), 155.

1046 Holling, C. S., 1973. Resilience and stability of ecological systems. Annual
1047 Review of Ecology and Systematics 4, 123.

1048 Holmberg, A. R., 1950. Nomads of the long bow: the Siriono of eastern
1049 Bolivia. U.S. Govt. Print. Off., Washington.

1050 Huckell, B. B., 1995. Of marshes and maize: preceramic agricultural settle-
1051 ments in the Cienega Valley, southeastern Arizona. University of Arizona
1052 Press, Tucson, AZ.

36
1053 Hunter-Anderson, R. L., 1986. Prehistoric adaptation in the American South-
1054 west. Cambridge University Press, Cambridge Cambridgeshire; New York.

1055 Janssen, M. A., Kohler, T. A., Scheffer, M., 2003. Sunk-cost effects and
1056 vulnerability to collapse in ancient societies1. Current anthropology 44 (5),
1057 722728.

1058 Keegan, W. F., 1986. The optimal foraging analysis of horticultural produc-
1059 tion. American Anthropologist 88 (1), 92107.

1060 Kennett, D. J., Winterhalder, B., 2006. Behavioral ecology and the transition
1061 to agriculture. University of California Press, Berkeley.

1062 Kohler, T. A., 1992a. Field houses, villages, and the tragedy of the commons
1063 in the early northern Anasazi Southwest. American Antiquity 57 (4), 617
1064 635.

1065 Kohler, T. A., 1992b. Prehistoric human impact on the environment in the
1066 upland North American southwest. Population and Environment 13 (4),
1067 255268.

1068 Kohler, T. A., Matthews, M. H., 1988. Long-term Anasazi land use and forest
1069 reduction: A case study from southwest Colorado. American Antiquity
1070 53 (3), 537564.

1071 Kroeber, A. L., 1953. Handbook of the Indians of California. Vol. 78. Cali-
1072 fornia Book Company, Berkeley.

1073 Lade, S. J., Tavoni, A., Levin, S. A., Schl

uter, M., 2013. Regime shifts in a
1074 social-ecological system. Theoretical ecology 6 (3), 359372.

1075 Laland, K. N., OBrien, M. J., 2010. Niche construction theory and archae-
1076 ology. Journal of Archaeological Method and Theory 17 (4), 303322.

1077 Layton, R., Foley, R., Williams, E., 1991. The transition between hunting and
1078 gathering and the specialized husbandry of resources: A socio-ecological
1079 approach. Current Anthropology, 255274.

1080 Lever, J. J., Nes, E. H., Scheffer, M., Bascompte, J., 2014. The sudden
1081 collapse of pollinator communities. Ecology letters 17 (3), 350359.

37
1082 Mabry, J. B., 2005. Diversity in early southwestern farming and optimization
1083 models of the in transition to agriculture. In: Diehl, M. (Ed.), Subsistence
1084 and Resource use Strategies of Early Agricultural Communities in South-
1085 ern, AZ. Center for Desert Archaeology, Tucson, AZ, pp. 113152.
1086 MacNeish, R. S., 1992. The origins of agriculture and settled life. University
1087 of Oklahoma Press, Norman.
1088 Madsen, D. B., Simms, S. R., 1998. The Fremont complex: A behavioral
1089 perspective. Journal of World Prehistory 12 (3), 255336.
1090 Matson, R. G., 1991. The origins of Southwestern Agriculture. University of
1091 Arizona Press, Tucson, AZ.
1092 Matson, R. G., Lipe, W. D., IV, W. R. H., 1988. Adaptational continuities
1093 and occupational discontinuities: The Cedar Mesa Anasazi. Journal of
1094 Field Archaeology 15 (3), 245264.
1095 Merrill, W. L., Hard, R. J., Mabry, J. B., Fritz, G. J., Adams, K. R., Roney,
1096 J. R., MacWilliams, A. C., 2009. The diffusion of maize to the southwestern
1097 United States and its impact. Proceeding of the National Acadademy of
1098 the United States of America 106 (50), 2101921026.
1099 Nichols, D. L., 1987. Risk and agricultural intensification during the Forma-
1100 tive Period in the northern Basin of Mexico. American Anthropologist 89,
1101 596616.
1102 Odling-Smee, F. J., Laland, K. N., Feldman, M. W., 2003. Niche construction
1103 :the neglected process in evolution. Vol. 37. Princeton University Press,
1104 Princeton, N.J.; Oxford.
1105 Odling-Smee, J., Erwin, D. H., Palkovacs, E. P., Feldman, M. W., Laland,
1106 K. N., 2013. Niche construction theory: a practical guide for ecologists.
1107 The Quarterly Review of Biology 88 (1), 328.
1108 Odling-Smee, J., Laland, K. N., 2011. Ecological inheritance and cultural
1109 inheritance: what are they and how do they differ? Biological Theory
1110 6 (3), 220230.
1111 Peeples, M. A., Barton, C. M., Schmich, S., 2006. Resilience lost: Intersecting
1112 land use and landscape dynamics in the prehistoric Southwestern United
1113 States. Ecology and Society 11 (2).

38
1114 Phillips Jr, D. A., 2009. Adoption and intensification of agriculture in the
1115 north american southwest: Notes toward a quantitative approach. Ameri-
1116 can antiquity, 691707.
1117 Piperno, D. R., Pearsall, D. M., 1998. The origins of agriculture in the low-
1118 land neotropics. Academic Press, San Diego.
1119 Rindos, D., 1980. Symbiosis, Instability, and the Origins and Spread of Agri-
1120 culture: A New Model [and Comments and Reply]. Current Anthropology,
1121 751772.
1122 Rindos, D., 1984. The origins of agriculture: an evolutionary perspective.
1123 Academic Press, Orlando, Fla.
1124 Roscoe, P., 2002. The hunters and gatherers of New Guinea 1. Current An-
1125 thropology 43 (1), 153162.
1126 Rowley-Conwy, P., Layton, R., 2011. Foraging and farming as niche con-
1127 struction: stable and unstable adaptations. Philosophical Transactions of
1128 the Royal Society B: Biological Sciences 366 (1566), 849862.
1129 Scheffer, M., 2009. Critical Transitions in Nature and Society. Princeton
1130 University Press, Princeton, N.J.; Oxford.
1131 Simms, S. R., 2008. Ancient Peoples of the Great Basin and Colorado
1132 Plateau. Left Coast Press, Inc., Walnut Creek, CA.
1133 Smith, B., 2011a. A cultural niche construction theory of initial domestica-
1134 tion. Biological Theory 6, 260271.
1135 Smith, B. D., 1995. The emergence of agriculture. Vol. 54. Scientific American
1136 Library, New York.
1137 Smith, B. D., 2001. Low-level food production. Journal of Archaeological
1138 Research 9 (1), 143.
1139 Smith, B. D., 2007. Niche construction and the behavioral context of plant
1140 and animal domestication. Evolutionary Anthropology 16 (5), 188199.
1141 Smith, B. D., 2011b. General patterns of niche construction and the man-
1142 agement of wildplant and animal resources by small-scale pre-industrial
1143 societies. Philosophical Transactions of the Royal Society B: Biological
1144 Sciences 366 (1566), 836848.

39
1145 Smith, B. D., 2015. A comparison of niche construction theory and diet
1146 breadth models as explanatory frameworks for the initial domestication of
1147 plants and animals. Journal of Archaeological Research, 148.

1148 Staver, A. C., Archibald, S., Levin, S., 2011. Tree cover in sub-saharan Africa:
1149 Rainfall and fire constrain forest and savanna as alternative stable states.
1150 Ecology 92 (5), 10631072.

1151 Steward, J. H., 1938. Basin-plateau aboriginal sociopolitical groups. Vol. 120.
1152 U.S. G.P.O., Washington.

1153 Stiger, M. A., 1979. Mesa Verde subsistence patterns from Basketmaker to
1154 Pueblo III. The Kiva 44 (2-3), 133144.

1155 Sullivan, A. P., 1982. Mogollon agrarian ecology. The Kiva 48 (1-2), 115.

1156 Talbot, R. K., Wilde, J. D., 1989. Giving form to the Formative: Shifting
1157 settlement patterns in the eastern Great Basin and northern Colorado
1158 Plateau. Utah Archaeology 1989, 318.

1159 Thomas, D. H., 1971. Prehistoric subsistence-settlement patterns of the reese

1160 river valley, central nevada. Unpublished ph.d. dissertation, University of
1161 California-Davis.

1162 Tucker, B. T., 2001. The behavioral ecology and economics of variation, risk,
1163 and diversification among Mikea forager-farmers of Madagascar. Ph.D. the-
1164 sis, University of North Carolina, Chapel Hill, NC.

1165 Ullah, I. A., Kuijt, I., Freeman, J., 2015. Toward a theory of punctuated
1166 subsistence change. Proceedings of the National Academy of Sciences, in
1167 press.

1168 Varien, M. D., 1999. Sedentism and Mobility in a Social Landscape. Univer-
1169 sity of Arizona Press, Tucson.

1170 Vierra, B. J., 2005. The Late Archaic across the borderlands: from foraging
1171 to farming. University of Texas Press, Austin.

1172 Wills, W. H., 1988. Early prehistoric agriculture in the American Southwest.
1173 School of American Research Press, Santa Fe, NM.

40
1174 Winterhalder, B., Baillargeon, W., Cappelletto, F., Daniel, I., Prescott, C.,
1175 1988. The population ecology of hunter-gatherers and their prey. Journal
1176 of Anthropological Archaeology 7 (4), 289328.

1177 Winterhalder, B., Goland, C., 1993. On population, foraging efficiency, and
1178 plant domestication. Current Anthropology 34 (5), 710715.

1179 Winterhalder, B., Goland, C., 1997. An evolutionary ecology perspective on

1180 diet choice. In: Gremillion, K. J. (Ed.), People, plants, and landscapes:
1181 Studies in paleoethnobotany. University of Alabama Press, Tuscaloosa,
1182 AL, pp. 123162.

1183 Winterhalder, B., Smith, E. A., 2000. Analyzing adaptive strategies: Human
1184 behavioral ecology at twenty-five. Evolutionary Anthropology Issues News
1185 and Reviews 9 (2), 5172.

1186 Zeder, M., 2006. Central questions in the domestication of plants and ani-
1187 mals. Evolutionary Anthropology: Issues, News, and Reviews 15 (3), 105
1188 117.

1189 Zeder, M., 2012. The broad spectrum revolution at 40: resource diversity, in-
1190 tensification, and an alternative to optimal foraging explanations. Journal
1191 of Anthropological Archaeology.

1192 Zeder, M. A., 2015. Core questions in domestication research. Proceedings of

1193 the National Academy of Sciences, 201501711.

1194 Zeenah, D. W., 2002. Central place foraging and prehistoric pinyon utilization
1195 in the great basin. In: Fitzhugh, B., Habu, J. (Eds.), Beyond Foraging and
1196 Collecting: Evolutionary Change in Hunter-Gatherer Settlement Systems.
1197 Klewer Academic/Plenum Publishers, New York, pp. 231256.

1198 Zvelebil, M., 1996. The agricultural frontier and the transition to farming
1199 in the Circum-Baltic. In: Harris, D. R. (Ed.), The Origins and Spread
1200 of Agriculture and Pastoralism in Eurasia. Smithsonian Institution Press,
1201 Washington DC, pp. 323345.

1202

41
Box 1: Constrained Optimization
We solve the constrained optimization problem to determine how households should allocate effort
to minimize the total effort required to meet K. First, we form the Lagrangian L(lc , ls , )

L(lc , ls , ) = lc + ls (Ac (p)lc + As (s)ls K) (12)

which yields the first order conditions

L 1
= 1 Ac (p) = 0 = (13)
lc Ac (p)
L 1
= 1 (As (s)ls(1) = 0 = (1)
. (14)
ls As (s)ls

Both equation (13) and (14) define . Equating these expressions for yields the following expression

Ac (p) = As (s)ls(1) (15)

which relates the labor allocated to seed gathering, ls , and the system state, p and s. This relation
simply says that at the optimal value of ls , the marginal productivity of labor in agriculture and
gathering are equal just as in equation 3. Some algebra yields an expression for the optimal amount
of labor to be directed toward gathering:
1
( 1
 )
As (s)
ls = (16)
Ac (p)
Note that the expression above is insensitive to the fact that ls may be greater than the total labor
available, l, or whether K is met. To account for these constraints, we define the unconstrained
optimal labor allocation given in equation (16) as ls and the constrained optimum, described below,
as ls . Applying the subsistence constraint yields

K As (s)(ls )
 
lc = max ,0 (17)
Ac
where lc is the constrained optimal labor alloction to cultivation. If lc is zero, this means that ls
is sufficient to meet K and cultivation is forgone. Put another way, the point at which it pays to
devote the next unit of labor to cultivation occurs after output is higher than K. In this case,
 (1)
K
ls = (18)
As (s)
Finally, the labor constraint must be checked to ensure that lc + ls l. The conditions in (16) and
(18) tell a household to first determine the best allocation of effort to foraging and farming, then
follow the decision heuristic below.
1. If ls is sufficient to meet K (i.e. lc = 0), then attempt to gather seeds until K is met (i.e. set
ls = ls ). If the effort required to meet K via gathering is greater than the maximum (ls > l),
then devote all available available effort to gathering (set ls = l).
2. If ls is not sufficient to meet K, then set ls = min(ls , l). and lc = min(lc , max(l ls , 0)).
3. If the quantity of land to accommodate the effort allocated to cultivation by the rule above
is available, cultivate it. Otherwise, cultivate what is available and devote some effort to
clearing more land in the effort to meet 42 the required cultivable land area.
Box 2: Model Analysis
We provide additional background to interpret Figures 3 & 4. Our model is a non-
linear dynamical system. A non-linear dynamical system can be studied analytically
and/or numerically using software (in our case a program called XPPAUT). Both
approaches can be used to determine how many solutions there are in a system (i.e.
equillibria) and the qualitative stability of each solution. One determines qualitative
stability by linearizing the the non-linear system around equilibrium points. Only very
near equillibria can we treat the system as if it is linear. We have used XPPAUT to
numerically determine the eigenvalues. If any of the real parts of the eigenvalues > 0,
then an equilibrium is unstable. If the real parts of all the eigenvalues < 0, then the
equilibrium is stable. We have used this theory to determine whether an equilibrium
is an attractor or repellor.

For example, absent humans (N=0), our model has one equilibrium (see Figure 3a).
We can show analytically that, at this equilibrium, the long-run seed and tree densities,
denoted as s and p are s = rs Kp /ds and p = Kp . This means that when seed density
is not changing, the density of seeds is determined by a ratio: The production of seeds
divided by the decay of seeds. That is, if we double Kp , we double the seed density
because we are looking at twice the area. This is a common feature of mathematical
models. Ratios control the dynamics of the model. However, we must note that when
we choose the value of Kp , this defines the scale of other parameters. For example,
setting Kp = 1 and rs = 2 means that the seed generation rate of trees is twice as fast
as the rate at which seeds are used up.

A non-linear dynamical system has both global and local properties. When we say
that an equilibrium is either stable or unstable, we are talking about a local property.
When we say that our model has one or three equillibria, we are talking about a
global property of the system. For example, when N = 0, the equilibrium seed and
tree density (s = rs Kp /ds , p = Kp ) defines a stable equilibrium. This means that
if the system is perturbed it will always move back toward this point. The system
is globally stable because there is only one equilibrium. Figure 3d is a metaphor for
these statements. On the valley bottom, the slope of the landscape is zero-signifying
that the system is in equilibrium. The valley walls represent the slope or how quickly
a system returns to equilibrium if perturbed. In this case, there is one valley (basin of
attraction) signifying that there is only one equilibrium in the model. In Figure 3b, N
is equal to 1.1 and the global structure of this system is defined by three equillibria.
Two are locally stable (what we call the productive and impoverished states), the
other is unstable. Notice on Figure 3e that the system may flip from one valley to the
other. In Figure 3e the hill between the valleys represents the unstable equilibrium.
At the top of the hill the slope is zero (the equilibrium). Any small perturbation will
initiate feedback loops that lead the system into one of the adjacent valleys.

43