PLoS One. 2021;
16(8): e0254539. Virginia Ahedo, Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review &
editing,1 Débora Zurro, Conceptualization, Investigation, Methodology, Writing – original draft,2,* Jorge Caro,
Conceptualization, Investigation, Methodology, Validation, Writing – review & editing,1 and José Manuel Galán, Conceptualization, Formal analysis, Investigation, Methodology, Software,
Supervision, Visualization, Writing – review & editing1 Ulysses Paulino Albuquerque, Editor The transition to agriculture is regarded as a major turning point in human history. In the present contribution we propose to look at it through the lens of ethnographic data by means of a machine learning approach. More
specifically, we analyse both the subsistence economies and the socioecological context of 1290 societies documented in the Ethnographic Atlas with a threefold purpose: (i) to better understand the variability and success of human economic choices; (ii) to assess the role of environmental settings in the configuration of the different subsistence economies; and (iii) to examine the relevance of fishing in the development of viable alternatives to cultivation. All data were extracted from the
publicly available cross-cultural database D-PLACE. Our results suggest that not all subsistence combinations are viable, existing just a subset of successful economic choices that appear recurrently in specific ecological systems. The subsistence economies identified are classified as either primary or mixed economies in accordance with an information-entropy-based quantitative criterion that determines their degree of diversification. Remarkably, according to our results, mixed economies are
not a marginal choice, as they constitute 25% of the cases in our data sample. In addition, fishing seems to be a key element in the configuration of mixed economies, as it is present across all of them. The Origins of Agriculture (OA) is a mainstay of archaeological research, being the transition to farming regarded as one of the major developments in
our past [1–5]. Notwithstanding, after more than a hundred years of research on the OA, we are only just beginning to understand the details of the process
[6], remaining plenty of questions unanswered and being the topic still considered one the most relevant scientific challenges for Archaeology [7]. Even though the adoption of agriculture has often been
described as a rapid, unidirectional and inexorable process, the vast expansion of knowledge on the OA of the last two decades has shown that it is a complex phenomenon that encompasses a continuum of plant and animal management strategies; such a continuum frequently extended over long periods of time and involved an intricate interplay of environmental, social and cultural factors
[8]. Therefore, the OA is no longer considered as a single monolithic research question, being now recognised as a higher order research domain comprising a wide range of different research questions, datasets, scales of analysis and analytical and theoretical approaches
[9]. The most outstanding questions within the OA research field include its chronology and geography, as well as its causes, pace of development and spread, being all aspects deeply intertwined. Despite the existence—at a high level of detail—of almost as varied approaches to these questions as researchers writing about them, in more general terms,
the OA state of play can be summarised around two main explanatory frameworks: traditional universalist explanations and more recent alternatives. As the adoption of agriculture occurred almost simultaneously in many different regions of the world, it has traditionally been explained in universalist terms, i.e., a single universal cause is proposed as the lever of the OA. The most renowned
universalist prime-mover explanations may be summarised around three main paradigms: The superiority of agriculture as a mode of production, irrespective of circumstances. Agriculture—a highly desirable development whose advantages were self-evident—would have been automatically adopted in favourable ecological conditions once the necessary knowledge—the limiting factor—had been reached
[10]. Within this framework, growing agriculturalist populations would have colonised new territories, absorbing or displacing local hunter-gatherer groups, and making them adopt agriculture rapidly
[11–15]. This, in turn, would have laid the ground for a series of socioeconomic changes that, later in time, resulted in urbanization processes and the emergence of civilisations
[16–18]. Accordingly, this first approach encompasses four main correlates that have long influenced the most widespread conception of the OA: (1) a conceptual dichotomy between “two mutually
incompatible ways of life” [19]: hunting and gathering on the one side, agriculture on the other, with no intervening options [20]. Intermediate positions between the two are regarded as transitory, short-lived intermediate stages from one
steady state to the other; over the long term, the great majority of populations are assumed to tend either to maintain a hunter-gatherer subsistence economy or to embrace agriculture, presenting a U-shaped distribution to the relative proportions of gathered/hunted versus produced food in the diet
[21,22]. (2) The notion that the agricultural transition was a radical and rapid switch in human evolution, arising from an express domestication once the necessary knowledge was acquired
[20,23,24], what enabled its consideration as a “revolution”: the Neolithic Revolution
[25–27]. (3) The conception of the adoption of agriculture as a “point of no return” [17]; for those
hunter-gatherer societies becoming agriculturalists, there would be no turning back unless they are compelled by environmental downturn or increased mortality [22]. And (4) a view of human history totally biased towards a classical idea of progress in which agriculture is clearly prioritised against hunting and gathering
[18,25,28], hence being hunter-gatherers stereotyped as inherently simple and
agriculturalists as culturally complex [29,30]. Standard-Evolutionary-Theory-based (SET-based) explanations for initial domestication. This second paradigm is based on the core assumption
of unidirectional adaptation (environments change and organisms adapt, never vice versa [31]); more precisely, a population-resources imbalance is proposed as a direct or underlying cause of the transition to farming. Such an imbalance could have originated either on the supply side, the demand side or both
[32,33]. Thereupon, domestication, intensification and/or the exploitation of suboptimal resources would have emerged as an adaptive response to the resource pressure induced by demographic changes in the form of a local
[34,35] or a global
[36,37] population increase, and/or by environmental changes such as a decline in resource availability
[38–41] and/or climatic variations like the Pleistocene-Holocene climatic transition—which resulted in warmer, wetter conditions, a significant reduction in climate fluctuations, and a 33% increase in atmospheric CO2
[42–49]. It is important to recall that the key assumption of this second paradigm is that hunter-gatherers would only have become agriculturalists under pressure, as farming is usually more labour-intensive, backbreaking
and time-consuming than hunting and gathering [32], often leading to no immediate change in quality of life
[32,50–52]. Social hypothesis. Under this paradigm, the onset of agriculture
would have been motivated by social forces in stress-free scenarios. More precisely, a set of ‘social disequilibrium models’
[30,53–56] proposes that farming would have been embraced to maintain social control, or in the struggle for
power, spouses and/or status [19]; this, in turn, would have resulted in the emergence of inequality and hierarchical societies. More recent alternatives derived from EE, NCT and integrative proposalsAlthough evolutionary frameworks have been used to assess human behaviour for decades (see [57] as an example), their popularity has increased substantially in recent decades. More specifically, the latest explanatory approaches to the OA are either derived from Evolutionary Ecology (EE)—inclusive of Human Behavioural Ecology (HBE)—[4,58–60], from Niche Construction Theory (NCT) [2,3,9,61,62], or from integrative approaches comprising EE, NCT and models of cultural transmission and gene-culture coevolution that envisage all the perspectives involved as complementary, synergetic and broadening each other [58,63–69]. EE is a selectionist, neo-Darwinian high-level theoretical framework that provides a well-defined set of general concepts, assumptions and analytical tools—such as optimisation theory—for the study of specific adaptations through the lens of the interaction between evolutionary forces and ecological variables. One of its subfields, HBE, which investigates human behaviour in relation to ecological conditions and assesses the different behavioural strategies in terms of fitness [4,70], has already made substantial contributions to OA research [71]. In this regard, particularly famous is the Diet Breadth Model (DBM), according to which human groups are supposed to have a list of all the resources in their environment ranked in descending order of net caloric return, being delayed-return strategies—such as resource management and production—only embraced when immediate-return alternatives are not productive enough [4,69,72–75]. According to the DBM, at the dawn of agriculture, a resource depression would have forced the inclusion in the diet of previously ignored resources such as the wild ancestors of present domesticates (small to medium-sized mammals, seeds and tubers), resources otherwise falling beneath the “optimal diet” boundary of human societies world-wide [63]. On its part, NCT postulates that organisms are capable of modifying their own evolutionary trajectories—and those of other species—by actively engineering their living environments, which can provide them with evolutionary advantages [76]. Thence, according to NCT, domestication arose from large-scale human efforts at ecosystem enhancement in the absence of any sort of population-resources disequilibrium [2]. More specifically, NCT representatives [9,61,62,77,78] consider that agriculture would have emerged in climatically stable resource-rich scenarios (usually near water), where small-scale societies would have established small semi-permanent to permanent central settlements, and within which a wide range of plant and animal species would have been comprehensively auditioned over many generations, evolving just a subset of them into domesticates [2]. Over the last few years, the increasing wealth of data on the OA has evidenced that the transition to farming was characterized by a great variability across time and space, i.e., that the different world regions followed independent developmental pathways, being each one of them shaped by a number of complex and locally contingent factors, and generally extending over long periods of time [63,79]. As a result, considerable controversy has emerged regarding the utility of the different explanatory approaches for the OA, existing to this day no consensus on which approach is most appropriate; in fact, the OA community continues to be characterised by a lack of consensus in many respects [6], with just a few major areas of agreement that may be summarised as follows:
In addition to the foregoing, a significant number of OA researchers agree that single causal scenarios based on prime movers fall short [3,8,63,79,91]. Clearly, the amelioration of climates at the end of the Pleistocene acted as a trigger; however, other relevant factors such as human demography, social systems and the biological characteristics of the auditioned species were operating simultaneously in tightly interconnected networks, thus being not possible nor convenient to select just one of them as the only cause. As a consequence, current approaches revolve around broader explanatory frameworks that try to integrate both the interplay of the different factors [3,6,8] and distinct theoretical approaches such as HBE and NCT [58,65,67–69]. For a succinct summary of the main theories on the OA described above and the most relevant references, please refer to the S1 Table the Supplementary Material. The vast middle-groundIn the previous section, we concluded by underscoring the inherent complexity of the transition to farming, which stems mainly from the protracted nature of the process, its marked variability across time and space, the multitude of factors involved at both the macro and micro scales, the intricate nature of the diverse middle ground, and the subsequent long-standing debates on the suitability of the different explanatory approaches. As pointed by Smith [86], from all the factors mentioned, the middle ground is probably the most largely neglected one, a fact that continues to hamper significantly our current understanding of the agricultural transition. Hence, in the present paper we intend to shed light on what the middle ground may have looked like. Given that the interest of the present contribution may be best understood in the light of what is known about the middle ground thus far, here we proceed to cover some of the key aspects. The middle ground refers to the conceptual territory between hunting, gathering and fishing on the one side, and agriculture and husbandry on the other. In accordance with the classic dualistic epistemology, foragers had no domesticates, and all human societies with domesticates had agriculture, being the transition between the two regarded as a mere shift [86]. Consistently, the middle ground has traditionally been regarded as some sort of hodgepodge of all the subsistence economies not meeting the criteria to be considered either hunter-gatherers or agriculturalists. Nevertheless, as discussed earlier, there is a global consensus that such dichotomic approaches are both insufficient and inefficient for several reasons, outstanding among which are (i) the difficulty of defining “hunter-gatherers” and “agriculturalists”: over the years, numerous definitions have been proffered (some of them inspired in real referents coming from Ethnography and/or Archaeology, whose attributes have been established as diagnostic); however, hitherto no single definition is considered to have a global or universal applicability [92,93]. (ii) The prolonged nature of the transition to agriculture, which indeed extended over millennia, hence having had to encompass many successful long-term subsistence alternatives that are fully disregarded in the binary approach [86]. And (iii) its implicit strait-ahead linear sense, which depicts the transition as an evolutionary process irreversibly leading to agriculture [94]. Therefrom, while the conceptual dichotomy can still be useful as chronological markers that delimit the extremes of the middle ground, it is also quite harmful, since it blurs and oversimplifies the multiple changes in subsistence behaviour that must have occurred during the Holocene in many different regions of the world. Moreover, even if different adjectives such as “complex” hunter-gatherers, “affluent” foragers, “incipient” agriculturalists, etc., have been used in an attempt to differentiate middle-ground strategies, those expressions still hold the imprint of the binary conceptualisation, and implicitly displace in-between societies to one margin or the other, hence concealing again the relevance of the middle ground in its own right. More recently, different authors have addressed the middle ground in various ways that transcend the dichotomous perspective. In overall terms, their proposals can be grouped around two main approaches: compartmental schemes and non-classificatory perspectives. As stated by Smith [86], compartmental schemes are aimed at “identifying and defining categories of human-plant and human-animal interaction” characteristic of a given subset of middle-ground societies. When taken together, all these interaction categories “form a continuum of increasing human intervention or involvement in the life cycle of targeted species […] that encompasses the landscape that lies between hunting-gathering and agriculture”. Over the years, many scholars have adopted the compartmental approach: see for instance, Ford [95], Harris [96–99], Higgs [100] and Zvelebil [24,101] (among others); however, the most renowned of all compartmentalists is decidedly Bruce D. Smith thanks to his proposal in [86] that if the binary categories food procurement and food production are maintained, the latter should be split into three main sub-categories: low-level food production without domesticates, low-level food production with domesticates and agriculture. Smith encompasses under food production all human actions aimed at intervening in the life cycle of targeted species, regardless of the presence or absence of domestication traits. Within the food production territory, subsistence strategies reliant on domesticates for less than a 30%–50% of the annual caloric intake are considered low-level food production, while those above such boundary are deemed fully agricultural economies. Regarding the partition between low-level food production without domesticates and low-level food production with domesticates, as their name suggests, the boundary marker is domestication, i.e., the presence of morphological/genetic changes associated with the domestication syndrome. Nonetheless, as recurrently highlighted by Smith [86], the definition of boundaries is an extremely troublesome task, since manifold management strategies have been documented across all categories at different intensity levels. Therefore, the relevance of Smith’s proposal is not to be found in the robustness of the definition of boundaries, but on the creation of the low-level food production category and its two subcategories; a conceptualisation that serves to highlight the importance of such a diverse array of stable subsistence strategies, whose persistence attests to their versatility and success in a wide range of different socioenvironmental contexts. Several ensuing publications incorporated Smith’s conceptualisation of the middle-ground either making use of the term low-level food production [3,87,102,103] or developing similar ones such as: low-level food-resource producers [104,105] or mixed economies [4,106–110]. Notwithstanding, other authors such as Terrell et al. [93] have argued that instead of a classification system whose inter-category boundaries are difficult to establish, it would be more appropriate to describe and compare subsistence strategies without first having to label them. Hence, they proposed the “provisions spreadsheet”, an interactive matrix of species and harvesting tactics intended to overcome restrictive definitions of domestication, and to accommodate the known diversity of human subsistence practices. Notably, the provisions spreadsheet covers four different aspects: goal (food or raw materials), primary variables (resource breadth and their yield, accessibility and reliability), secondary variables (behavioural or environmental manipulation skills necessary to exploit those resources) and observations, to faithfully reflect the particularities of each socio-economic strategy and of the environment within which they develop(ed). To this day, no-one of the two approaches has proven to be better than the other. However, this is not necessarily a problem, as both proposals serve to place on the middle ground the emphasis it deserves, and highlight the difficulty of establishing clear-cut frontiers between the different subsistence possibilities. In fact, in the last decades, plenty of empirical evidence—both ethnographic and archaeological—is showing that a great deal more research is needed to understand the different subsistence combinations, and to unravel the particulars of the middle ground. Regarding the most outstanding recent findings on the middle ground, particular mention must be made of the innate complexity of hunter-gatherer resource management techniques, the sophistication of complex hunter-gatherer societies and the relevance of mixed economic choices. As regards hunter-gatherer complexity, several recent contributions have served to emphasise both the diversity inherent in hunter-gatherer subsistence practices, and the great variety of their socio-political systems [111–113]. One of the major advances has certainly been the recognition that no human societies are simple by definition, as even the smallest hunter-gatherer societies present intricate social relationships and elaborated economic solutions that comprise scheduling, mobility, task differentiation, gender division of labour, etc. [112]. In addition, the engagement in efforts at landscape engineering such as regular burning, tending, tilling, transplanting, weeding, sowing and selective harvesting has been documented for both past and present hunter-gatherers and low-level food producers, which supports the existence of a wide spectrum of hunter-gatherer and low-level-food-production subsistence strategies without clear frontiers between them [109,114,115]. Particularly outstanding among the different landscape modification techniques is the so-called slash-and-burn cultivation, a common practice in tropical forested areas that consists in cutting the trees and undergrowth of the selected site, leaving them to dry for a few months, and eventually burning them to restore minerals and nutrients to the soil prior to sowing—which is done at the beginning of the rainy season [116]. In the Amazon basin, for instance, slash-and-burn is widely reported in the ethnographic record in connection with manioc and/or maize cultivation; in addition, numerous works point to its use in pre-Columbian times as either an adaptation of late Holocene immigrating agriculturalists or as an extension of Amazonian house gardens. Be that as it may, the fact is that Amazonia hosts Holocene age archaeological evidence for anthropogenic landscape modification via burning, the vast majority of which is located in the vicinity of natural concentrations of aquatic resources, which underscores the well-known Amazonian complementarity between fish and cultivated plants [117]. From all the above, hunter-gatherers are no longer regarded as failed agriculturalists living in marginal environments; quite the opposite, at present, hunting-gathering and low-level food production are conceptualised as a broad range of successful long-term socioeconomic solutions characterised by their flexibility, resilience and adaptability to very different contexts [113]. As regards mixed economies, plenty of past and present empirical examples attest to their great variety and endpoint solution status. In the context of Archaeology and Prehistory, mixed economies tend to refer to economies developed during the Mesolithic and the Early and Middle Neolithic. Although some publications refer to mixed economies whenever fishing complements hunting and gathering [118], most researchers designate under mixed economies a combination of foraging (hunting, gathering and/or fishing) with different food-production activities such as farming at different intensity levels and/or small-scale herding [119–122]; more precisely, the most widespread definition of a mixed economy is probably the one proffered by Winterhalder and Kennet [60] as “an economy (…) that includes, in long-enduring combinations, both foraging and the low-level use of cultivars or domesticates“. Some of the most paradigmatic archaeological examples of mixed economies include the Jomon (Japan) and Chulmun (Korea), whose resource procurement strategies included hunting, gathering, fishing and resource production strategies ranging from annual plant encouragement and tree management, to domestication of particular species and cultivation of others [105,123–125]; the Okhotsk (preceded in time by the Jomon), who hunted, gathered, raised pigs and had a few crops [105]; the peoples from the Northwest Coast of North America (typically classified as complex hunter-gatherers), who were strongly reliant on mariculture and conducted plenty of resource management activities to sustain and enhance both marine and plant resources: beach clearance, clam and root gardening, transplanting salmon eggs, tree modification, landscape burning, soil tilling, etc. [126–129]; and the late Prehispanic peoples from Sierras of Córdoba (Argentina), whose subsistence economy was a mix of small-scale farming and broad-scale foraging [120]. In the ethnographic record, some of the most well-known examples of mixed economies include the Ituri forest foragers of the Democratic Republic of Congo (who engage in partnership relationships with agriculturalists), the Agta from the Philippines (famous for their exchange of meat for rice), the savanna Pumé (a Venezuelan mobile hunter-gatherer society that incorporates manioc cultivation as a fallback strategy) [106], the Mlabri (skilled hunter-gatherers now living in north-eastern Thailand and the western Laos, who establish symbiotic trading relationships with more settled groups) [130], and present-day indigenous communities in the Arctic (whose economy combines foraging with trade or other economic activities such as full-time or part-time paid work, seasonal labour, craft-making, commercial fishing and/or tourism) [107,130–132]. It is clear from the foregoing that the study of the middle ground is key to understand prehistoric societies in general, and the transition to agriculture in particular. Remarkably, both in the ethnographic and archaeological records, we count with plenty of examples that illustrate the whole economic variability that human societies have implemented around the globe for millennia. In this regard, it is important to recall that even if Ethnography and Archaeology tend to deal with different time periods and rhythms [133], and being fully aware that available ethnographic information does not necessarily and directly inform about prehistoric societies, the fact is that ethnographic data allows to test archaeological theory by confronting it with real case studies. Concretely, as regards subsistence strategies, ethnographic examples enable to identify which economic choices are plausible, and in which contexts they are effectively taking place, thus contributing to a general theory of human and social behaviour that can be used as a frame of reference for prehistoric studies [134,135].
Research proposalFrom a methodological perspective, our contribution is framed within the increasingly relevant research line of integration, mining, analysis and interpretation of archaeological, ethnographic and anthropological data to explore global patterns through advanced computer-based approaches [136–138]. In particular, in the present work we intend to explore the middle ground by means of a comprehensive analysis of the subsistence strategies of 1290 societies extracted from the Ethnographic Atlas [139–141], which are either historical or ethnographically documented. The purpose of such an approach is twofold: (i) increasing our understanding of the different combinations of subsistence strategies that have been developed across the world over time; and (ii) providing new perspectives from which to explore prehistoric economies, the persistence of hunter-gatherer(-fisher) economies, and the agricultural transition. Note that we are not aiming at establishing direct formal analogies between our data and Prehistory, but at better understanding human economic choices, the contexts in which they develop, and their success in the long term. Therefore, our study can be considered to belong to the theory-building realm, and it may help to look at prehistoric economies in a new light, and/or to hypothesise in a different way about particular archaeological contexts. More precisely, we are interested in the following research questions:
Materials and methodsData sourcesAll data used in the present study were extracted from the Database of Places, Language, Culture, and Environment—D-PLACE—[138] accessible at https://d-place.org/. In words of its creators, D-PLACE is “an attempt to bring together the dispersed corpus of information describing human cultural diversity” [142], being its main goal to facilitate cross-cultural analyses. More specifically, D-PLACE integrates information from multiple datasets and presents it in a unified and consistent manner. The sample of societies selected is that of the Ethnographic Atlas (EA) [139–141], which in its original form includes up to 1291 pre-industrial societies [143] ranging from small hunter-gatherer groups to societies with complex agricultural economies—i.e., market economies are not included. However, in the present contribution we reduced the EA sample size to 1290, since that is the number of societies for which we had information on subsistence strategies. It is important to note that while these societies are globally distributed, the EA has especially good coverage of Africa and western North America [144], and that as a consequence of the pre-industrial nature of the societies in the EA, the conclusions obtained from the analyses in this paper cannot be extrapolated to market economies. As regards the focal time period, the different time frames and the percentages of EA societies falling under each of them can be found in Table 1. Table 1Focal time periods of the societies documented in the EA.
From Table 1 it becomes clear that the time span covered is quite circumscribed. This aspect is particularly convenient, since it avoids problems associated with longer time frames, namely changing ecological and social conditions, different occupations of the same site over time, long-term transitions in subsistence economies, social organisation, etc. The different subsistence-related and socio-ecological variables used in this work and the datasets from which they were extracted are detailed in Table 2. Note that the column Data type indicates if the variable is originally continuous, categorical or ordinal in D-PLACE. Subsequent transformations of the variables are duly explained in the corresponding Rmarkdowns available at the GitHub repository of the paper: https://github.com/Virahe/Lets-go-Fishing. Table 2Variables selected for the present study from the different sources available at D-PLACE.
Methods: Data analysisThe analysis of the database described in the previous section was performed in two stages:
The details of the analytical methods applied in each of the two stages are summarised below. Exploratory data analysis by means of unsupervised learning techniquesUnsupervised learning is a subfield of machine learning conceived to look for patterns in a dataset with no predefined response variable, i.e., it is referred to as unsupervised since no output variable guides the analyses [152,153]. Under the umbrella of unsupervised learning we find a great variety of methodologies, many of which seek to find relationships either between variables or between observations, hence being generally applied within the framework of exploratory data analysis. Regarding the present contribution, we have drawn on two of the most commonly used unsupervised learning techniques, namely Principal Components Analysis (PCA) and clustering, to address our first research question: Are all subsistence combinations viable or do specific patterns exist? Principal Components Analysis (PCA)The principal components of a dataset are the dimensions in feature space along which the original data present the greatest variation; these dimensions also define the lines and subspaces that are closest to the data cloud (minimum squared distance). More precisely, for a dataset X with n observations and p features (all standardised to have mean zero and standard deviation one), each component is a linear combination of the p features, and all components together constitute an orthonormal basis, which ensures that they are linearly uncorrelated [152]. PCA is the unsupervised learning technique consisting in the computation of the principal components and their subsequent use in understanding the data. It is frequently used for dimensionality reduction and visualisation purposes, since by keeping just the first few principal components and projecting all data points onto them, a low-dimensional representation that captures as much of the information as possible is obtained. Note that the interpretation of the different components is generally conducted by projecting the features on the new low-dimensional space, and by thereafter assessing the correlations between the different features and the corresponding components. In the context of the present study, we took all the societies in the EA, focused exclusively on the variables ranging from EA001 to EA005, (i.e., the percentages of dependence on gathering, hunting, fishing, husbandry and agriculture, what we have called the subsistence dataset), and performed PCA to ascertain whether the variability in our data could be explained by just a few components, and if so, to identify which of the variables had a more significant role in explaining such variability. ClusteringThe term clustering encompasses a broad set of algorithms aimed at identifying groups (clusters) in a dataset, so that the observations within each group share some common features, while observations in different groups are quite different from each other [152,154]. In relation to our first research question, should only a subset of subsistence economies be successful and persistent, it would translate into the data having cluster structure. Therefore, at this stage of the analyses, we took again the subsistence dataset and conducted cluster analysis to ascertain whether it had structure, and if so, to explore the subgroups identified among the observations. More specifically, we first assessed the clustering tendency of the subsistence dataset following Kassambara [155]; then, we opted for the hierarchical clustering methodology [152], since instead of committing to a specific number of clusters, it results in a dendrogram that allows conducting analysis at different levels of resolution; and finally, we tried to determine the optimal number of clusters—the height at which to cut the dendrogram—by means of the multiple techniques proposed in [155] and consensus clustering [156]. Assessment of clustering tendencyA noteworthy issue in clustering analysis is that clustering techniques will return clusters even if the data do not contain any meaningful clusters. Therefore, before applying any clustering algorithm, it is necessary to check whether the data contain non-random structures. To that end, Kassambara [155] proposes two alternatives:
Hierarchical clusteringAs its name would suggest, hierarchical clustering is a clustering technique intended to build a hierarchy of clusters of the observations in a dataset [152]. Its main advantages are (i) that it results in a dendrogram (an intuitive tree-based representation of the observations); and (ii) that it is not necessary to specify the number of clusters in advance, as the dendrogram covers all possibilities from the lowest level—at which each cluster contains a single observation—to the highest one—at which all the observations belong to the same cluster. Consequently, hierarchical clustering is extremely useful to conduct analysis at different levels of resolution, which is precisely the reason why we chose it to explore the different subsistence economies. Nevertheless, it is noteworthy that not committing to a number of clusters beforehand can also be regarded as a drawback, namely the lack of an objective criterion to determine the number of clusters, which can result in the biases of the analyst influencing the choice. Determination of the optimal number of clustersOnce the data have proven to be clusterable and the clustering algorithm has been selected, the next step is trying to ascertain the optimal number of clusters. However, there is no one answer to this question, as the optimal number of clusters depends on both the partitioning algorithm chosen and the similarity/dissimilarity metric used. In fact, there exists a great variety of indices and methods conceived to determine the optimal number of clusters, see, for instance, the elbow method [155], which selects the number of clusters that minimises the total intra-cluster variation; the average silhouette method [158], which maximizes the quality of clustering, i.e., how well each observation lies within its cluster; the gap statistic [159], which chooses the clustering furthest away from the random uniform distribution of points; all the indices included in the NbClust R package [160]; and consensus clustering approaches [156]. For further details on the NbClust R package, please refer to its reference manual available at [161]. As regards consensus clustering, it is a resampling-based method designed to determine the consensus number of clusters in the data, and to assess and represent cluster stability. In particular, consensus clustering takes the original dataset, creates a given number of perturbed datasets via the resampling technique of choice, applies the clustering algorithm selected to those perturbed datasets, and finally assesses the agreement among the multiple runs. Its underlying assumption is that the more robust clusters are to sampling variability, the more likely it is that they reflect the real structure of the data. In the present work, we opted for computing all the above-mentioned metrics and selecting the best number of clusters either in accordance with the majority rule, with consensus clustering and/or with both. Cluster entropyAfter choosing the number of clusters, in the present contribution we have drawn on the concept of information entropy to formally define mixed economies. Within the framework of information theory, the entropy of a random variable is defined as the uncertainty intrinsic in the variable’s possible outcomes [162]. Formally, the entropy of a discrete random variable X is calculated as follows: Where X = {x1,⋯,xn}, i.e., x1,⋯,xn are the possible outcomes of X, and they occur with probability P (x1),⋯,P (xn). Note that the entropy is maximal when all possible outcomes are equiprobable (uniform probability distribution). In the context of the present contribution, a subsistence economy is defined by five different variables—namely the percentages of dependence on gathering, hunting, fishing, husbandry and agriculture—that are interpreted as the probability of relying on those food sources. Therefore, given that the maximal entropy is attained when all possible outcomes follow a uniform distribution, the subsistence economies with the same percentage of dependence on the five alternatives will be the ones with maximal entropy. Consequently, mixed economies can be formally defined as those subsistence combinations with high/the highest information entropy. Supervised learning approachSupervised learning techniques are intended for the type of problems in which each observation—codified as a vector with the corresponding values for the different predictor variables—has an associated response measurement. In particular, supervised learning methods seek to fit a model that relates the response to the predictors, either for prediction purposes or to better understand the relationships between the response and the predictors [152]. Supervised learning problems can be, in turn, divided into two main categories based on their response variable: regression problems (quantitative response) and classification problems (qualitative response); specific machine learning techniques have been developed for each problem typology, existing also methods that can be used for both quantitative and qualitative responses. As regards the present contribution, once structure has been found in the data, and the prevailing subsistence economies have been identified via clustering (i.e., the number of clusters has been selected and hence each observation has been assigned to a cluster), we can use such information as the response variable to conduct supervised learning analyses. More precisely, we have opted for supervised learning approaches to address our second research question, i.e., to assess the role played by ecological settings in the configuration of the different subsistence economies. First, it is necessary to clarify that this second stage of the analyses has been conducted on what we called the supervised learning database: a dataset including all the variables in Table 2 except for the percentages of dependence on the different subsistence strategies (variables from EA001 to EA005), plus the response variable (the cluster to which each observation belongs); note that we have chosen the response variable for k = 15, which is the highest level of resolution. Therefore, given that our response variable is qualitative, we are facing a classification problem. Notably, we did not include variables from EA001 to EA005 in the supervised learning database because the clustering analysis has been conducted exclusively on the five of them, and thus, they are closely related to the response variable, which could hinder the identification of other possible relationships between the rest of socioecological variables and the clusters found. When joining the variables in Table 2 to configure the supervised learning database, we found that we had missing data, i.e., that certain societies in the EA had no information available for some of the variables coming from other data sources. To solve this problem, instead of dropping those societies with missing values, we kept all of them and applied a multiple imputation technique (as opposed to single imputation) to estimate the statistical uncertainty attributable to the missing data. This approach consists in creating multiple complete datasets by independently imputing the missing data via stochastic draws from the distributions of the observed data. In particular, in our work we have used the Multiple Imputation by Chained Equations (MICE) method [163], which is implemented in the mice R package [164]. MICE assumes that data are missing at random (MAR)—i.e., that the probability that a value is missing depends only on the observed data—and consists in iteratively estimating the conditional distributions of each variable from the rest of the variables. To increase stability, the imputation process of all variables is repeated through cycles. In our analysis, we have generated 100 different datasets, being each one of them the result of 50 cycles. As for the MICE imputation method chosen, we have used a recursive partitioning approach: random forests [165,166], so as to take into account possible nonlinearities, interactions and both numerical and categorical data. Recall that even though variables from EA001 to EA005 have not been considered for the supervised learning analyses, we did include them in the database on which multiple imputation was conducted, so as to help get more accurate estimates of the missing values. Regarding the supervised learning model to be fitted to our data, from all techniques suited for classification problems, we have chosen random forest [167]; the rationale behind this choice is to be found in four main reasons: (i) as previously stated, random forest can handle both numerical and categorical predictors; (ii) it is an ensemble method, which translates into the resulting model having a good bias-variance trade-off—i.e., the resulting model is neither too general nor overfitting the data [152,153]; (iii) in an exhaustive evaluation of up to 179 classifiers, random forest was found to be the best family of classifiers [168]; and (iv) it allows to conduct both individual and group variable importance analyses. Nevertheless, to be sure that our choice was appropriate, we have compared random forest with several benchmark and highly performant classification algorithms. More specifically, we have conducted an ANOVA test and the Duncan’s New Multiple Range Test [169] on the accuracies obtained by stratified 10-fold cross-validation with the following classifiers: ZeroR [170], OneR [171], AdaBoost [172], Support Vector Machine with polynomial kernel [152,173], rotation forest [174] and random forest [167]. Note that in the Duncan’s New Multiple Range Test, if two classifiers belong to the same group, no statistically significant differences exist between them, which implies that the choice of the classifier can be based on criteria other than accuracy—see for instance interpretability, model simplicity, etc. As far as variable importance analyses are concerned, their raison d’être is that predictors are seldom equally important in supervised learning models; in fact, usually only a subset of them is relevant to determine the response. Therefore, it is of interest to quantify the relative contribution of each variable (individual variable importance analysis) or group of variables (group variable importance analysis) in predicting the response. In particular, variable importance is calculated as the mean decrease in accuracy after randomly permuting a given predictor [167] or a group of predictors [175]. Recall that the more the accuracy is reduced after the permutation, the more important the predictor/group of predictors. Remarkably, variable importance analyses become even more important in the context of ensemble models, since ensembles result into better accuracy than single-tree models, but they do so at the expense of interpretability. ResultsExploratory data analysis by means of unsupervised learning techniquesPCAThe results of the PCA conducted on the subsistence dataset (variables from EA001 to EA005) can be found in Table 3, where we can see that the total variance of the data is effectively explained by the first four PCA dimensions. Table 3Results from PCA on the subsistence dataset.
In Fig 1 we can see how the different variables considered (percentage of dependence on gathering, hunting, fishing, husbandry and agriculture) correlate with the first four PCA dimensions. The actual correlation values and their respective p-values can be found in S2 Table from the Supplementary Material. PCA graph of variables for dimensions 1:2 and 3:4 respectively. A contextualised interpretation of these four PCA dimensions may be summarised as follows:
Assessment of clustering tendency
Hierarchical clusteringHaving shown in the previous section that the data are clusterable, here we proceed to explain the hierarchical clustering implementation that we applied to our subsistence dataset. We have worked with the R package factoextra [176], and more precisely, we have used the the eclust function selecting the Euclidean distance as dissimilarity metric, and Ward’s method as the agglomerative method. Determination of the optimal number of clustersAs emphasised in the Methods section, one of the main strengths of hierarchical clustering—and the reason why we chose it—is that it enables to conduct analyses at different scales of resolution. Accordingly, in the present contribution we have decided to explore the resulting dendrogram at three levels: k = 2, 7 and 15. The reasons justifying these choices may be summarised as follows:
Detail of the clusters obtained for k = 2, 7 and 15In the present section we succinctly interpret the clusters obtained at the three levels selected. 3. Sunburst of the different subsistence strategies identified via hierarchical clustering for k = 2, 7, 15. A short description is provided for each of the clusters. Note that the size of the circular sectors is proportional to the number of societies falling under each cluster. is a sunburst diagram that summarises the cluster precedence relationships. Note that the lowest level of detail (k = 2) is to be found in the inner circle, and that the greatest detail (k = 15) is presented in the outermost circle. Recall as well that the percentage values accompanying the cluster descriptions are the mean values obtained for the given variable across all societies in that cluster. At the lowest level of resolution, we find the traditional dichotomy between hunting-gathering-fishing and agropastoralism. When moving to the intermediate level, the differences between the distinct subsistence alternatives become more evident: the hunter-gatherer-fisher (HGF) group is divided into three depending on which of the three activities is more heavily relied upon: (1) gathering, (2) hunting or (3) fishing; as for the agropastoralist group, it is splitted into four clusters: (4) the strictly pastoralists, whose strategy is based predominantly on animal husbandry and complement it with a low percentage of agriculture; (5) the agropastoralists, who are strongly dependent on agriculture and complement their subsistence economy with animal husbandry; (6) the group which we have agreed to name agrofishers since their subsistence economy is a mix of fishing and agriculture, being the rest of possible activities—hunting, gathering and husbandry—underrepresented; and (7) those with what we have called the Whole Spectra Economy (WSE), i.e., whose subsistence economy encompasses hunting, gathering and fishing at approximately par value, and a great percentage of agriculture. Eventually, at the highest level of resolution we identify the more subtle nuances according to which the previous seven clusters are further subdivided. As for the three HGF clusters, they are partitioned into six clusters for k = 15: the HGF’s more dependent on gathering are divided into two clusters: (1.A.) those who are just hunter-gatherers (HG’s) more reliant on gathering (we claim that they are simply HG’s since fishing represents a negligible value in this cluster); and (1.B.) the archetypal HGF’s with a similar percentage of dependence on each of the three sources; the group of HGF’s more heavily reliant on hunting is maintained and now named (2.BIS.); as regards the cluster of HGF’s more heavily reliant on fishing, they are splitted into three clusters: (3.A.) the eminently fishers, with a percentage of dependence on fishing greater than 70%; (3.B.) hunter-fisher groups with similar percentages of dependence on each of the two strategies and a negligible level of gathering; and (3.C.) HGF’s–Fishers, who in spite of being more heavily dependent on fishing resources, present significant percentages (greater than 20%) of gathering and hunting. Concerning the other four clusters (the ones with a given dependence on agriculture), for k = 15 they have been further splitted as follows: the pastoralist cluster has been divided into two different groups: (4.A.) the outstandingly pastoralists, which in this smaller cluster present average values of up to a 75% dependence on animal husbandry and just a 13% complementation with agriculture; and (4.B.) agropastoralists that are more strongly dependent on animal husbandry than in any other source (up to 49% on average), and that complement their strategy not only with agriculture (22%) but also with hunting (12%) and fishing (11%), what allows to consider them as having a mixed economy. Regarding the agropastoralist cluster, it has been divided into three clusters: (5.A.) the strictly agropastoralists—again—whose percentages of dependence on both agriculture and husbandry are almost maintained with respect to its homonymous cluster at k = 7; (5.B.) a second group of agropastoralists more heavily reliant on animal husbandry; and (5.C.) what we have called agrohunting, which is still heavily reliant on agriculture, but instead of complementing the subsistence economy with just husbandry, it also presents a significant degree of hunting (in fact, the percentage of dependence on hunting is greater than that of husbandry). As for the agrofishing cluster, it has been partitioned into two clusters in accordance with the percentages of dependence on agriculture, fishing and husbandry: (6.A.) agrofishers with an average 47% reliance on agriculture, almost 40% dependence on fishing and a marginal contribution of husbandry to the diet of up to 7%; and (6.B.) agrofishers more strongly dependent on agriculture (54%), with a 20% percentage of dependence on fishing, and a greater weight of husbandry (13%). Lastly, the cluster of the WSE has been splitted into two: (7.A.) the truly agrohunters, which exhibit an average 45% reliance on agriculture, 28% on hunting and complement their strategy with an 18% of fishing; and (7.B.) the strictly WSE, which is not significantly altered with respect to its homonymous cluster at k = 7, presenting now an average reliance on agriculture of up to 35% and of 21% on hunting, 25% on gathering and 15% on fishing. At this point, it would be interesting to recall that even though for k = 2 there is no trace of mixed economies, and at k = 7 they are restricted to two clusters (agrofishers and the WSE), when we increase the level of detail and work at a fine-grained scale (k = 15), they make up a significant percentage of the possible subsistence economies, in fact, 5 out of 15 clusters can be considered as mixed: clusters (4.B.), (6.A.), (6.B.), (7.A.) and (7.B.). For further details on this aspect, please refer to the discussion, where it is examined in depth. Additionally, we decided that it could be of interest to represent the different clusters in a map with the terrestrial ecoregions of the world proposed by Dinerstein et al. [150]. Therefore, we created Figs 3 and 4. In Fig 3 we can see all societies considered in the present study placed in accordance with their latitude and longitude and coloured according to the cluster they belong to for k = 7; note that the map has intentionally been coloured to represent the different biomes, and that a legend with the average percentages of the different variables for each cluster has been provided. As for Fig 4, it is the replication of Fig 3 for k = 15, with the caveat that instead of representing the 15 clusters, as the interest of the present paper is on the agricultural transition and the middle ground, we have just represented the 9 clusters with a given level of dependence on agriculture and/or husbandry. Map with the subsistence clusters obtained for k = 7. The different societies have been placed in accordance with their latitude and longitude and coloured according to the cluster they belong to. The different world regions have been coloured to represent the different biomes. Two legends provided: the upper right one presents the detail of the average percentages of dependence on gathering, hunting, fishing, husbandry and agriculture of each of the clusters. The lower right one is the legend of the biomes. Map source–[150]. Biome cartography is licensed under CC-BY 4.0 (https://ecoregions2017.appspot.com/). Map with 9 of the subsistence clusters obtained for k = 15. The different societies are placed in accordance with their latitude and longitude and coloured according to the cluster they belong to. Only those subsistence strategies with a significant level of agriculture and/or husbandry have been considered. Two legends provided: The upper right one presents the detail of the average percentages of dependence on gathering, hunting, fishing, husbandry and agriculture of each of the clusters. The lower right one is the legend of the biomes. Map source–[150]. Biome cartography is licensed under CC-BY 4.0 (https://ecoregions2017.appspot.com/). Cluster entropyAs stated in the Methods section, in this work we have formally defined mixed economies as those subsistence strategies with high/the highest information entropy—remember that maximal entropy is achieved when all possible outcomes follow a uniform distribution, which in our case would translate into subsistence economies with the same percentage of reliance on the five variables considered (gathering, hunting, fishing, husbandry and agriculture). Note that in our results, the WSE is the subsistence economy closest to the case with maximum theoretical entropy. To illustrate the concept, we have calculated both the entropy of each society’s subsistence strategy and the entropy of each cluster’s average strategy for k = 2, 7 and 15. (Recall that a cluster’s average strategy is calculated by averaging the percentages of dependence on gathering, hunting, fishing, husbandry and agriculture across all societies in the cluster). For each level of analysis, we have obtained: (i) a summary table with the clusters’ average strategies, their entropy, their standard deviation, two columns quantifying the number of variables with a percentage of dependence equal or greater than 15% and 10% respectively, and a concise interpretation of each cluster; as well as (ii) a figure with the entropy distributions of all the clusters in that resolution level. For simplicity, here we just present the results obtained for k = 7, which are provided in Table 4 and Fig 5 respectively. Nevertheless, the detail of the entropy results obtained for the other two levels of resolution can be found in the Supplementary Material S5 Table and S1 Fig. for k = 2, and S6 Table and S2 Fig for k = 15. Ridgeline plot of the entropy distributions of each cluster for k = 7. Recall that the distributions have been sorted in ascending order of entropy along the vertical axis. Table 4Summary table for k = 7.
Supervised learning approachComparison of different benchmark and highly performant classification algorithmsAs stated in the Methods section, we have compared random forest (our intended algorithm of choice) with several benchmark and high-performance classification algorithms on our supervised learning database; the scale of resolution selected has been k = 15, since it is at this level that the diversity of the middle ground strategies becomes more evident. Note that since the supervised learning database had missing data, and hence we obtained 100 imputed datasets via the MICE multiple imputation method, the comparison of the algorithms has been conducted on the 100 imputed datasets. As we explained, the classifiers selected were the following: ZeroR, OneR, AdaBoost, Support Vector Machine (SVM) with polynomial kernel, rotation forest and random forest. ZeroR and OneR are rule-based classifiers typically used to establish the baseline performance, as ZeroR trivially predicts the most frequent class, while OneR creates one rule for each of the predictors in the dataset and establishes as its ‘one rule’ the one with the smallest classification error. The rest of the classifiers chosen are high-performant classifiers [168,177]. Our experiments have been conducted in Weka [170]; stratified 10-fold nested cross-validation was used for parameter tuning in AdaBoost and the SVM with polynomial kernel (for the details of the parametrizations explored please refer to our GitHub repository: https://github.com/Virahe/Lets-go-Fishing); as regards rotation forest and random forest, in accordance to [177] tuning makes no significant improvement to them as long as reasonable parameter values are set; therefore, given that parameter tuning is extremely time-consuming, we used the parametrization they propose in Table 2, and evaluated the two algorithms by means of simple stratified 10-fold cross-validation. To assess if statistically significant differences existed between the six classifiers on our data, we conducted one-way ANOVA tests on the 10-fold cross-validation accuracies obtained for the six of them on each imputed dataset. In all the 100 cases the null hypothesis of equality of means could be rejected. Thereupon, having proved that statistically significant differences existed between the algorithms, we conducted the Duncan’s New Multiple Range Test—again on the results for the 100 datasets—to get the detail of the differences. In all the 100 cases, random forest belonged to group “a”, the group with the highest performance. In 99 out of 100 datasets group “a” was composed by random forest, rotation forest and the SVM with polynomial kernel. In the remaining dataset, random forest belonged to group “a”, rotation forest to group “ab”, and the SVM with polynomial kernel to group “b” ─recall that classifiers with the same letter are not statistically significantly different. Hence, we can conclude that since random forest belongs to the group with the highest performance in 100 out of 100 of the datasets, it is perfectly valid to justify its choice on grounds of the variable importance analyses it enables to conduct. Variable importance analysesAs we have already discussed, since random forest is not straightforwardly interpretable, it is of interest to conduct variable importance analyses to assess the relevance of a given variable or a group of variables in predicting the response. In coherence with the previous section, variable importance analyses have been conducted for k = 15. Individual variable importance analysisWe have conducted individual variable importance analysis on the 100 imputed datasets. The results provided in Fig 6 show the importance of each variable averaged across the 100 datasets. Remember that we have conducted individual variable importance analysis in accordance to Breiman [167], and that therefore, the importance of each variable is the percent increase in misclassification rate obtained after randomly permuting the values for that variable. Breiman’s individual variable importance averaged across the 100 imputed datasets. One standard deviation error bar. Group variable importance analysisGroup variable importance analysis is intended to assess the impact of an aggregated set of variables in the resulting predictive accuracy of the random forest algorithm. In the present contribution, four groups of variables have been considered. Please find below the detail of the variables that constitute each group.
The results of our group variable importance analyses averaged across the 100 imputed datasets can be found in Fig 7. Group variable importance averaged across the 100 imputed datasets. One standard deviation error bar. Group variable importance averaged across the 100 imputed datasets. One standard deviation error bar. DiscussionIn spite that standardised databases for the Digital Humanities in general, and all those available at D-PLACE [138] in particular, are not exhaustive, they are representative enough of human societies’ variability and allow to identify previously uncharted regions, hence being particularly suitable for cross-cultural studies such as the present one. Nevertheless, it is also noteworthy that since the contents in these databases are commonly quite generalist, one should be extremely careful when extrapolating conclusions to specific cases. That said, we proceed to revise our three research questions in the light of the results obtained. Question 1: Regarding subsistence strategies, are all combinations viable or do specific patterns exist?This question has been specifically addressed by means of the unsupervised learning approach; the answer we found is that clear patterns exist in the data, hence not being all combinations viable. More specifically, the PCA analysis has allowed us to identify the four main axes of variability in the subsistence database—which explain up to 100% of the variance—and the data have proved to be highly clusterable. As regards PCA analysis, remarkably, the first PCA dimension—which explains up to 52.48% of the variability—correlates quite clearly with the traditional dichotomy between hunting-gathering and agropastoralism. Thereupon, we can assert that at least for our particular EA sample of 1290 societies, the binary conceptualisation has proved to be both useful and insufficient. It is indeed useful as it explains approximately half of the variance in our data, but it is at the same time insufficient as it leaves an equally important half unexplored. Assuming that our sample may be in fact representative enough of the entire phenomenon, this finding could suggest that because of embracing the traditional dichotomic approach, half of the picture could have been disregarded. Additional PCA-derived highlights would be—as noted in the Results section—(i) the exclusive relationship found between fishing and animal husbandry in dimension 2, which corresponds to the fact that societies with a high reliance on fishing generally do not have the need to breed livestock and vice versa (i.e., pastoralists are mostly found in arid areas with no access to fishing resources); and (ii) the disjunctive from dimension 3 between agriculture and husbandry-fishing, which is clearly related to intensification strategies, namely, if it is decided to intensify on plant resources or on animal protein. Regarding cluster analysis, the fact that the subsistence dataset is significantly different from a random distribution and highly clusterable implies that recurrent specific subsistence combinations exist. As stated in the methods section, we opted for hierarchical clustering because it allows to explore the problem under consideration at different scales. More precisely, the three levels of granularity selected (k = 2, 7, 15) could be regarded as corresponding to the macro-scale, the mesoscale and a more detailed scale respectively. The macro-scale—which provides a general overview of the phenomenon—finds again the traditional division between HGF and agropastoralists; this fact complements and reinforces the reading that we made in the context of the percentage of variance explained by the first PCA dimension: that the classic binary conceptualisation corresponds to a broad-brush approach, hence being quite oversimplistic and incomplete. For this reason, here we will focus primarily on the mesoscale, since it is detailed enough to distinguish the different subsistence strategies, while being at the same time general enough so as to get a global view of the matter. Nevertheless, when of interest, we will also delve into the specifics of certain strategies at the greatest level of detail, i.e., for k = 15. Cluster interpretation for k = 7Recall that in the present section, in addition to exploring the economic structure of each of the clusters found, we will also provide some insights in relation to the biomes where they appear. For the details on geographical location and entropy values, unless otherwise specified, please refer to Fig 3 and Table 4 respectively. HGF strategies
Farming strategies
Mixed strategies
Thus far we have analysed the seven clusters found at the mesoscale, and our main conclusion is that they could be divided into two main groups: primary economies and mixed economies. More precisely, we have realised that it is the internal distribution of dependencies on the different subsistence strategies that holds the key to distinguishing between the classical binary categories and the middle ground. This is because the configuration of such internal distribution ultimately informs about the different types and degrees of specialisation. Even though specialisation is usually defined in relation to the exploitation of a narrow subset of resources [190], here we refer by specialisation to exhibiting high percentages of dependence—around 50%—on one to two subsistence strategies. Therefore, under primary economies we refer to those subsistence economies showing a strong reliance (i.e., a high specialisation) on one to two foraging or farming strategies—recall that for the case of two dominant strategies, both must be either foraging or productive. Consequently, the categories derived from the traditional HG-versus-farming divide fit reasonably well into this definition. Remarkably, in accordance with our results 75% of the societies in the EA present a primary economy, stemming the differences between the distinct primary-economy clusters from the strategy that is prioritised (if hunting is given priority over gathering/fishing, agriculture over animal husbandry, etc.). In this vein, it is also important to note that although the HG-agropastoralist disjunctive is already established by the first PCA dimension, further PCA dimensions have been considered for the identification and characterisation of the different clusters falling under primary economies. On the other hand, we refer under mixed economies to all those subsistence economies that combine both foraging and productive strategies, and that generally rely on a wider spectrum of subsistence strategies (typically three or four). Additionally, as stated in the methods section, mixed economies can alternatively be defined as a function of the information entropy; more specifically, as a consequence of their reliance on a greater number of subsistence strategies, mixed economies correspond to those alternatives with higher/the highest information entropy. Unquestionably, the most distinguishing feature of mixed economies is the combination of foraging and farming within the same subsistence economy, something which does not occur among the primary economies, and which turns them into extremely valuable examples of what the middle ground may have looked like. Ultimately, it is remarkable that up to 25% of the societies in our EA database have a mixed economy; such a relevant percentage proves once again that instead of a single intermediate category between hunting-gathering and agriculture, mixed economies constitute a complex phenomenon worthy of study in its own right. Consequently, since one of the aims of the present contribution is to gain a better understanding of human subsistence variability in general, and of the middle ground in particular, hereunder we proceed to explore the particularities of the mixed economies in greater detail (for k = 15). Mixed economies for k = 15As we saw in the Results section, at k = 15 we find further divisions of the previous seven clusters that help profile more clearly the particularities of the different alternatives. However, as previously stated, here we will limit our discussion to those economies that we have defined as mixed. In accordance with Fig 8 (the sunburst diagram), for k = 15 the agrofishing strategy is further divided into two clusters, and the same happens to the WSE. Consequently, in this section we will succinctly explore these four subdivisions, together with an additional cluster that is obtained after splitting the pastoralist cluster into two—recall that this fifth cluster will be also delved into since it happens to perfectly match the definition of mixed economies too. For the details on their geographical location and entropy values please refer to Fig 5 (map of the subsistence strategies with a relevant percentage of agriculture and/or husbandry for k = 15) and S6 Table (the summary of information entropies for k = 15) respectively. Sunburst of the different subsistence strategies identified via hierarchical clustering for k = 2, 7, 15. A short description is provided for each of the clusters. Note that the size of the circular sectors is proportional to the number of societies falling under each cluster. Agrofishing subdivisions
WSE subdivisions
Pastoralist mixed subdivision
The conclusions to be drawn from the different clusters into which mixed economies are subdivided for k = 15 may be summarised as follows:
Question 2: What is the role played by ecological settings in the configuration of the different subsistence strategies?There has been much debate about the role played by ecological variables in shaping human behaviour—see Nettle et al. [199] and a review in Ahedo et al [200]. At present, one of the most widely accepted views is that even though they condition human agency, they do not determine it. In this line, despite human cultural variability might not be directly explained on the basis of ecological traits, it is true that ecological constraints shape subsistence activities [201], and that these configure several social organisational aspects. Given our omnivorous nature, human nutritional requirements can be fulfilled in very different ways. Here is where the environmental conditions come into play, since while several ecological contexts allow the development of different and/or flexible economic strategies, many others offer limited possibilities, condition the complementarity choices to be made and restrain population growth. More specifically, whilst the carrying capacity of some settings can be further increased just through innovation, diversification, specialisation and/or intensification [202], in other contexts it is simply unfeasible to expand it (see for instance desertic areas in which agriculture cannot be put into practice). Nevertheless, it is important to note that not only the ecological variables set limits to growth; in fact, subsistence economies themselves present limitations as well, since they offer different growth possibilities depending on their structure. In HG societies with no food storage techniques, for instance, intensification is, to a major extent, limited, as increasing the hunting/gathering rates above the needs of the group would directly translate into those resources becoming spoiled [203,204]. By contrast, in fully agricultural societies with storage systems, a priori there would be no limitations to intensification other than those imposed by the land’s productivity and availability. As far as our results are concerned, they show that specific socioeconomic choices appear recurrently in certain ecological systems. While this is obvious for coastal economies, our findings may also shed light into possible macro-behavioural patterns associated with human ecology, socioecological systems and their persistence. More specifically, in the cluster interpretation from question 1 (based on Figs 3 and 4) we found that (i) pastoralism and gathering are the prevailing subsistence strategies in arid areas; (ii) agriculture is extremely versatile, being viable in a wide range of biomes as long as they are warm and have water; (iii) the primacy of hunting appears recurrently in Subarctic contexts; (iv) those HGF whose dominant strategy is fishing are commonly found in cold coastal areas; (v) in agrofishing strategies, the different percentages of dependence on fishing resources correspond to their inland or coastal location; and that (vi) WSE more heavily reliant on hunting correspond to inland locations in the vicinity of lakes and/or rivers. Additionally, these findings are in perfect coherence with the results from the group variable importance analysis (Fig 6) which show that from all the groups of variables considered, the group of the ecological variables is the most discriminant one. Lastly, as regards the results from the individual variable importance analysis (Fig 5), it is worth noting that even though the variable biome itself contributes just marginally to predicting the response (when the predictors are arranged by importance, it appears in the last third), the fact is that many of the variables within the top half are to some extent related to the biome—see agriculture intensity, crop type, temperature and precipitation-related variables, resource richness, etc. Question 3: The role of fishing in the development of viable alternatives to cultivationArchaeological and ethnographic narratives of the past 150 years have traditionally relegated fishing and aquatic-oriented societies to a marginal role; the reasons behind such relegation are to be found (among others) in (i) the postglacial sea-level rise, which submerged most of the shorelines older than 10,000 years B.P. and thus their archaeological evidence, a fact that has led to assumption that aquatic resource exploitation emerged very late in human prehistory; (ii) the differential preservation and reporting of material evidence of aquatic adaptations (such evidence is very scanty as most tools were made from biodegradable materials, and its documentation is considerably patchy since aquatic resources were held to be economically unimportant); (iii) all the additional biases of the archaeological and ethnographic records; (iv) the predominant role attributed to hunting in most traditional hunter-gatherer models; and (v) the reluctance to abandon the deep-seated HG-farmer classification and the ladder of economic and technological progress it implies [129,195]. However, the results of our analyses show that the role of fishing could have been far more relevant than previously thought. In fact, in all but the purest primary economies, the percentage of dependence on fishing is greater than 10%. For the resolution level corresponding to k = 7, we see that fishing constitutes a relevant source of food in 4 out of 7 (60%) of the clusters identified, being its contribution only marginal in clusters 2 (hunters), 4 (pastoralists) and 5 (agriculturalists). This 60% ratio is maintained for k = 15, where the percentage of dependence on fishing is found to be either equal to or greater than 10% in 9 out of 15 of the clusters. In this vein, it is also remarkable that with the exception of the strictly fishing economies (i.e., the HGF heavily reliant on fishing and the sheer fishers), the presence of fishing is a common factor across all those subsistence economies that diversify more, that is, that are more ‘mixed’. As a matter of fact, if we look at the clusters obtained for k = 15 (please refer to S6 Table in the Supplementary Material) from cluster 6.A. (agrofishing) onwards, we identify both a gradient of increasingly mixed subsistence economies, and the presence of a relevant percentage of fishing across all them. Therefore, we can assert that fishing is the common denominator of all the strategies lying between agrofishing and the WSE. Actually, agrofishing seems to play a ‘hinge’ role between the primary economies (i.e., those heavily reliant on one to two subsistence strategies and that generally correspond to the traditional stereotypes of HFG, pastoralism, agriculture and/or agropastoralism), and the mixed economies, the most paradigmatic of which is the WSE. Recall that this finding is aligned with previous contributions by Erlandson et al. [205,206] and Lepofsky et al. [196], in which they already noted that in accordance with recent findings, our ancestors would have relied on aquatic resources more heavily and for longer periods of time than the twentieth century anthropological theory once suggested. As previously mentioned, aquatic environments (coastal, riverine and/or lacustrine) constitute exceptionally rich ecosystems [195] in which—at least a priori—neither resource restoration nor limits of carrying capacity pose a problem, since both can generally be solved by simply enlarging and/or changing the catchment area. In these contexts, the main limitations are linked either to geographical constraints and/or to the development of specific nautical/resource-exploitation technologies. Due to all the above, aquatic environments can be considered singular locations that present very specific subsistence behaviours and that foster particular social dynamics. Regarding subsistence economies, aquatic societies are known to have developed sophisticated management systems to sustain or increase resource diversity and/or yields. The broad spectrum of activities, actions and strategies encompassed under management systems may be organised around four main pillars: (i) harvesting methods (see for instance clam harvesting with digging sticks, en masse fish harvesting by means of weirs, traps and nets, the extension of harvest times through the construction of holding ponds into intertidal fish traps, the establishment of harvesting rules to prevent overharvesting, etc.) [196]; (ii) enhancement strategies (such as transplanting eggs, size selection, habitat conditioning and extension: boulder clearance, construction of rock walls in the lowest intertidal zone to create clam gardens, etc.) [126,207]; (iii) tenure systems (ownership of fish harvesting locations and/or rights to catch fish from certain areas) [208]; and (iv) world view and social realm (discouragement of overharvesting, pursuit of ecosystem equilibrium, initiation ceremonies) [128]. In the ethnographic and archaeological records, paradigmatic examples of aquatic management techniques are documented—among others—for the Chulmun people in Korea [123], several coastal societies from the Atacama Desert [209] and indigenous peoples from both the NW Coast of North America [210] and the Southern Coasts of Eastern North America [211]. As far as social dynamics are concerned, aquatic-oriented societies are typically characterised by their reduced mobility [212,213] (which in most cases would have led to sedentarism or semi-sedentarism [214]), their big population numbers [111,196], and/or the emergence of social organisational changes such as higher territoriality [215] and/or increased political complexity [216,217], traits long assumed to require agriculture to develop. Eventually, we would like to return to the fact that as a consequence of their high resilience, successful aquatic/maritime adaptations could have acted as a viable alternative to cultivation [41]. Coastal/riverine/lacustrine societies would have retained their hunter-gatherer strategies or would have adopted different subsistence combinations strongly reliant on aquatic resources, since fishing would have been significantly more reliable and cost-efficient than other inland alternatives [218,219]. In this regard, it is also remarkable that the adoption of subsistence economies significantly dependent on fishing would not have been the immediate consequence of living in a coastal/aquatic environment; quite the contrary, it would have been the result of a thoughtful choice after carefully considering the risks and advantages of other alternatives; for instance, several examples attest to the retention of the original subsistence economies after having continuous contact with farming populations [130,219,220]. ConclusionsCurrent research on the transition to agriculture is characterised by both a burgeoning corpus of data and the pursuit of broader explanatory frameworks through the integration of different theoretical perspectives and the adoption of computational approaches. In this context, the present contribution proposes to explore prehistoric economies in general, and the middle ground in particular, through the quantitative analysis of ethnographically documented subsistence choices. More specifically, by means of both unsupervised and supervised learning techniques, we have assessed (i) the viability and relevance of the different subsistence combinations, (ii) the existence of associations between specific subsistence economies and certain ecological settings, and (iii) the importance of fishing in the configuration of mixed economic choices. The main conclusions obtained may be summarised as follows:
Ultimately, our contribution serves to illustrate the potential of advanced computational methods as theory-building tools; more precisely, we have showed that the application of machine learning techniques to a comprehensive standardised cross-cultural database such as D-PLACE can provide us with new insights into problems as relevant, complex and multidimensional as the transition to agriculture. Supporting informationS1 FigRidgeline plot of the entropy distributions of each cluster for k = 2. Recall that the distributions have been sorted in ascending order of entropy along the vertical axis. (EPS) S2 FigRidgeline plot of the entropy distributions of each cluster for k = 15. Recall that the distributions have been sorted in ascending order of entropy along the vertical axis. (EPS) S1 TableSuccinct summary of the main theories on the Origins of Agriculture described in the manuscript and the most relevant references. (DOCX) S2 TableCorrelation coefficients and their respective p-values (in brackets) of the first PCA dimensions with the variables in the dataset: Percentage of dependence on hunting, gathering, fishing, husbandry and agriculture. (DOCX) S3 TableOptimal number of clusters proposed by elbow method, average silhouette method, gap statistic and NbClust following the majority rule. (DOCX) S4 TableMost frequent alternatives proposed as the optimal number of clusters by the 30 different indices computed by NbClust. Note that this table is some sort of contingency table and that proposals receiving just one vote have not been included. (DOCX) S5 TableSummary table for k = 2. It includes cluster number, each cluster’s average strategy, -the average of the percentages of dependence on gathering, hunting, fishing, husbandry and agriculture across all societies in the cluster-, their entropy, standard deviation, the number of variables with a percentage of dependence equal or greater than 15% and 10%, and a succinct interpretation of the cluster. Note that the table has been sorted in ascending order of entropy. (DOCX) S6 TableSummary table for k = 15. It includes cluster number, each cluster’s average strategy, -the average of the percentages of dependence on gathering, hunting, fishing, husbandry and agriculture across all societies in the cluster-, their entropy, standard deviation, the number of variables with a percentage of dependence equal or greater than 15% and 10%, and a succinct interpretation of the cluster. Note that the table has been sorted in ascending order of entropy. (DOCX) S1 AppendixChoice of k = 7 and k = 15. (DOCX) Funding StatementThe authors acknowledge support from the Spanish Ministry of Science and Innovation: Excellence Networks (HAR2017-90883-REDC) (VA, DZ, JC, JMG) and (RED2018-102518-T) (VA, JMG), as well as the CULM Project (HAR2016-77672-P) (DZ, JC); from the Catalan Government - AGAUR through 2017 SGR 212 (DZ); from the Junta de Castilla y León – Consejería de Educación through BDNS 425389 (VA, JMG); and from the Research Foundation – Flanders (FWO) through the NASA project (VA, JC, JMG). In addition, this work was partially supported by the European Social Fund, as VA is the recipient of a predoctoral grant from the Department of Education of Junta de Castilla y León. Lastly, the publication fee was partially supported by the CSIC Open Access Publication Support Initiative through its Unit of Information Resources for Research (URICI). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. References1. Fuller DQ. An Emerging Paradigm Shift in the Origins of Agriculture. Gen Anthropol. 2010;17: 1–12. doi: 10.1111/j.1939-3466.2010.00010.x [CrossRef] [Google Scholar] 2. Smith BD. Neo-Darwinism, niche construction theory, and the initial domestication of plants and animals. Evol Ecol. 2016;30: 307–324. doi: 10.1007/s10682-015-9797-0 [CrossRef] [Google Scholar] 3. Zeder MA. Core questions in domestication research. Proc Natl Acad Sci. 2015;112: 3191–3198. doi: 10.1073/pnas.1501711112 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 4. Winterhalder B, Kennett DJ. Behavioral ecology and the transition from hunting and gathering to agriculture. In: Winterhalder B, Kennett D J., editors. Behavioral Ecology and the Transition to Agriculture. First. Berkeley and Los Angeles, CA: University of California Press; 2006. pp. 1–21. [Google Scholar] 5. Flannery K V. The Origins of Agriculture. Annu Rev Anthropol. 1973;2: 271–310. doi: 10.1146/annurev.an.02.100173.001415 [CrossRef] [Google Scholar] 6. Douglas Price T, Bar-Yosef O. The origins of agriculture: New data, new ideas. Curr Anthropol. 2011;52. doi: 10.1086/659964 [CrossRef] [Google Scholar] 7. Kintigh KW, Altschul JH, Beaudry MC, Drennan RD, Kinzig AP, Kohler TA, et al. Grand challenges for archaeology. PNAS. 2014;111: 879–880. doi: 10.1073/pnas.1324000111 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 8. Larson G, Piperno DR, Allaby RG, Purugganan MD, Andersson L, Arroyo-Kalin M, et al. Current perspectives and the future of domestication studies. Proc Natl Acad Sci. 2014;111: 6139–6146. doi: 10.1073/pnas.1323964111 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 9. Smith BD. A Comparison of Niche Construction Theory and Diet Breadth Models as Explanatory Frameworks for the Initial Domestication of Plants and Animals. J Archaeol Res. 2015;23: 215–262. doi: 10.1007/s10814-015-9081-4 [CrossRef] [Google Scholar] 10. Caldwell JR. Cultural evolution in the Old World and the New, leading to the beginnings and spread of agriculture. In: Reed CA, editor. Origins of Agriculture. The Hague: De Gruyter Mouton; 1977. pp. 77–88. [Google Scholar] 11. Darwin C. The Variation of Animals and Plants under Domestication. Vol I. London: John Murray; 1868. [Google Scholar] 12. Childe VG. The most ancient East: the Oriental prelude to European prehistory. London: Kegan Paul, Trench, Trübner & Co; 1928. [Google Scholar] 13. Braidwood RJ. Near Eastern Prehistory: The swing from food-collecting cultures to village-farming communities is still imperfectly understood. Science (80-). 1958;127: 1419–1430. doi: 10.1126/science.127.3312.1419 [PubMed] [CrossRef] [Google Scholar] 14. Ammerman AJ, Cavalli-Sforza LL. The Neolithic Transition and the Genetics of Populations in Europe. Princeton University Press; 1984. [Google Scholar] 15. Ammerman AJ, Cavalli-Sforza LL. Measuring the Rate of Spread of Early Farming in Europe. Man. 1971;6: 674. doi: 10.2307/2799190 [CrossRef] [Google Scholar] 16. Harlan JR. Crops and man. 2nd ed. Madison: American Society of Agronomy and Crop Science Society of America; 1992. [Google Scholar] 17. Tanno K, Maeda O. The Origins of Agriculture. Ancient West Asian Civilization. Singapore: Springer Singapore; 2017. pp. 87–98. doi: 10.1007/978-981-10-0554-1_6 [CrossRef] [Google Scholar] 18. Fuller DQ, Stevens CJ. Between domestication and civilization: the role of agriculture and arboriculture in the emergence of the first urban societies. Veg Hist Archaeobot. Springer New York LLC; 2019;28: 263–282. doi: 10.1007/s00334-019-00727-4 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 19. Zvelebil M, Zvelebil PEPM, Audouze F, Broodbank JDPAC, Renfrew C, Schlanger N, et al. Hunters in Transition: Mesolithic Societies of Temperate Eurasia and Their Transition to Farming. Cambridge University Press; 1986. [Google Scholar] 20. Williams NM, Hunn ES. Resource Managers: North American and Australian Hunter-Gatherers. 1st ed. Williams NM, Hunn ES, editors. New York: Routledge; 1982. doi: 10.4324/9780429304569 [CrossRef] [Google Scholar] 21. Bellwood P. First Farmers: the Origins of Agricultural Societies. Oxford, United Kingdom: Blackwell Publishing; 2004. [Google Scholar] 22. Bellwood P, Oxenham M. The Expansions of Farming Societies and the Role of the Neolithic Demographic Transition. In: Bocquet-Appel Jean-Pierre, Bar-Yosef O, editor. The Neolithic Demographic Transition and its Consequences. 1st ed. Dordrecht: Springer Netherlands; 2008. pp. 13–34. [Google Scholar] 23. Sørensen Lasse. From Hunter to Farmer in Northern Europe. Migration and adaptation during the Neolithic and Bronze Age. Acta Archaeologica Supplementa. 2014. [Google Scholar] 24. Zvelebil M. The agricultural frontier and the transition to farming in the circum-Baltic region. In: Harris DR, editor. The origins and spread of agriculture and pastoralism in Eurasia. London: UCL Press; 1996. pp. 323–45. [Google Scholar] 25. Weisdorf JL. From Foraging To Farming: Explaining The Neolithic Revolution. J Econ Surv. 2005;19: 561–586. doi: 10.1111/j.0950-0804.2005.00259.x [CrossRef] [Google Scholar] 26. Simmons AH. The Neolithic Revolution in the Near East. University of Arizona Press; 2007. [Google Scholar] 27. Childe VG. Man makes himself. New York: The New American Library; 1951. [Google Scholar] 28. Weisdorf JL. Stone Age Economics: The Origins of Agriculture and the Emergence of Non-Food Specialists. Discuss Pap 03–34, Univ Copenhagen Dep Econ. 2003. 29. Bharucha Z, Pretty J. The roles and values of wild foods in agricultural systems. Philos Trans R Soc B Biol Sci. 2010;365: 2913–2926. doi: 10.1098/rstb.2010.0123 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 30. Bender B. Prehistoric Developments in the American Midcontinent and in Brittany, Northwest France. Prehistoric Hunters-Gatherers. Elsevier; 1985. pp. 21–57. doi: 10.1016/B978-0-12-564750-2.50007-X [CrossRef] [Google Scholar] 31. Williams GC. Gaia, Nature Worship and Biocentric Fallacies. Q Rev Biol. University of Chicago Press; 1992;67: 479–486. doi: 10.1086/417796 [CrossRef] [Google Scholar] 32. Binford LR. Post-Pleistocene Adaptations. In: Binford LR, Binford SR, editors. New Perspectives in Archaeology. Chicago: Aldine; 1968. pp. 313–41. [Google Scholar] 33. Flannery K V. Origins and ecological effects of early domestication in Iran and the Near East. In: Ucko PJ, Dimblebly GW, editors. The Domestication and Exploitation of Plants and Animals. Chicago: Aldine; 1969. pp. 73–100. [Google Scholar] 34. Hassan F. Demographic Archaeology. London: Academic Press; 1981. [Google Scholar] 35. Binford LR. In Pursuit of the Past. London: Thames and Hudson; 1983. [Google Scholar] 36. Cohen MN. The Food Crisis in Prehistory: Overpopulation and the Origins of Agriculture. New Haven, Conn.: Yale University Press; 1977. [Google Scholar] 37. Cohen MN. Introduction: Rethinking the Origins of Agriculture. Curr Anthropol. 2009;50: 591–595. doi: 10.1086/603548 [PubMed] [CrossRef] [Google Scholar] 38. Piperno DR. The Origins of Plant Cultivation and Domestication in the Neotropics: A Behavioural Ecological Perspective. In: Kennett DJ, Winterhalder B, editors. Behavioral Ecology and the Transition to Agriculture. 1st ed. University of California Press; 2006. pp. 137–166. [Google Scholar] 39. Piperno DR. The Origins of Plant Cultivation and Domestication in the New World Tropics. Curr Anthropol. 2011;52: 453–S470. doi: 10.1086/659998 [CrossRef] [Google Scholar] 40. Zvelebil M. From Forager to Farmer in the Boreal Zone: reconstructing economic patterns through catchment analysis in prehistoric Finland. Oxford: BAR International Series 115 (I and II); 1981. [Google Scholar] 41. Zvelebil M, Rowley‐Conwy P. Transition to farming in Northern Europe: A hunter‐gatherer perspective. Nor Archaeol Rev. 1984;17: 104–128. doi: 10.1080/00293652.1984.9965402 [CrossRef] [Google Scholar] 42. Richerson PJ, Boyd R, Bettinger RL. Was Agriculture Impossible during the Pleistocene but Mandatory during the Holocene? A Climate Change Hypothesis. Am Antiq. 2001;66: 387–411. doi: 10.2307/2694241 [CrossRef] [Google Scholar] 43. Bar-Yosef O. Climatic Fluctuations and Early Farming in West and East Asia. Curr Anthropol. 2011;52: S175–S193. doi: 10.1086/659784 [CrossRef] [Google Scholar] 44. Ferrio JP, Voltas J, Araus J. Global change and the origins of agriculture. Crop Stress Management and Global Climate Change. 2011. pp. 1–14. [Google Scholar] 45. Sage RF. Was low atmospheric CO2 during the Pleistocene a limiting factor for the origin of agriculture? Glob Chang Biol. 1995;1: 93–106. doi: 10.1111/j.1365-2486.1995.tb00009.x [CrossRef] [Google Scholar] 46. Willcox G, Buxo R, Herveux L. Late Pleistocene and early Holocene climate and the beginnings of cultivation in northern Syria. The Holocene. 2009;19: 151–158. doi: 10.1177/0959683608098961 [CrossRef] [Google Scholar] 47. Feynman J, Ruzmaikin A. Climate stability and the development of agricultural societies. Clim Change. 2007;84: 295–311. doi: 10.1007/s10584-007-9248-1 [CrossRef] [Google Scholar] 48. Gupta AK. Origin of agriculture and domestication of plants and animals linked to early Holocene climate amelioration. Curr Sci. Temporary Publisher; 2004;87: 54–59. [Google Scholar] 49. Bar-Yosef O. The Natufian culture in the Levant, threshold to the origins of agriculture. Evol Anthropol Issues, News, Rev. 1998;6: 159–177. [Google Scholar] 50. Lee R. B., & DeVore I, editor. Man the hunter. Chicago: Aldine; 1968. [Google Scholar] 51. Lee RB. Work effort, group structure and land use in contemporary hunter-gatherers. In: Ucko PJ, Tringham R, Dimbleby GW, editors. Man, Settlement and Urbanism. London: Duckworth; 1972. pp. 177–85. [Google Scholar] 52. Lee RB. The !Kung San. Cambridge: Cambridge University Press; 1979. [Google Scholar] 53. Bender B. Gatherer‐hunter to farmer: A social perspective. World Archaeol. 1978;10: 204–222. doi: 10.1080/00438243.1978.9979731 [CrossRef] [Google Scholar] 54. Sahlins M. Stone Age Economics. London: Tavistock Publications; 1972. [Google Scholar] 55. Godelier M. Perspectives in Marxist Anthropology. Cambridge: Cambridge University Press; 1977. [Google Scholar] 56. Bender B. Emergent Tribal Formations in the American Midcontinent. Am Antiq. Society for American Archaeology; 1985;50: 52–62. doi: 10.2307/280633 [CrossRef] [Google Scholar] 57. Alexander RD. Evolution, Human Behavior, and Determinism. PSA Proc Bienn Meet Philos Sci Assoc. 1976;1976: 3–21. doi: 10.1086/psaprocbienmeetp.1976.2.192370 [CrossRef] [Google Scholar] 58. Gremillion KJ. Human Behavioral Ecology and Plant Resources in Archaeological Research. Handbook of Evolutionary Research in Archaeology. Cham: Springer International Publishing; 2019. pp. 255–273. doi: 10.1007/978-3-030-11117-5_13 [CrossRef] [Google Scholar] 59. Codding BF, Bird DW. Behavioral ecology and the future of archaeological science. J Archaeol Sci. 2015;56: 9–20. doi: 10.1016/j.jas.2015.02.027 [CrossRef] [Google Scholar] 60. Winterhalder B, Kennett DJ. Four Neglected Concepts with a Role to Play in Explaining the Origins of Agriculture. Curr Anthropol. 2009;50: 645–648. doi: 10.1086/605355 [PubMed] [CrossRef] [Google Scholar] 61. Zeder MA. Domestication as a model system for niche construction theory. Evol Ecol. 2016;30: 325–348. doi: 10.1007/s10682-015-9801-8 [CrossRef] [Google Scholar] 62. Smith BD. A Cultural Niche Construction Theory of Initial Domestication. Biol Theory. 2011;6: 260–271. doi: 10.1007/s13752-012-0028-4 [CrossRef] [Google Scholar] 63. Gremillion KJ, Barton L, Piperno DR. Particularism and the retreat from theory in the archaeology of agricultural origins. Proc Natl Acad Sci. 2014;111: 6171–6177. doi: 10.1073/pnas.1308938110 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 64. Stiner MC, Kuhn SL. Are we missing the “sweet spot” between optimality theory and niche construction theory in archaeology? J Anthropol Archaeol. 2016;44: 177–184. doi: 10.1016/j.jaa.2016.07.006 [CrossRef] [Google Scholar] 65. Broughton JM, Cannon MD, Bartelink EJ. Evolutionary Ecology, Resource Depression, and Niche Construction Theory: Applications to Central California Hunter-Gatherers and Mimbres-Mogollon Agriculturalists. J Archaeol Method Theory. 2010;17: 371–421. doi: 10.1007/s10816-010-9095-7 [CrossRef] [Google Scholar] 66. Freeman J, Peeples MA, Anderies JM. Toward a theory of non-linear transitions from foraging to farming. J Anthropol Archaeol. 2015;40: 109–122. doi: 10.1016/j.jaa.2015.07.001 [CrossRef] [Google Scholar] 67. Mohlenhoff KA, Coltrain JB, Codding BF. Optimal foraging theory and niche-construction theory do not stand in opposition. Proc Natl Acad Sci. 2015;112: E3093–E3093. doi: 10.1073/pnas.1507637112 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 68. Mohlenhoff KA, Codding BF. When does it pay to invest in a patch? The evolution of intentional niche construction. Evol Anthropol Issues, News, Rev. 2017;26: 218–227. doi: 10.1002/evan.21534 [PubMed] [CrossRef] [Google Scholar] 69. Piperno DR, Ranere AJ, Dickau R, Aceituno F. Niche construction and optimal foraging theory in Neotropical agricultural origins: A re-evaluation in consideration of the empirical evidence. J Archaeol Sci. 2017;78: 214–220. doi: 10.1016/j.jas.2017.01.001 [CrossRef] [Google Scholar] 70. Foley R. Optimality Theory in Anthropology. Man. [Wiley, Royal Anthropological Institute of Great Britain and Ireland]; 1985;20: 222–242. doi: 10.2307/2802382 [CrossRef] [Google Scholar] 71. Mulder MB. Human Behavioural Ecology. eLS. Chichester, UK: John Wiley & Sons, Ltd; 2005. doi: 10.1038/npg.els.0003671 [CrossRef] [Google Scholar] 72. Macarthur RH, Pianka ER. On Optimal Use of a Patchy Environment. Am Nat. 1966;100: 603–609. [Google Scholar] 73. Schoener TW. Theory of Feeding Strategies. Annu Rev Ecol Syst. 1971;2: 369–404. doi: 10.1146/annurev.es.02.110171.002101 [CrossRef] [Google Scholar] 74. Hardesty DL. The Human Ecological Niche. Am Anthropol. 1972;74: 458–466. doi: 10.1525/aa.1972.74.3.02a00150 [CrossRef] [Google Scholar] 75. Hardesty D. The Use of General Ecological Principles in Archaeology. Adv Archaeol Method Theory. 1980;3: 157–187. [Google Scholar] 76. Odling-Smee FJ, Laland KN, Feldman MW. Niche Construction: The neglected process in evolution. Princeton: Princeton University Press; 2003. [Google Scholar] 77. Smith BD. Niche construction and the behavioral context of plant and animal domestication. Evol Anthropol Issues, News, Rev. 2007;16: 188–199. doi: 10.1002/evan.20135 [CrossRef] [Google Scholar] 78. Zeder MA. The Broad Spectrum Revolution at 40: Resource diversity, intensification, and an alternative to optimal foraging explanations. J Anthropol Archaeol. 2012;31: 241–264. doi: 10.1016/j.jaa.2012.03.003 [CrossRef] [Google Scholar] 79. Zeder MA, Smith BD. A Conversation on Agricultural Origins: Talking Past Each Other in a Crowded Room. Curr Anthropol. 2009;50: 681–690. doi: 10.1086/605553 [CrossRef] [Google Scholar] 80. Barker G. The agricultural revolution in Prehistory. Oxford: Oxford University Press; 2006. [Google Scholar] 81. Aiello LC. The Origins of Agriculture: New Data, New Ideas. Curr Anthropol. 2011;52: S161–S162. doi: 10.1086/660154 [CrossRef] [Google Scholar] 82. Nadel D, Piperno DR, Holst I, Snir A, Weiss E. New evidence for the processing of wild cereal grains at Ohalo II, a 23 000-year-old campsite on the shore of the Sea of Galilee, Israel. Antiquity. 2012;86: 990–1003. doi: 10.1017/S0003598X00048201 [CrossRef] [Google Scholar] 83. Weiss E, Kislev ME, Simchoni O, Nadel D. Small-grained wild grasses as staple food at the 23 000-year-old site of Ohalo II, Israel. Econ Bot. 2004;58: S125–S134. doi: 10.1663/0013-0001(2004)58[S125:SWGASF]2.0.CO;2 [CrossRef] [Google Scholar] 84. Diamond J. Evolution, consequences and future of plant and animal domestication. Nature. 2002;418: 700–707. doi: 10.1038/nature01019 [PubMed] [CrossRef] [Google Scholar] 85. Fuller DQ, Denham T, Arroyo-Kalin M, Lucas L, Stevens CJ, Qin L, et al. Convergent evolution and parallelism in plant domestication revealed by an expanding archaeological record. Proc Natl Acad Sci. 2014;111: 6147–6152. doi: 10.1073/pnas.1308937110 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 86. Smith BD. Low-Level Food Production. J Archaeol Res. 2001;9: 1–43. doi: 10.1023/A:1009436110049 [CrossRef] [Google Scholar] 87. Zeder MA. The Origins of Agriculture in the Near East. Curr Anthropol. 2011;52: S221–S235. doi: 10.1086/659307 [CrossRef] [Google Scholar] 88. Zvelebil M. The agricultural transition and the origins of Neolithic society in Europe. Doc Praehist. 2001;28: 1–26. doi: 10.4312/dp.28.1 [CrossRef] [Google Scholar] 89. Fuller DQ, Kingwell-Banham E, Lucas L, Murphy C, Stevens CJ. Comparing Pathways to Agriculture. Archaeol Int. 2015;18: 61–66. doi: 10.5334/ai.1808 [CrossRef] [Google Scholar] 90. Bueno Ramírez P, de Balbín Behrmann R. De cazadores a productores. Transiciones y tradiciones. In: de Valencia Diputación de P de V M, editor. Del Neolític a l’Edat de Bronze en el Meditarrani occidental. Valencia; 2016. pp. 465–480. [Google Scholar] 91. Zeder MA. Central questions in the domestication of plants and animals. Evol Anthropol Issues, News, Rev. 2006;15: 105–117. doi: 10.1002/evan.20101 [CrossRef] [Google Scholar] 92. Smith B. Documenting domesticated plants in the archaeological record. In: Zeder MA, Bradley DG, Emshwiller E, Smith BD, editors. Documenting Domestication: New Genetic and Archaeological Paradigms. 1st ed. Berkeley: University of California Press; 2006. pp. 15–24. [Google Scholar] 93. Terrell JE, Hart JP, Barut S, Cellinese N, Curet A, Denham T, et al. Domesticated Landscapes: The Subsistence Ecology of Plant and Animal Domestication. J Archaeol Method Theory. 2003;10: 323–368. doi: 10.1023/B:JARM.0000005510.54214.57 [CrossRef] [Google Scholar] 94. Bogucki P. Prelude to agriculture in north-central Europe. Before Farming: Hunter-Gatherer Society and Subsistence, MASCA Research Papers in Science and Archaeology, Supplement to Volume 12. Philadelphia: University of Pennsylvania Museum of Archaeology and Anthropology; 1995. pp. 105–116. [Google Scholar] 95. Ford RI. The processes of plant food production in prehistoric North America. In: Ford RI, editor. Prehistoric Food Production in North America. Ann Arbor: Anthropological Papers, No. 75, Museum of Anthropology, University of Michigan; 1985. pp. 1–18. [Google Scholar] 96. Harris DR. An evolutionary continuum of people-plant interaction. Foraging and Farming: The Evolution of Plant Exploitation. London: Unwin Hyman; 1989. pp. 11–26. [Google Scholar] 97. Harris DR. Settling Down and Breaking Ground: Rethinking the Neolithic Revolution. Amsterdam: Twaalfde Kroon-Voordracht, Stichting Nederlands Museum voor Anthropologie en Praehistorie; 1990. [Google Scholar] 98. Harris DR. Introduction: Themes and concepts in the study of early agriculture. The origins and spread of agriculture and pastoralism in Eurasia. Washington: Smithsonian Institution Press; 1996. pp. 1–11. [Google Scholar] 99. Harris DR. Domesticatory Relationships of People, Plants and Animals. In: Ellen RF, Fukui K, editors. Redefining Nature: Ecology, Culture, and Domestication. Oxford: Berg; 1996. pp. 437–463. [Google Scholar] 100. Higgs ES. Paleoeconomy. In: Higgs ES, editor. Paleoeconomy. Cambridge: Cambridge University Press; 1975. pp. 1–7. [Google Scholar] 101. Zvelebil M. Hunting, gathering, or husbandry? Management of food resources by the late Mesolithic communities of temperate Europe. In: Campana D, editor. Before Farming: Hunter-Gatherer Society and Subsistence, MASCA Research Papers in Science and Archaeology, Supplement to Volume 12. Philadelphia: University of Pennsylvania Museum of Archaeology and Anthropology; 1995. pp. 79–104. [Google Scholar] 102. Jiao T. Toward an alternative perspective on the foraging and low-level food production on the coast of China. Quat Int. 2016;419: 54–61. doi: 10.1016/j.quaint.2015.06.060 [CrossRef] [Google Scholar] 103. Rowley-Conwy P, Layton R. Foraging and farming as niche construction: stable and unstable adaptations. Philos Trans R Soc B Biol Sci. 2011;366: 849–862. doi: 10.1098/rstb.2010.0307 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 104. Crawford GW. East Asian Plant Domestication. Archaeology of Asia. Oxford, UK: Blackwell Publishing Ltd; 2006. pp. 77–95. doi: 10.1002/9780470774670.ch5 [CrossRef] [Google Scholar] 105. Crawford GW. Advances in Understanding Early Agriculture in Japan. Curr Anthropol. 2011;52: S331–S345. doi: 10.1086/658369 [CrossRef] [Google Scholar] 106. Greaves RD, Kramer KL. Hunter–gatherer use of wild plants and domesticates: archaeological implications for mixed economies before agricultural intensification. J Archaeol Sci. 2014;41: 263–271. [Google Scholar] 107. Burnsilver S, Magdanz J, Stotts R, Berman M, Kofinas G. Are Mixed Economies Persistent or Transitional? Evidence Using Social Networks from Arctic Alaska. Am Anthropol. 2016;118: 121–129. doi: 10.1111/aman.12447 [CrossRef] [Google Scholar] 108. Tharakan C. G. The Mixed Economy of the South Indian Kurumbas. Ethnology. 2003;42: 323–334. doi: 10.2307/3773832 [CrossRef] [Google Scholar] 109. Svizzero S. Foraging Wild Resources: Evolving Goals of an Ubiquitous Human Behavior. Anthropology. 2016;04. doi: 10.4172/2332-0915.1000161 [CrossRef] [Google Scholar] 110. Svizzero S. Hunting Strategies with Cultivated Plants as Bait and the Prey Pathway to Animal Domestication. Int J Res Sociol Anthropol. 2016;2. doi: 10.20431/2454-8677.0202007 [CrossRef] [Google Scholar] 111. Sassaman KE. Complex Hunter–Gatherers in Evolution and History: A North American Perspective. J Archaeol Res. 2004;12: 227–280. doi: 10.1023/B:JARE.0000040231.67149.a8 [CrossRef] [Google Scholar] 112. Kim J, Grier C. Beyond Affluent Foragers. In: Grier C, Kim J, Uchiyama J, editors. Beyond Affluent-foragers: Rethinking Hunter-gatherer complexity. Oxbow Books; 2006. pp. 192–200. doi: 10.2307/j.ctt1w0df3b.16 [CrossRef] [Google Scholar] 113. Sapignoli M. Mobility, Land Use, and Leadership in Small-Scale and Middle-Range Societies. Rev Anthropol. 2014;43: 35–78. doi: 10.1080/00938157.2014.872463 [CrossRef] [Google Scholar] 114. Ames KM. Complex Hunter-Gatherers. In: Smith C, editor. Encyclopedia of Global Archaeology. New York, NY: Springer New York; 2014. pp. 1613–1621. doi: 10.1007/978-1-4419-0465-2_964 [CrossRef] [Google Scholar] 115. Harris DR, Hillman GC, editors. Foraging and Farming: The Evolution of Plant Exploitation. London: Unwin Hyman; 1989. [Google Scholar] 116. Carneiro RL. Slash-and-burn Agriculture: A Closer Look at its Implications for Settlement Patterns. In: Wallace AFC, editor. Men and Cultures. Philadelphia: University of Pennsylvania Press; 1960. doi: 10.9783/9781512819526-037 [CrossRef] [Google Scholar] 117. Arroyo-Kalin M. Slash-burn-and-churn: Landscape history and crop cultivation in pre-Columbian Amazonia. Quat Int. 2012;249: 4–18. doi: 10.1016/j.quaint.2011.08.004 [CrossRef] [Google Scholar] 118. Palmisano A, Bevan A, Shennan S. Comparing archaeological proxies for long-term population patterns: An example from central Italy. J Archaeol Sci. 2017;87: 59–72. doi: 10.1016/j.jas.2017.10.001 [CrossRef] [Google Scholar] 119. Robb J. The Early Mediterranean Village. Cambridge: Cambridge University Press; 2007. doi: 10.1017/CBO9780511499647 [CrossRef] [Google Scholar] 120. Medina ME, Pastor S, Recalde A. The archaeological landscape of Late Prehispanic mixed foraging and cultivation economy (Sierras of Córdoba, Argentina). J Anthropol Archaeol. 2016;42: 88–104. doi: 10.1016/j.jaa.2016.04.003 [CrossRef] [Google Scholar] 121. Madsen DB, Simms SR. The Fremont Complex: A Behavioral Perspective. J World Prehistory. 1998;12: 255–336. [Google Scholar] 122. Svizzero S, Tisdell C. Theories about the Commencement of Agriculture in Prehistoric Societies: A Critical Evaluation. Riv di Stor Econ. 2014; 255–280. [Google Scholar] 123. Lee G-A. The Transition from Foraging to Farming in Prehistoric Korea. Curr Anthropol. 2011;52: S307–S329. doi: 10.1086/658488 [CrossRef] [Google Scholar] 124. Crawford GW, Takamiya H. ASIA, EAST | Japanese Archipelago, Prehistoric Hunter-Fisher-Gatherers. Encyclopedia of Archaeology. Elsevier; 2008. pp. 637–641. doi: 10.1016/B978-012373962-9.00260-0 [CrossRef] [Google Scholar] 125. Crawford GW. The Jomon in early agriculture discourse: issues arising from Matsui, Kanehara and Pearson. World Archaeol. 2008;40: 445–465. doi: 10.1080/00438240802451181 [CrossRef] [Google Scholar] 126. Lepofsky D, Smith NF, Cardinal N, Harper J, Morris M, (Elroy White) G, et al. Ancient Shellfish Mariculture on the Northwest Coast of North America. Am Antiq. 2015;80: 236–259. doi: 10.7183/0002-7316.80.2.236 [CrossRef] [Google Scholar] 127. Turner NJ, Lepofsky D, Deur D. Plant management systems of British Columbia’s first peoples. BC Stud Br Columbian Q. 2013; 107–133. [Google Scholar] 128. Mathews DL, Turner NJ. Ocean cultures: Northwest Coast ecosystems and indigenous management systems. Conservation for the Anthropocene Ocean. Elsevier; 2017. pp. 169–206. [Google Scholar] 129. Erlandson JM. The Archaeology of Aquatic Adaptations: Paradigms for a New Millennium. J Archaeol Res. Springer; 2001;9: 287–350. doi: 10.1023/A:1013062712695 [CrossRef] [Google Scholar] 130. Svizzero S, Tisdell C. The Persistence of Hunting and Gathering Economies. Soc Evol Hist. 2015;15. [Google Scholar] 131. BurnSilver S, Magdanz J. Heterogeneity in mixed economies. Hunt Gatherer Res. 2019;3: 601–633. doi: 10.3828/hgr.2017.31 [CrossRef] [Google Scholar] 132. Wenzel GW. Canadian Inuit subsistence: Antinomies of the mixed economy. Hunt Gatherer Res. 2019;3: 567–581. [Google Scholar] 133. Cunningham J. Ethnoarchaeology Beyond Correlates. Ethnoarchaeology. 2009;1: 115–136. doi: 10.1179/eth.2009.1.2.115 [CrossRef] [Google Scholar] 134. Skibo J. Archaeological Theory and Snake-Oil Peddling: The Role of Ethnoarchaeology in Archaeology. Ethnoarchaeology. 2009;1: 27–56. doi: 10.1179/eth.2009.1.1.27 [CrossRef] [Google Scholar] 135. Binford LR. Constructing Frames of Reference: An analytical method for archaeological theory building using ethnographic and environmental data sets. First. London: University of California Press; 2019. [Google Scholar] 136. Fischer MD, Lyon SM, Sosna D, Henig D. Harmonizing Diversity: Tuning Anthropological Research to Complexity. Soc Sci Comput Rev. 2013;31: 3–15. doi: 10.1177/0894439312455311 [CrossRef] [Google Scholar] 137. Fischer MD, Ember CR. Big Data and Research Opportunities Using HRAF Databases. In: Chen S-H, editor. Big Data in Computational Social Science and Humanities. Cham: Springer; 2018. pp. 323–336. doi: 10.1007/978-3-319-95465-3_17 [CrossRef] [Google Scholar] 138. Kirby KR, Gray RD, Greenhill SJ, Jordan FM, Gomes-Ng S, Bibiko HJ, et al. D-PLACE: A global database of cultural, linguistic and environmental diversity. PLoS One. 2016;11: 1–14. doi: 10.1371/journal.pone.0158391 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 140. Ethnographic Atlas. Ethnology. 1962;1: 113. doi: 10.2307/3772932 [CrossRef] [Google Scholar] 141. Gray PJ. Ethnographic Atlas Codebook. World Cult. 1998;10: 86–136. [Google Scholar] 142. D-PLACE—About [Internet]. 143. Giuliano P, Nunn N. Ancestral Characteristics of Modern Populations. Econ Hist Dev Reg. 2018;33: 1–17. doi: 10.1080/20780389.2018.1435267 [CrossRef] [Google Scholar] 144. D-PLACE—Dataset Ethnographic Atlas [Internet]. 145. Ember CR, Page H, Martin MM, O’Leary T. A computerized concordance of cross-cultural samples. New Haven: Human Relations Area Files; 1992. [Google Scholar] 146. Jenkins CN, Pimm SL, Joppa LN. Global patterns of terrestrial vertebrate diversity and conservation. Proc Natl Acad Sci. 2013;110: E2602–E2610. doi: 10.1073/pnas.1302251110 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 147. Kreft H, Jetz W. Global patterns and determinants of vascular plant diversity. Proc Natl Acad Sci. 2007;104: 5925–5930. doi: 10.1073/pnas.0608361104 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 148. NASA. Net Primary Productivity (1 month—TERRA/MODIS) [Internet]. 150. Dinerstein E, Olson D, Joshi A, Vynne C, Burgess ND, Wikramanayake E, et al. An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm. Bioscience. 2017;67: 534–545. doi: 10.1093/biosci/bix014 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 151. Lima-Ribeiro MS. EcoClimate: a database of climate data from multiple models for past, present, and future for macroecologists and biogeographers. Biodivers Informatics. 2015;10. doi: 10.17161/bi.v10i0.4955 [CrossRef] [Google Scholar] 152. James G, Witten D, Hastie T, Tibshirani R. An Introduction to Statistical Learning. 1st ed. Casella G, Fienberg S, Olkin I, editors. New York, NY: Springer New York; 2013. doi: 10.1007/978-1-4614-7138-7 [CrossRef] [Google Scholar] 153. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. 2nd ed. New York, NY: Springer New York; 2009. doi: 10.1007/978-0-387-84858-7 [CrossRef] [Google Scholar] 154. Jain AK, Dubes RC. Algorithms for Clustering Data. Englewood Cliffs: Prentice Hall; 1988. [Google Scholar] 155. Kassambara A. Multivariate Analysis 1: Practical Guide To Cluster Analysis in R. Taylor & Francis Group. STHDA; 2015. [Google Scholar] 156. Monti S, Tamayo P, Mesirov J, Golub T. Consensus Clustering: A Resampling-Based Method for Class Discovery and Visualization of Gene Expression Microarray Data. Mach Learn. 2003;52: 91–118. doi: 10.1023/A:1023949509487 [CrossRef] [Google Scholar] 157. Bezdek JC, Hathaway RJ. VAT: a tool for visual assessment of (cluster) tendency. Proceedings of the 2002 International Joint Conference on Neural Networks IJCNN’02 (Cat No02CH37290). IEEE; 2002. pp. 2225–2230. 158. Kaufman L, Rousseeuw PJ. Finding Groups in Data: An Introduction to Cluster Analysis. New York: Wiley; 1990. [Google Scholar] 159. Tibshirani R, Walther G, Hastie T. Estimating the number of clusters in a data set via the gap statistic. J R Stat Soc Ser B (Statistical Methodol. 2001;63: 411–423. doi: 10.1111/1467-9868.00293 [CrossRef] [Google Scholar] 160. Charrad M, Ghazzali N, Boiteau V, Niknafs A. NbClust : An R Package for Determining the Relevant Number of Clusters in a Data Set. J Stat Softw. 2014;61. doi: 10.18637/jss.v061.i06 [CrossRef] [Google Scholar] 162. Shannon CE. A Mathematical Theory of Communication. Bell Syst Tech J. 1948;27: 379–423. doi: 10.1002/j.1538-7305.1948.tb01338.x [CrossRef] [Google Scholar] 163. van Buuren S. Multiple imputation of discrete and continuous data by fully conditional specification. Stat Methods Med Res. 2007;16: 219–242. doi: 10.1177/0962280206074463 [PubMed] [CrossRef] [Google Scholar] 164. Buuren S van, Groothuis-Oudshoorn K. mice : Multivariate Imputation by Chained Equations in R. J Stat Softw. 2011;45. [Google Scholar] 165. Doove LL, Van Buuren S, Dusseldorp E. Recursive partitioning for missing data imputation in the presence of interaction effects. Comput Stat Data Anal. Elsevier B.V.; 2014;72: 92–104. doi: 10.1016/j.csda.2013.10.025 [CrossRef] [Google Scholar] 166. Shah AD, Bartlett JW, Carpenter J, Nicholas O, Hemingway H. Comparison of random forest and parametric imputation models for imputing missing data using MICE: A CALIBER study. Am J Epidemiol. 2014;179: 764–774. doi: 10.1093/aje/kwt312 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 167. Breiman L. Random Forests. Mach Learn. 2001;45: 5–32. doi: 10.1023/A:1010933404324 [CrossRef] [Google Scholar] 168. Fernández-Delgado M, Cernadas E, Barro S, Amorim D. Do we Need Hundreds of Classifiers to Solve Real World Classification Problems? J Mach Learn Res. 2014;15: 3133–3181. [Google Scholar] 169. Duncan DB. Multiple Range and Multiple F Tests. Biometrics. 1955;11: 1. doi: 10.2307/3001478 [CrossRef] [Google Scholar] 170. Frank E, Hall MA, Witten IH. The WEKA Workbench. Online Appendix. Data Mining: Practical Machine Learning Tools and Techniques. Fourth. San Francisco: Morgan Kaufmann; 2016. [Google Scholar] 171. Holte RC. Very Simple Classification Rules Perform Well on Most Commonly Used Datasets. Mach Learn. 1993;11: 63–90. doi: 10.1023/A:1022631118932 [CrossRef] [Google Scholar] 172. Schapire RE. Explaining AdaBoost. In: Schölkopf B, Luo Z, Vovk V, editors. Empirical Inference. Berlin, Heidelberg: Springer Verlag; 2013. pp. 37–52. doi: 10.1007/978-3-642-41136-6_5 [CrossRef] [Google Scholar] 173. Cortes C, Vapnik V. Support-vector networks. Mach Learn. 1995;20: 273–297. doi: 10.1007/BF00994018 [CrossRef] [Google Scholar] 174. Rodriguez JJ, Kuncheva LI, Alonso CJ. Rotation Forest: A New Classifier Ensemble Method. IEEE Trans Pattern Anal Mach Intell. 2006;28: 1619–1630. doi: 10.1109/TPAMI.2006.211 [PubMed] [CrossRef] [Google Scholar] 175. Gregorutti B, Michel B, Saint-Pierre P. Grouped variable importance with random forests and application to multiple functional data analysis. Comput Stat Data Anal. Elsevier B.V.; 2015;90: 15–35. doi: 10.1016/j.csda.2015.04.002 [CrossRef] [Google Scholar] 176. Kassambara A, Mundt F. factoextra: Extract and Visualize the Results of Multivariate Data Analyses [Internet]. 2020. 177. Bagnall A, Flynn M, Large J, Line J, Bostrom A, Cawley G. Is rotation forest the best classifier for problems with continuous features? 2018. [Google Scholar] 178. Olson DM, Dinerstein E, Wikramanayake ED, Burgess ND, Powell GVN, Underwood EC, et al. Terrestrial Ecoregions of the World: A New Map of Life on Earth: A new global map of terrestrial ecoregions provides an innovative tool for conserving biodiversity. Bioscience. 2001;51: 933–938. doi: 10.1641/0006-3568(2001)051[0933:TEOTWA]2.0.CO;2 [CrossRef] [Google Scholar] 179. Clarke PA. Aboriginal people and their plants. Kenthurst, N.S.W.: Rosenberg Publishing; 2011. [Google Scholar] 180. Knight MH. Tsama melons, Citrullus lanatus, a supplementary water supply for wildlife in the southern Kalahari. Afr J Ecol. 1995;33: 71–80. doi: 10.1111/j.1365-2028.1995.tb00782.x [CrossRef] [Google Scholar] 181. Wills WH. Early prehistoric agriculture in the American Southwest. School of American Research Press; 1988. [Google Scholar] 182. Nabhan GP, Riordan EC, Monti L, Rea AM, Wilder BT, Ezcurra E, et al. An Aridamerican model for agriculture in a hotter, water scarce world. Plants, People, Planet. 2020;2: 627–639. doi: 10.1002/ppp3.10129 [CrossRef] [Google Scholar] 183. Speth JD, Spielmann KA. Energy source, protein metabolism, and hunter-gatherer subsistence strategies. J Anthropol Archaeol. 1983;2: 1–31. doi: 10.1016/0278-4165(83)90006-5 [CrossRef] [Google Scholar] 184. Koblents-Mishke OI. Primary production in the Pacific. Oceanology. 1965;5: 104–116. [Google Scholar] 185. Stone GD, Downum CE. Non-Boserupian Ecology and Agricultural Risk: Ethnic Politics and Land Control in the Arid Southwest. Am Anthropol. 1999;101: 113–128. doi: 10.1525/aa.1999.101.1.113 [CrossRef] [Google Scholar] 186. Berbesque JC, Marlowe FW, Shaw P, Thompson P. Hunter–gatherers have less famine than agriculturalists. Biol Lett. 2014;10: 20130853. doi: 10.1098/rsbl.2013.0853 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 187. Nixon SW, Oviatt CA, Frithsen J, Sullivan B. Nutrients and the productivity of estuarine and coastal marine ecosystems. J Limnol Soc South Africa. 1986;12: 43–71. [Google Scholar] 188. Fitzhugh B. Complex Hunter-Gatherers on the Kodiak Archipelago. In: Fitzhugh B, editor. The Evolution of Complex Hunter-Gatherers. Springer US; 2003. pp. 67–85. [Google Scholar] 189. Pestle WJ, Torres-Rouff C, Gallardo F, Ballester B, Clarot A. Mobility and Exchange among Marine Hunter-Gatherer and Agropastoralist Communities in the Formative Period Atacama Desert. Curr Anthropol. 2015;56: 121–133. doi: 10.1086/679594 [CrossRef] [Google Scholar] 190. Byers DA, Ugan A. Should we expect large game specialization in the late Pleistocene? An optimal foraging perspective on early Paleoindian prey choice. J Archaeol Sci. 2005;32: 1624–1640. doi: 10.1016/j.jas.2005.05.003 [CrossRef] [Google Scholar] 191. Western D. The Environment and Ecology of Pastoralists in Arid Savannas. Dev Change. 1982;13: 183–211. doi: 10.1111/j.1467-7660.1982.tb00117.x [CrossRef] [Google Scholar] 192. Larson DO, Johnson JR, Michaelsen JC. Missionization among the Coastal Chumash of Central California: A Study of Risk Minimization Strategies. Am Anthropol. 1994;96: 263–299. doi: 10.1525/aa.1994.96.2.02a00020 [CrossRef] [Google Scholar] 193. Jochim MA. Strategies for Survival: Cultural Behavior in an Ecological Context. New York: Academic Press; 1981. [Google Scholar] 194. Ember CR, Ringen EJ, Dunnington J, Pitek E. Resource stress and subsistence diversification across societies. Nat Sustain. 2020;3: 737–745. doi: 10.1038/s41893-020-0542-5 [CrossRef] [Google Scholar] 195. Bailey G, Milner N. Coastal hunter-gatherers and social evolution: marginal or central? Before Farming Archaeol Old World Hunter-Gatherers. 2002;3: 1–15. [Google Scholar] 196. Lepofsky D, Caldwell M. Indigenous marine resource management on the Northwest Coast of North America. Ecol Process. 2013;2: 12. doi: 10.1186/2192-1709-2-12 [CrossRef] [Google Scholar] 197. Trosper RL. Northwest coast indigenous institutions that supported resilience and sustainability. Ecol Econ. Elsevier; 2002;41: 329–344. doi: 10.1016/S0921-8009(02)00041-1 [CrossRef] [Google Scholar] 198. Lancelotti C, Zurro D, Whitehouse NJ, Kramer KL, Madella M, García-granero JJ, et al. Resilience of small-scale societies ‘ livelihoods : a framework for studying the transition from food gathering to food production. Ecol Soc. 2016;21: art8. doi: 10.5751/ES-08757-210408 [CrossRef] [Google Scholar] 199. Nettle D, Gibson MA, Lawson DW, Sear R. Human behavioral ecology: Current research and future prospects. Behav Ecol. 2013;24: 1031–1040. doi: 10.1093/beheco/ars222 [CrossRef] [Google Scholar] 200. Ahedo V, Caro J, Bortolini E, Zurro D, Madella M, Galán JM. Quantifying the relationship between food sharing practices and socio-ecological variables in small-scale societies: A cross-cultural multi-methodological approach. PLoS One. 2019;14: e0216302. doi: 10.1371/journal.pone.0216302 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 201. De Souza RG, Mirazón Lahr M. Climatic correlates of human subsistence: A global analysis. Appl Ecol Environ Res. 2015;13. [Google Scholar] 202. Morgan C. Is it Intensification Yet? Current Archaeological Perspectives on the Evolution of Hunter-Gatherer Economies. J Archaeol Res. 2015;23: 163–213. doi: 10.1007/s10814-014-9079-3 [CrossRef] [Google Scholar] 203. Mannino MA, Thomas KD. Depletion of a resource? The impact of prehistoric human foraging on intertidal mollusc communities and its significance for human settlement, mobility and dispersal. World Archaeol. 2002;33: 452–474. doi: 10.1080/00438240120107477 [CrossRef] [Google Scholar] 204. Wroe S, Field J, Fullagar R, Jermin LS. Megafaunal extinction in the late Quaternary and the global overkill hypothesis. Alcheringa An Australas J Palaeontol. 2004;28: 291–331. doi: 10.1080/03115510408619286 [CrossRef] [Google Scholar] 205. Erlandson JM, Rick TC. Archaeology Meets Marine Ecology: The Antiquity of Maritime Cultures and Human Impacts on Marine Fisheries and Ecosystems. Ann Rev Mar Sci. 2010;2: 231–251. doi: 10.1146/annurev.marine.010908.163749 [PubMed] [CrossRef] [Google Scholar] 206. Erlandson JM, Fitzpatrick SM. Oceans, Islands, and Coasts: Current Perspectives on the Role of the Sea in Human Prehistory. J Isl Coast Archaeol. 2006;1: 5–32. doi: 10.1080/15564890600639504 [CrossRef] [Google Scholar] 207. Williams J. Clam gardens: aboriginal mariculture on Canada’s west coast. Vancouver: New Star Books; 2006. [Google Scholar] 208. Turner N, Jones JT. Occupying the Land: Traditional Patterns of Land and Resource Ownership among First Peoples of British Columbia. 2000. 209. Flores C, Olguin L, Rebolledo S, Torres J, Broitman BR. Prehistoric coastal fisheries on the Atacama Desert (Taltal, 25°S). A Trans-Holocene record of paleoceanographic and archaeological fish remains. Quat Int. 2020; doi: 10.1016/j.quaint.2020.09.011 [CrossRef] [Google Scholar] 210. Caldwell ME, Lepofsky D, Combes G, Washington M, Welch JR, Harper JR. A Bird’s Eye View of Northern Coast Salish Intertidal Resource Management Features, Southern British Columbia, Canada. J Isl Coast Archaeol. 2012;7: 219–233. doi: 10.1080/15564894.2011.586089 [CrossRef] [Google Scholar] 211. Thompson VD, Worth JE. Dwellers by the Sea: Native American Adaptations along the Southern Coasts of Eastern North America. J Archaeol Res. 2011;19: 51–101. doi: 10.1007/s10814-010-9043-9 [CrossRef] [Google Scholar] 212. Kelly RL. Hunter-Gatherer Mobility Strategies. J Anthropol Res. 1983;39: 277–306. doi: 10.1086/jar.39.3.3629672 [CrossRef] [Google Scholar] 213. Kelly RL. The Foraging Spectrum: Diversity in Hunter-Gatherer Lifeways. Washington, D.C.: Smithsonian Institution; 1995. [Google Scholar] 214. Borrero LA, Barberena R. Hunter‐Gatherer Home Ranges and Marine Resources. Curr Anthropol. 2006;47: 855–868. doi: 10.1086/507186 [CrossRef] [Google Scholar] 215. Yesner DR, Ayres WS, Carlson DL, Davis RS, Dewar R, Hassan FA, et al. Maritime Hunter-Gatherers: Ecology and Prehistory [and Comments and Reply]. Curr Anthropol. 1980;21: 727–750. doi: 10.1086/202568 [CrossRef] [Google Scholar] 216. Arnold JE. Complex Hunter-Gatherer-Fishers of Prehistoric California: Chiefs, Specialists, and Maritime Adaptations of the Channel Islands. Am Antiq. 1992;57: 60–84. doi: 10.2307/2694835 [CrossRef] [Google Scholar] 217. Marquet PA, Santoro CM, Latorre C, Standen VG, Abades SR, Rivadeneira MM, et al. Emergence of social complexity among coastal hunter-gatherers in the Atacama Desert of northern Chile. Proc Natl Acad Sci. 2012;109: 14754–14760. doi: 10.1073/pnas.1116724109 [PMC free article] [PubMed] [CrossRef] [Google Scholar] 218. King CL, Millard AR, Gröcke DR, Standen VG, Arriaza BT, Halcrow SE. Marine resource reliance in the human populations of the Atacama Desert, northern Chile–A view from prehistory. Quat Sci Rev. 2018;182: 163–174. doi: 10.1016/j.quascirev.2017.12.009 [CrossRef] [Google Scholar] 219. Roberts A, Donald Pate F, Petruzzelli B, Carter C, Westaway MC, Santoro CM, et al. Retention of hunter–gatherer economies among maritime foragers from Caleta Vitor, northern Chile, during the late Holocene: evidence from stable carbon and nitrogen isotopic analysis of skeletal remains. J Archaeol Sci. 2013;40: 2360–2372. doi: 10.1016/j.jas.2013.01.009 [CrossRef] [Google Scholar] 220. Díaz-Zorita Bonilla M, Drucker DG, Richardin P, Silva-Pinto V, Sepúlveda M, Bocherens H. Marine food consumption in coastal northern Chilean (Atacama Desert) populations during the Formative Period: Implications of isotopic evidence (C, N, S) for the Neolithic process in south central Andes. J Archaeol Sci Reports. 2016;6: 768–776. doi: 10.1016/j.jasrep.2016.01.026 [CrossRef] [Google Scholar] Articles from PLoS ONE are provided here courtesy of Public Library of Science What are the types of subsistence strategies?In Cultural Anthropology, subsistence strategies are the ways that people obtain food from their environment. There are five basic subsistence strategies: foraging, pastoralism, horticulture, agriculture, and industrialism.
Which strategy of subsistence has been the most common throughout human history?Before 11 000 years ago, all human societies foraged for food, but more recently plant-based agriculture or animal husbandry became the dominant subsistence strategy for most human groups.
What is the most common mode of subsistence among early state societies?The earliest subsistence strategy that was used is foraging. This strategy involves hunting for meat and gathering wild plants. There is no cultivation of plants or animals, leading to a nomadic lifestyle and small community groups.
What are the 4 main methods of subsistence for humans?Beyond this basic division, anthropologists recognize four general types of food system known as modes of subsistence. The four modes of subsistence are foraging, pastoralism, horticulture, and agriculture.
|