research paper on military history

An official website of the United States government

Official websites use .gov A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

• Publications
• Account settings
• Advanced Search
• Journal List

Rise of the war machines: Charting the evolution of military technologies from the Neolithic to the Industrial Revolution

Peter turchin, daniel hoyer, andrey korotayev, nikolay kradin, sergey nefedov, gary feinman, jill levine, jenny reddish, enrico cioni, chelsea thorpe, james s bennett, pieter francois, harvey whitehouse.

• Author information
• Article notes
• Copyright and License information

Competing Interests: The authors have declared that no competing interests exist.

* E-mail: [email protected]

Received 2021 May 28; Accepted 2021 Aug 18; Collection date 2021.

This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

What have been the causes and consequences of technological evolution in world history? In particular, what propels innovation and diffusion of military technologies, details of which are comparatively well preserved and which are often seen as drivers of broad socio-cultural processes? Here we analyze the evolution of key military technologies in a sample of pre-industrial societies world-wide covering almost 10,000 years of history using Seshat : Global History Databank . We empirically test previously speculative theories that proposed world population size, connectivity between geographical areas of innovation and adoption, and critical enabling technological advances, such as iron metallurgy and horse riding, as central drivers of military technological evolution. We find that all of these factors are strong predictors of change in military technology, whereas state-level factors such as polity population, territorial size, or governance sophistication play no major role. We discuss how our approach can be extended to explore technological change more generally, and how our results carry important ramifications for understanding major drivers of evolution of social complexity.

Introduction

From simple sharpened stone projectiles in the Paleolithic to the weapons of mass destruction in the modern world, what have been the main factors driving the evolution of military technology? Many have argued that the evolution of military technologies is just one aspect of a much broader pattern of technological evolution driven by increasing size and interconnectedness among human societies [ 1 – 3 ]. Several cultural evolutionary theories, conversely, highlight military technologies as a special case, arguing that steep improvements in both offensive and defensive capabilities of technologies along with accompanying tactical and organizational innovations resulted in “Military Revolutions” (note the plural), which in turn had major ramifications on the rise and, of particular concern here, the spread of state formations globally [ 4 – 8 ] and the evolution of religion and other cultural phenomena [ 9 , 10 ]. But the evolutionary mechanisms underlying general technological innovation, adoption, and transmission (especially in pre-industrial societies) are not well understood. Moreover, available theories have drawn on evidence that is limited both in geographical scope and temporal depth and deployed in ways that are subject to selection bias. Here we explore a variety of factors that previous scholarship suggests may have played a role in the evolution of military technologies by systematically quantifying the effects of those factors for thousands of years of world history.

Earlier efforts to quantify levels of technological complexity in eastern and western ends of Eurasia [ 11 , 12 ] have been criticized for being unduly subjective [ 13 ], especially when it comes to measuring rates of innovation in military technology, and are obviously limited in spatial coverage. Here we propose an alternative methodology for quantifying technological evolution and expand the geographic scope from just these two broad regions to 35 “Natural Geographic Areas” across all ten major world regions, using Seshat : Global History Databank , a major resource for studying patterns of sociocultural evolution in world history (see Materials and Methods below).

This article has two related goals. The first is to establish broad spatio-temporal patterns in the evolution of military technologies in pre-industrial societies. By technological evolution we mean here the dynamics of uptake (and possible loss) of technologies used by societies at significant scale (rather than simply whether the technology was known at all), regardless of how that society came to acquire that technology (indigenous innovation or adoption from another culture). For those interested in the study of technological evolution in general, focusing specifically on military technologies in pre-industrial societies has many practical benefits. Warfare was one of the most intensive activities of human societies, leaving abundant traces in the archaeological and historical record.

The second goal is to explore why these important military technologies developed or were adopted in the places, at the times, and as part of the technological packages as we observe in the historical and archaeological record. There have been several theoretical conjectures (discussed below) about the main causal drivers of technological innovation that we test. As our approach will show, the pattern of military technological evolution shows great variation in time and space, with different regions assuming a leading role in innovation at different moments in time.

Delineating the possible causes and observed consequences of changes in levels of military technologies will have far-reaching implications for understanding the evolution of technology broadly. To encourage further progress towards that ultimate goal, we present here a detailed methodology for testing theories about technological change in human history. This paper serves as a crucial step along this path.

Theoretical background

Here, we review several competing theoretical perspectives on the evolution of technologies offered in the past. Technological change is one of the fundamental drivers in social and cultural evolution and of long-term economic growth [ 14 – 17 ]. Many have pointed to technology’s ramifying effects on warfare, state formation, and the development of information processing systems [ 1 , 18 – 22 ]. Military technologies and their widespread application in particular have been shown to foment rises in social complexity and to spur related ideological developments [ 23 – 27 ]. But what processes are responsible for the evolution–the development, spread, and cumulative adoption–of military technology globally and across time?

Following this link between military technologies and socio-cultural development, we might expect to find a positive feedback between technological innovation and population growth at the global scale [ 2 , 28 – 32 ] see also [ 33 – 36 ]. Indeed, a well-known and much discussed theory proposed by economist Michael Kremer and expanded by others suggests exactly this causal link [ 2 ]. According to Kremer, “high population spurs technological change because it increases the number of potential inventors.” Kremer notes that "this implication flows naturally from the nonrivalry of technology… The cost of inventing a new technology is independent of the number of people who use it. Thus, holding constant the share of resources devoted to research, an increase in population leads to an increase in technological change. Thus, in a larger population there will be proportionally more people lucky or smart enough to come up with new ideas" [ 2 : 685]. This innovation, in turn, can spur further population growth, creating a positive feedback loop between technological and population growth; for instance, the proliferation of iron axes facilitated the clearing of agricultural land from forests [ 37 ], while the iron ploughshare improved the quality of plowing allowing for increased productivity [ 38 ] and, hence, larger populations to develop further innovations. Note that what is described by Kremer is virtually identical with what David Christian calls “collective learning” [ 39 ].

This process was expressed mathematically by Taagepera, Kremer, Podlazov and Tsirel in the following way:

This equation states that the technological growth rate at a given moment in time ( dT / dt ) is proportional to the global population, N (the larger the population, the larger the number of potential inventors) and to the current technological level, T . The second factor is included in the model to reflect the assumption that the wider the existing technological base, the greater the number of inventions that can be made on its basis. This model explicitly refers to global population level, rather than regional or localized populations of specific societies. To account both for the effect of global population size and the existing stock of technology, the Taagepera-Kremer model assumes that the rate of technology growth is proportional to the product of these two quantities. Taagepera and Kremer did not test this hypothesis empirically in a direct way. Empirical tests of this hypothesis performed by other researchers, however, have found support [ 34 , 40 ]. Note that Kremer observed that these new technologies would, in turn, likely generate population growth, suggesting a positive feedback between technological innovation and population. Here, however, we are concerned only with the effect of population on the evolution of technology, rather than the reverse.

A limitation of such population-focused theories, however, is the assumption that world population can be treated as having been an integrated information-exchanging system for many centuries, if not millennia. To address this problem, world-systems analysts have advanced an additional cluster of hypotheses. Chase-Dunn and Hall, for example, distinguish four types of networks of world-system communications: bulk good networks, political-military networks, prestige good networks, and information networks (IN) [ 41 ]. Korotayev et al. [ 42 – 44 ] explicitly focus on INs as technological innovation diffusion networks, proposing that a systematic diffusion of technological innovations within a certain set of societies is a sufficient condition to consider them a “world-system”. Thus, an as-yet unexplored synthesis of these ideas is that, while population may be one factor in the pace and location of technological evolution, membership in such an information diffusion network may play an additional role in facilitating the exchange of ideas and propensity for wide-spread adoption of new technologies.

One important advantage of the population-driven model advocated by Kremer, Taagepera, and others is that it explicitly includes the effect of the existing stock of technologies on technological growth rate. The greater the existing stock, the greater number of new technologies the model expects to be developed in the next time period. Although this is only one, relatively straightforward, way to model the impact of the existing technology stock, there is substantial historical evidence to make it a strong contender to be tested empirically. For example, the improvement of metallurgy and metal processing led not only to the emergence of new tools such as iron ploughs, but also to the proliferation of various types of weapons—starting with knives, daggers, swords, battle axes, up to the appearance of rifles and artillery. Nevertheless, the model assumes that the means and knowledge to adapt and improve upon existing technologies are readily accessible as well as the organizational capacity to deploy these technologies at large scales, which are open questions requiring further scrutiny.

Further, once a military technology had proven advantageous in inter-state competition, there arose an existential pressure on nearby societies to adopt that technology as well, so as not to be left behind. This sort of mimetic diffusion has been observed with respect to key technologies such as horse-mounted warfare that spread initially among nomadic confederations and nearby agrarian societies located along the central Eurasian Steppe [ 45 – 48 ]. Indeed, the domestication of the horse and its use in the civil and military sphere–including both the material components of horse-mounted archery as well as the tactical and organizational means to wield these weapons–appear to be of particular importance in the evolution of technologies and social complexity during the pre-industrial era, improving transportation, agriculture, and military capacities alike [ 47 ]. Further, the creation of new and more lethal weapons in one society could force people in their “strike zone” [ 27 ] to invent more sophisticated defenses while also often adopting the offensive technology themselves, prompting further technological advances. Following the invention of increasingly powerful, armor-piercing projectiles from bows and crossbows, for instance, we tend to see the means of protection improved as well to include chain mail, scaled armor, and plate armor.

Similarly, some work suggests that location is a critical factor in this process, as societies on the periphery, or semi-periphery [ 41 ], of larger, more complex imperial states will tend to be hotbeds of innovation, as they have both the incentive to increase (typically military) capability to compete with regional powers as well as the requisite flexibility to explore more radical innovation by being removed from the institutionalized practices and path-dependencies experienced by the larger societies “locked in” to the tools and habits that won them their hegemony [ 41 , 49 , 50 ].

Overall, previous theoretical work suggests that the evolution of military technologies depends on the total number of potential innovators involved in this process, the connectedness of distinct centers of innovation as well as of spheres of inter-state competition, and on the already existing stock of technologies, especially such fundamental developments as metal processing and transportation. In Materials and Methods below we discuss how we operationalize an empirical test of these hypotheses.

Materials and methods

A general approach to quantifying the evolution of pre-industrial societies.

This article follows the general philosophy and procedures that have been developed by the Seshat : Global History Databank project [ 51 – 54 ]. The Seshat Databank stores large volumes of historical and archaeological data on a growing number of variables for past polities going back to the late Neolithic. Supplementary Information ( SI ) contains a detailed description of the core methods and workflows underpinning the Seshat project, including how we incorporate differing levels of uncertainty and disagreement and data quality procedures involving experts and research assistants. We make the data used for the analyses presented here available online through a DataBrowser site ( seshatdatabank.info/databrowser ) and we encourage scholars to make use of and to augment our dataset.

The principal unit for data collection and analysis is a polity, defined as any independent political unit ranging from autonomous villages (local communities) through simple and complex chiefdoms to states and empires, regardless of degree of centralization [ 51 , 52 ]. Our sample of historic polities was developed using a stratified sample of the globe using the concept of ‘Natural Geographic Area’ (NGA). An NGA is a fixed spatial location of roughly 100 x 100 km delimited by naturally occurring geographical features (for example, a river basin, a coastal plain, a mountain valley, or an island). All polities that occupied the NGA, or part thereof, at a century mark (e.g. 200 CE), are included in our sample. This strategy avoids oversampling (redundantly repeating information across time points) while still capturing meaningful changes in the variables of interest. Although this granularity is relatively coarse, it is suitable for uncovering macro-level patterns in societal dynamics and exploring pathways of cultural evolution [ 24 , 53 ]. The data used in the analyses presented here come from 373 historic polities covering 35 NGAs.

Aggregation of military technology data into “Warfare Characteristics”

We quantified the sophistication of war-making capacity by encoding 46 binary variables indicating the presence or absence of different military technologies by a polity. These variables were aggregated into six general categories, termed Warfare Characteristics (WCs): Metals used in producing weapons and armor, the variety of Projectiles and hand-held Weapons, the sophistication of Armor, the use of transport Animals, and different kinds of Defensive Fortifications. Finally these WCs were aggregated into a composite, temporal MilTech variable for each NGA. See SI for details of the aggregation.

Of the six WCs, two (Metal and Animal) have a much broader area of application than specifically warfare. In some analyses below we investigate another measure (CoreMil) that focuses more narrowly on the sophistication of core military technologies by aggregating only the Projectiles, Weapon, Armor, and Defense WCs. As described below, we explore the impact of the spread of Iron and Cavalry in particular. Because Iron and Cavalry are correlated with the Metals and Animals WCs, analyzing CoreMil allows us to disentangle any potentially spurious effects of these WCs on overall military technology.

Hypotheses to be tested: Defining predictor variables

Our review in Theoretical Background suggested that the evolution of military technologies may be a function of the total number of potential innovators, the connectedness of innovation/adoption centers, and/or the existing stocks of technology. We measure these various potential explanatory factors in the following ways:

Following Taagepera and Kremer, we proxy the number of potential innovators with the world population (WorldPop, defined as log(10) of the global population at time t ). We take data on the dynamics of world population during the Holocene from [ 55 ].

Connectedness is a harder variable to quantify. Here we build on the concept of IN used by Chase-Dunn and Hall and other world-systems theorists [ 34 , 43 ] who define the extent of any particular IN as the zone within which spatially and culturally distinct regions exchange information, so that technological innovations made in one society diffuse relatively rapidly (on the time scale of centuries) to all other societies within the system than to societies that may be close (spatially and culturally) but fall outside of the IN. As an example, the contacts between Western and Eastern Eurasia (mediated via Central Asia) in the third and especially the second millennia BCE led to the spread of multiple technological innovations between the western and eastern parts of Eurasia: wheat, cattle, horses, bronze metallurgy, wheeled chariots, among others [ 44 , 56 ]. Here, we constructed a predictor variable proxying the Centrality of each region within the evolving (eventually global) IN by calculating the distance between each of our NGAs and the system of Silk Routes that connected East and West Eurasia for the majority of the period under study [ 57 – 60 ]. Our measure of Centrality is the inverse of the distance between an NGA and the nearest node on the Silk Route (see Fig 1 ; and SI for details).

Fig 1. Location of nodes on Silk Routes used in quantifying Centrality (red) along with NGA locations (black).

In addition to Centrality within the IN, we capture two additional kinds of connectivity, namely the possible influence of spatial proximity (Space) as well as cultural affinity (Phylogeny). These terms not only allow us to control for possible autocorrelations and phylogenetic effects in our response variable (see Dynamic Regression Analysis below), but can also carry important information about processes influencing the evolution of military technologies. In particular, Space captures the process by which technological innovations may travel between geographically proximate societies–separately from the possible mediating influence of an expanding IN described above–measuring the likelihood that neighboring regions will share similar levels of military technology. Phylogeny focuses on the cultural similarity between polities, however spatially close, proxied by the relatedness of their dominant languages.

Another possible factor in the evolution of technology identified in the theoretical review is the effect of the current technology stock. We measure this in two ways. First, we model MilTech as a temporal autoregressive process, in which past values of MilTech affect its future values (for the details of the statistical model, see the next section). Second, we focus on the potential effects of two specific fundamental technologies: horse-riding and iron smelting.

According to the Cavalry Revolution theory, the invention of effective horse-riding in the Pontic-Caspian steppes, combined with powerful recurved bows and iron-tipped arrows, triggered a process of military evolution that spread from the steppes south to the belt of farming societies over several centuries throughout the first millennia BCE and CE [ 8 , 47 , 61 ]. Specifically, the threat of nomadic warriors armed with this advanced (for the period) military technology spurred the development of counter-measures designed to mitigate the cavalry advantage, while also producing an incentive to adopt horse-riding and effective accompanying combat tactics in areas further and further away from the location of their initial invention within the Steppe. The history of the military use of the horse went through several stages: the use of the chariot, the development of riding, the formation of light auxiliary cavalry, the development of nomadic riding, the appearance of the hard saddle, armored cataphracts, stirrups and, finally, heavy cavalry—a major branch of troops across Afro-Eurasian societies between c. 550 and 1400 CE [ 62 ]. As a result, effective horse-riding had far-reaching consequences for the evolution of military technologies, and specifically armor, projectiles such as crossbows, and fortifications. We use the data from [ 63 ] to encode the Cavalry variable (see Fig 2 ).

Fig 2. Spread of horse-mounted Cavalry.

Data from [ 63 ].

The effect of Iron is similarly widespread. Multiple authors [ 64 – 66 ] have suggested that the availability of iron had a major impact on the evolution of technologies, as this strong and malleable material served as an input for a host of important technologies, military and otherwise, throughout the period under investigation here. We use data from [ 67 ] to encode the Iron variable (see Fig 3 ).

Fig 3. Spread of iron metallurgy.

Data from [ 67 ].

Note that these two variables, Cavalry and Iron, are highly correlated with each other (compare Figs 2 and 3 ) and it may be difficult to estimate their effects separately (the problem of collinearity). To address this potential issue we created a synthetic variable, IronCav, that combines the two effects (by adding Cavalry and Iron together). IronCav, thus, takes the maximum value for societies with both mounted warfare and iron weapons, intermediate value for societies having one characteristic and not the other, and minimum for societies with neither characteristic. We explored with dynamic regressions whether IronCav turns out to be a better predictor than either of its constituent variables, reported below.

In addition to the theoretically-motivated predictors–WorldPop, Centrality, Iron, and Cavalry, along with our autocorrelation terms Space and Phylogeny–we explore other potential polity- and NGA-specific predictors to proxy interesting subsidiary hypotheses, as explained below. These measures are taken from previously published analyses using Seshat data [ 68 ] and enable us to reduce the potential “hidden variable” (or omitted variable bias) problem, which arises when analysis implicates X as a causal factor for Y , while in reality the true cause could be Z , with which X is closely correlated [ 69 , 70 ]. The additional predictor variables include the following:

Social scale (Scale) represents the first principal component (PC) of the Seshat variables polity population , polity territory , population of the largest settlement , and the number of hierarchical levels . The hypothesis here is that larger and more complexly organized and productive societies (in both population, territory) will have more resources to both generate new inventions and to implement them, or adopt them from elsewhere, especially the costly ones like sophisticated siege engines or elaborate fortifications. This measure also reflects having larger shares of the population not mainly engaged in primary production, proxied by the population of the largest settlement [ 71 , 72 ]. Further, more stratified and administratively complex societies–measured by the number of levels in administrative, military, and settlement hierarchies (combined here as one measure of hierarchical levels–see SI )–are hypothesized to be better equipped to implement useful technologies along with developing or adopting effective tactical and organizational models at scale. Thus, by this logic, increases in military technology should occur preferentially in larger scale societies. Previous analysis [ 53 ] reveals that these four dimensions are highly correlated within the Seshat sample and so represent an effective cross-cultural measure of societal scale to explore this hypothesis.

SocSoph (“social sophistication”) represents the first PC of the Seshat variables governance , infrastructure , information systems , and money . This measure likewise derives from previous analysis of the dimensions of social complexity [ 53 ], capturing the important non-scale institutional and informational aspects. The hypothesis here is that societies with more sophisticated, pre-existing mechanisms for the exchange and implementation of ideas will generate and/or adopt innovations into widespread use at a faster pace.

Agricultural productivity (Agri) is the estimated yield of different regions, measured as tonnes per hectare of the major carbohydrate source consumed in each of our NGAs. These data are taken from the analyses in [ 73 ]. The term is included here to test the possibility that productivity affects the amount of resources that are available for technological advances.

Social scale and productivity, thus, give us two complementary views of the resource base that may drive the evolution of technology. Agri tracks the underlying material resource base in a given geographical region (our NGAs) needed to support development, including technological evolution. Social scale, on the other hand, is a measure of the territorial and population size of specific historical polities. Larger polities can gather resources from a large territory, including the human energy from large populations, even where agricultural productivity is low. Separately SocSoph represents the sophistication of infrastructure and exchange media that could conceivably facilitate the flow of ideas from invention (whether within or outside of the society) to widespread adoption.

Statistical analysis

In addition to standard correlational statistical analyses of our response and predictor variables, we used a general non-linear dynamic regression model to investigate factors affecting the evolution of military technology. This dynamic regression analysis distinguishes correlation from causation by estimating the influence potential causal factors at a previous time have on the response variable at a later time (known generally as Wiener-Granger causality [ 74 , 75 ]). While an improvement over ‘static’ correlations, where causal direction remains ambiguous, this method is, nevertheless, insufficient for making absolute claims of causality. Further scrutiny will be required to provide additional support for the provisional causal interpretations suggested below.

Our model takes the following form [ 70 ]:

Here Y i , t is the response variable (MilTech) for location (NGA) i at time t . We construct a spatio-temporal series for Seshat response and predictor variables by following polities (or quasi-polities, such as archaeologically attested cultures) that occupied a specific NGA at each century mark during the sampled period. Thus, the time step in the analysis is100 years.

On the right-hand side, a is the regression constant (intercept). The next term captures the influences of past history (“autoregressive terms”), with τ = 1, 2, … indexing time-lagged values of Y (as time is measured in centuries, Y i , t – 1 refers to the value of MilTech 100 years before t ).

The third term represents potential effects resulting from geographic diffusion using our Space term. We used a negative-exponential form to relate the distance between location i and location j , δ i , j , to the influence of j on i . Unlike a linear kernel, the negative-exponential does not become negative at very large δ i , j , instead approaching 0 smoothly. The third term, thus, is a weighted average of the response variable values in the vicinity of location i at the previous time step, with weights falling off to 0 as distance from i increases. Parameter d measures how steeply the influence falls with distance, and parameter c is a regression coefficient measuring the importance of geographic diffusion. For an overview of potential effects resulting from geographic diffusion, see [ 69 , 76 ]; for a description of how we avoided the problem of endogeneity, see [ 70 ].

The fourth term detects autocorrelations due to any shared cultural history at location i with other regions j using our Phylogeny variable. Here w represents the weight applied to the phylogenetic (linguistic) distance between locations (set to 1 if locations i and j share the same language, 0.5 if they are in the same linguistic genus, and 0.25 if they are in the same linguistic family). Linguistic genera and families were taken from The World Atlas of Language Structures and Glottolog [ 77 ].

The next term on the right-hand side represents the effects of the main predictor variables X k [ 70 ]; g k are regression coefficients. These variables (described in the previous section) are of primary interest because they enable us to test various theories about the evolution of MilTech against each other. Finally, ε i , t is the error term. We also include quadratic versions of these terms at a time lag (not shown) in order to explore non-linear responses to response and predictor factors.

Model selection (choosing which terms to include in the regression model) was accomplished by exhaustive search: regressing the response variable on all possible linear combinations of predictor variables. The degree of fit was quantified by the Akaike Information Criterion (AIC). Standard diagnostic tests were performed for the best-fitting models [ 70 ].

Missing values, estimate uncertainty, and expert disagreement in the predictors were dealt with by multiple imputation [ 78 , 79 ]. The response variable, MilTech, however, was not imputed as that can result in biased estimates [ 76 ]. For details of the multiple imputation procedure see SI . Because diagnostic tests indicated that the distribution of residuals are not gaussian, we used nonparametric bootstrap to estimate the P -values associated with various regression terms (see the SI for details). Additional robustness checks are similarly detailed in the SI .

Spatio-temporal patterns

We first examined the frequency distributions of the variables of interest and the cross-correlations between WCs, overall MilTech, which combines all WCs, as well as calendar Time and the various aspects of social complexity and productivity. As expected, we find that all WCs are closely correlated with each other and with the overall MilTech variables. Plotting MilTech as a function of time for each NGA ( Fig 4 ), we observe that there is a general upward trend, as expected. However, there are also periods when some technologies are lost, for a time. Most importantly, there is a great amount of variation between different geographic regions in the timing of MilTech increases. Interestingly, all WCs are more strongly correlated with the two dimensions of social complexity specified here–Scale and SocSoph–rather than with Time, suggesting that key drivers of MilTech evolution go beyond merely the additive nature of technology through the ‘march of time’. The nature of any causality between complexity and MilTech is discussed below.

MilTech trajectories in Seshat NGAs, divided by major world region: (A) Europe and Africa; (B) Western Asia; (C) East and SE Asia; (D) Americas and Oceania.

We next focus on “technology leaders”, NGAs that at some point in their history had the highest value of MilTech available at the time. Fig 5 shows them, roughly in the order that they achieved world leadership (note that this order is also affected by how far back in the past we have data). The hot spot of technological development, through either innovation or adoption, appears to roughly coincide with the “Imperial Belt” of the Old World, located just south of the Great Eurasian Steppe (and in places, impinging into it, as in Sogdiana), which can be seen by the location of the ‘leader’ NGAs (mapped in Fig 5 ).

Fig 5. “Technological leaders”: NGAs that at some point achieved the highest MilTech score available at that time.

This same territory also of course corresponds roughly to the path of the overland silk routes used in our analyses ( Fig 1 ). We return to this pattern below. Overall, the pattern is that most of the leading regions exhibit an increase in their overall MilTech levels roughly together and at a fairly regular, almost linear pace (after the 4 th millennium BCE), with late comers accelerating at various points to merge with leaders. This is seen clearly in this graph on the example of Sogdiana, but it is a general pattern discernible in the regional examples ( Fig 4 ).

We explored the “similarity” between NGAs by calculating the number of MilTech variables in each NGA shared with other NGAs at each time-step. As explained in SI (see Similarity Analysis and S1 Fig in S1 File ) we trace how NGAs join the expanding Eurasian (eventually global) IN by noting the time when they achieve a similarity index of 10, that is, when they share 10 or more specific MilTech variables with one or more other NGAs. As the histograms in S1 Fig in S1 File show, the first NGAs that achieve this threshold of similarity appear between 3000 and 2500 BCE. As time progresses, more and more NGAs cross this threshold. Fig 6 maps the expansion of this IN–initially restricted to central Eurasia but growing eventually into a global network–by color coding the date when the NGA cross this threshold. Thus, the similarity analysis reveals that different regions not only saw rapid increases in their overall level of MilTech, but these areas came increasingly to share specific technologies. A plausible interpretation for this pattern is that, as the IN expands, each new region accelerates its development of MilTech to join the level achieved by the network leaders, until, eventually, all regions in diverse areas around the globe adopt similar ‘MilTech packages’. Future work is needed to disentangle occasions where these late-comer regions adopt or adapt existing technologies from cases where ‘leader’ societies simply take over others, imposing their technologies (along with a host of other socio-political and cultural traits) onto this new regions.

Fig 6. Results of similarity analysis.

NGAs are binned into 6 categories according to the earliest time they share 10 or more MilTech variables with another NGA, displayed by color: dark red = 2500 BCE or before; orange = between 2500 and 1500 BCE; yellow = between 1500 and 500 BCE; green = between 500 BCE and 500 CE; blue = after 500 CE; grey = did not display any similarities during our sample period. Unfilled red circles indicate Silk Route Nodes as in Fig 3 .

Dynamic regression results

The best fitting model from our general dynamic regression analysis is shown in Table 1 .

Table 1. Regression results for the best (lowest AIC) regression model.

Estimate shows the standardized regression coefficients, which provide a direct measure of relative effects by the lagged predictors on the response variable. Thus, MilTech here represents the linear autoregressive term, AR(1). The column “t value” lists t-statistics, a measure of statistical significance of regression terms associated with various predictors. Pr(>|t|) is the statistical significance for regression assuming the Normal distribution of residuals, while Bootstrap estimated P is the result of nonparametric bootstrap that does not make this assumption.

Our analysis identifies the following variables as having the strongest causal influence on MilTech:

Autocatalytic effects (the value of MilTech in the previous time step).

Global population size (WorldPop).

Connection to an expanding (eventually global) Information Network (Centrality).

Spread of Iron+Cavalry (IronCav), revealing both the importance of prior technology stock on continued technological evolution as well as the incentive that these advances placed on societies within connected information and competition spheres to adopt or develop additional technologies in response.

Cultural similarity (Phylogeny), revealing that polities linguistically similar to polities with high MilTech are more likely to have high MilTech themselves. This effect could be a result of either common inheritance or easier diffusion of technology between culturally similar polities, or, most likely, both.

Productivity of the resource base (Agri).

Investigation of the effects of Cavalry and Iron as predictor variables indicate that either, separately, has a statistically significant effect of similar strength on the evolution of MilTech. The synthetic variable, IronCav, however, is a better predictor than either of its constituents. For this reason, the results here are reported for IronCav only.

We estimated how location with respect to the system of Silk Routes affects the evolution of MilTech in each region. Our measure of Centrality (inverse distance to the nearest Silk Route node) finds strong empirical support, although we ran analyses using alternate methods of proxying this type of spatial effect (see SI for details). Overall our best model predicts the level of military technology with regression coefficient of determination ( R 2 ) of 0.96. While some of this high predictability is a result of strong temporal autocorrelation, rerunning the regression omitting all autocorrelation terms nevertheless yields an R 2 of 0.72. Thus, more than 70% of the variation in MilTech is explained by WorldPop, Centrality, IronCav, and Agri.

We performed several supplemental analyses and robustness checks to detect any biases in our results. Several of these checks are discussed below and detailed in the SI .

Table 2 shows a comparison between the best fitting model and other models with Δ AIC ≤ 2. Strong effects are detected in these alternative models for all terms in the best model including Agri, which, though its standardized coefficient is the smallest, remains statistically significant at the conventional P < 0.05 level. However, additional robustness tests using multiple datasets built by random sampling from among the different WCs comprising the MilTech variable indicate that Agri is not always supported (see SI for details). Neither measures of social complexity, Scale and SocSoph, appear to have a consistent significant positive effect on MilTech evolution (they show up in several of the alternative models, but with small t -values and negative signs for Scale).

Table 2. Alternative model selection results.

The table shows t-statistics associated with each of the predictors (column headings) for the best models with ΔAIC (the AIC difference with respect to the best model) less than 2. The best model by AIC is included as the top row. An empty entry indicates that the term associated with this predictor is not included in the model.

Further checks indicate that these results are robust to the inclusion of additional spatial and temporal autocorrelation effects: Neither geographic diffusion (Space) nor higher temporal lags ( τ = 2 centuries or greater) are significant. In addition, as we discussed in Materials and Methods , because our measure of MilTech includes the Metals and Animal WCs, which might confound the effect of IronCav due to a potential circularity, we re-ran the analysis using CoreMil, our measure of military technologies that does not include these WCs. This analysis yields essentially identical results (see SI ), thus suggesting that the effect of IronCav is not spurious.

What is remarkable is that neither Scale nor SocSoph variables, which characterize polities, have any detectable effect on the level of MilTech. Overall, these results suggest that MilTech evolves almost entirely as an exogenous variable: it is little, if at all, affected by such polity characteristics as the population, territory size, the sophistication of information systems or administrative institutions, or provision of infrastructure and public goods.

As noted, our dynamic regression approach cannot offer a definitive demonstration that the factors in the best model are the central causal forces driving the evolution of military technologies. These variables may simply be highly correlated with the ‘true’ causal factors, not included in our analyses, or the causal link may be indirect, as these factors, as well as MilTech, could be caused separately by additional factors whose effects were felt at different time-scales. We explored such a possible ‘hidden variable bias’ as much as possible through supplemental analyses of several variables for which we had reliable information. As our findings remain robust to various tests, we provisionally conclude that these results offer appealing and parsimonious causal explanation for the long-run and global evolution of military technologies. Future research will need to scrutinize whether these results hold up to the inclusion of additional factors and exploration using alternate statistical methods or mechanistic models perhaps using agent-based modelling [ 24 , 27 , 80 ].

Our goals were to investigate the global spatio-temporal evolution of key pre-industrial military technologies to illuminate the major forces driving the evolution of these critical tools, whether by innovation, adoption and adaption, or a combination of these processes. Further, our approach to testing theoretically-informed hypotheses against a broad and diverse set of empirical historical data taken from Seshat : Global History Databank serves as an example of how more general patterns of technological evolution can be explored in future research, as well as more fine-grained analyses seeking to distinguish these different processes or explore the pathways taken by individual regions or societies. Here we surveyed various causal hypotheses, which together suggested that the evolution of military technology would be a function of some combination of global population size, connectedness to information exchange networks, involvement in inter-state competition networks, and prior histories of technological innovation and adoption (especially major breakthroughs such as iron metallurgy and horse riding), along with, perhaps, various properties of polities and their resource base. We set out to test these theories empirically against the evidence from world history, using a stratified sample of polities in Seshat , dating from the Neolithic to the Industrial Revolutions.

While we found some empirical support for each of these hypotheses, no one theory alone accounted for the observed dynamics of military technology as well as a combination of the factors suggested by these various proposals. Our results not only explain why these theories have found support in previous studies, but also why a general understanding of the evolution of military technology has proven elusive. Our robust historical sample and extensive dynamic analyses allowed us to compare and combine elements of different theories proposed as critical drivers of military technology. Specifically, we found that global population size is a strong predictor of the subsequent levels of MilTech. While this result supports the Kremer-Taagepera model, it does not rule out other possible causal explanations based on additional variables, which, while correlated with global world population, could turn out to be a better predictor of MilTech. One such addition in future work could be to distinguish societies by their general affluence or social mobility [ 81 ], rather than treating populations as indistinguishable, which may play such a causal role driving both population increases and technological evolution.

Our analysis found that stock of prior technological innovations played an important role in the observed levels of military technology, not only from the autoregressive terms (again, supporting the Kremer-Taagepera model) but critically because the combination of iron metallurgy and horse riding had a particularly strong effect on innovation and adoption of militarily technologies in the periods under investigation here.

Importantly, we found that location within the central Eurasian IN was also a strong predictor of our response measure, in line with the insights of World Systems and cultural evolutionary theorists. This result supports the impact of being connected to other major centers of development and innovation, as well as being incorporated into spheres of inter-state competition.

However, it is noteworthy that geographic proximity between NGAs itself (proxied here by our Space measure) does not appear to be a strong predictor of the evolution of military technologies, contrary to what might be expected from certain cultural evolution theories and ideas of mimetic diffusion. This underscores the significance of iron and cavalry diffusion in particular, which have a strong effect on subsequent levels of MilTech, supporting previous work highlighting the unique role of the nomadic pastoralists of the Eurasian Steppe, early adopters of mounted archery tactics, in driving not only technological innovation among nearby agrarian populations, but in driving the expansion of social complexity and, relatedly, technological evolution throughout Afro-Eurasia [ 8 , 24 , 26 , 27 , 45 – 48 , 82 ]. The development of iron-smelting, as an input material for so many valuable weapons, appears to play a similarly crucial role [ 64 – 67 ]. These findings suggest that iron and cavalry were particularly critical technologies that conferred an important enough advantage that they fomented widespread adoption as well as sparked ‘arms races’ among competitors that included a host of other, related technologies as discussed above, which would explain the observed patterns.

This interpretation gains further support from our similarity analysis. Our main result indicate that the overall level of MilTech–measured with our aggregate MilTech score–generally rose over time (with some losses, noted above), with more and more regions coming to exhibit the same level of MilTech over time. Our similarity analysis unpacks this finding, demonstrating that not only did regions increasingly exhibit the same overall MilTech score, but they also came to share the same ‘packages’ of specific military technologies. Further, as expected, the regions with the highest combined similarity scores followed the same pattern as seen in the Centrality measure, as the NGAs closest to a Silk Route node both appeared as sharing MilTech variables with other NGAs earlier and continued to display similar MilTech packages with other NGAs that joined the IN over time, resulting in their larger combined scores (see Fig 4 ).

An interesting and somewhat surprising finding is that the properties of polities, including such seemingly important characteristics as their scale (population and territory) and sophistication (e.g., information systems), have no significant impact on the evolution of military technologies wielded by the polity (with the partial exception of Phylogeny, discussed below). We expected both scale (Scale) and non-scale (SocSoph) aspects of social complexity to play a significant role in these processes, due to an increased availability of populations and resources to put towards technological development as well as how developments in organizational and informational-exchange capacities could facilitate the adoption and adaption of existing technologies from elsewhere. However, these terms display no significant effect on subsequent levels of MilTech, suggesting that the level of technology characterizing a particular polity (whether invented or adopted) depends not on the polity’s characteristics, but rather on the characteristics of the inter-polity informational and competitive interaction spheres to which it belongs, along with the other factors identified above. The Arabian Peninsula, for example, despite being relatively low-scale in the early first millennium CE, adopts much of the ‘military package’ seen in other parts of Eurasia around 300 CE (see Fig 4 and the Similarity Analysis in the SI ), as it became increasingly incorporated into Silk Route trade connections via the Persian and Roman imperial systems, before becoming its own seat of imperial power with the rise of Islam a few centuries later.

The only polity-related term that is included in the best regression model is Phylogeny, which can reflect an operation of one of two (or both) processes: inheritance of technological sophistication from a “common ancestor” (for example, Italy and France inheriting technologies from the Roman Empire), or easier spread of innovations between culturally similar countries (such as between Romance-speaking Italy and France, or between Arabic-speaking Egypt and Mesopotamia). The latter process likely reflects the greater likelihood that an innovation developed in one polity will be more compatible with existing institutional, social, cultural, and economic systems of a culturally similar polity than those of a more distant one [ 83 ]. One major component of this effect might be that military technologies require specific tactical and organizational apparatus to wield effectively. Cultural similarity then could not only facilitate exchange of information about a new, useful technology across societies, but facilitate the spread of knowledge of and increase the ability to acquire these more ephemeral aspects accompanying the material components of this new technology. Alternatively, linguistically similar polities might have engaged in more frequent and intense competition, which could lead to a similar impact (likely correlated strongly with the IronCav and Centrality effects) on overall MilTech. This is less plausible than the other processes, however, as interstate competition has been shown to be most intense involving culturally dissimilar polities [ 45 , 47 , 61 ]. Additional study is needed to fully clarify the different possible causal forces driving this effect and to explore the possible causal role that each of these potential processes play in the overall development and spread of these key military technologies, along with technological evolution more generally. Nevertheless, the finding that Phylogeny is a significant predictor of MilTech further speaks to the importance of connection-mediated information exchange, over and above closeness in space.

Lastly, we find that agricultural productivity, measured here as per-hectare tons of major carbohydrates, displays a significant effect on subsequent levels of MilTech. While we had no strong theoretical motivations for this idea, we included the term in analyses to test the possibility that an increased resource base would impact technological development. Its inclusion in our best model suggests that a certain level of agriculture productivity may have been a necessary component in generating and adopting new technologies. Perhaps a more efficient productivity was required to support large enough populations not primarily employed in agriculture, or expanding a society’s general resource base and extractive capacity provided the raw materials and intermediate goods used in constructing key military technologies. As noted, however, this factor displays a much weaker effect compared to the others, and is least robust to supplemental analyses. Thus, this result must remain tentative. Exploring more deeply the impact of agricultural productivity on the evolution of technology stands out as an important avenue for future research.

While these findings constitute an important first step towards identifying some of the major long-term drivers of technological evolution in general, and in the domain of military capacity in particular, and finding broad support for previously somewhat speculative theories, there is still much to be done to build on this line of research. First, it would be desirable to extend the geographical coverage beyond the stratified global sample used in the present study, particularly relating to the phylogenic connections in the spread of existing technologies and the different possible processes that lead to this interesting effect. Second, it would be important to explore the downstream consequences of changes in military technology for other aspects of human life, including levels of peacefulness (or, alternatively, mortality rates due to violence), equality (e.g. distributions of wealth, rights of citizenry, levels of exploitation and oppression based on class or ethnicity) and public health (e.g. longevity, infant mortality, nutrition, infection rates, etc.). Third, our goal was to offer a preliminary exploration of some key causal forces proposed to support the evolution of military technology, ignoring differences between the initial innovation of new technologies and subsequent adoption by other societies. Future work is needed to pinpoint the source(s) of invention and distinguish advances made by innovation from advances by later spread to assess whether the same or different factors drive each of these separate processes. Fourth, additional potential drivers of technological innovation in general should be explored, over and above the effects of population size, connectivity, and existing stocks of critical innovations, as well as analyzing further the potential causal role played by rising agricultural productivity. These explorations would include factors impacting resource scarcity (e.g. due to drought, pestilence, and other natural disasters), more direct measures of intergroup competition (e.g. levels and intensities of external warfare, cultural distance between competitors, and other exogenous factors), identifying various different regional INs which might (partially) overlap in time and space.

Finally, it is important for future studies to ‘narrow in’ on the details of some of the more macro-level processes suggested by the present study. In particular, it will useful to explore the possible impact of regional-level factors along with a broader range of technological innovations within the polity (e.g. in energy, construction, transportation, and information sectors). Seshat data is relatively coarse, resolved here to 100-year intervals. While this granularity is well suited to exploring broad, global-level dynamics over thousands of years, it likely misses some of the nuances and outlying patterns. Future effort can hopefully generate more fine-grained temporal data allowing for meso- and even micro-level scrutiny of the pathways to technological evolution taken by different societies in various times and places. Alongside this, we require more qualitative investigation into the details of the specific items as well as the less material, tactical and managerial aspects of technological development employed in a host of specific historical cases.

Beyond the insights gained from these analyses on the development of military technologies over the very long-term, we hope that the approach presented here, which explores likely casual theories against a wide body of empirical data gathered by the Seshat project, will provide a roadmap to these important future studies, allowing scholars to delve deeper into not only the critical ‘Military Revolutions’ throughout history, but into the evolution of technology generally.

Supporting information

Acknowledgments.

The authors are grateful to Sergey Nefedov who reviewed data and provided helpful comments. We thank also Christopher Chase-Dunn, Peter Grimes, Gene Anderson, and the SetPol project for their constructive critique on earlier versions of the manuscript, as well as Jennifer Larson and Alan Covey for helpful comments on previous drafts. We gratefully acknowledge the contributions of our team of research assistants, post-doctoral researchers, consultants, and experts. Additionally, we have received invaluable assistance from our collaborators. Please see the Seshat website ( www.seshatdatabank.info ) for a comprehensive list of private donors, partners, experts, and consultants and their respective areas of expertise.

Data Availability

All relevant data are within the manuscript and its Supporting Information files. Additionally, all data are available alongside a preprint publication of this article, accessible at: https://osf.io/mkhde/

Funding Statement

This work was supported by: a John Templeton Foundation grant to the Evolution Institute, entitled "Axial-Age Religions and the Z-Curve of Human Egalitarianism" (HW, PF, PT); a Tricoastal Foundation grant to the Evolution Institute, entitled "The Deep Roots of the Modern World: The Cultural Evolution of Economic Growth and Political Stability" (PT); an Economic and Social Research Council Large Grant to the University of Oxford, entitled "Ritual, Community, and Conflict" (REF RES-060-25-0085) (HW); a grant from the European Union Horizon 2020 research and innovation programme (grant agreement No 644055 [ALIGNED, www.aligned-project.eu ]) (HW, PF); a European Research Council Advanced Grant to the University of Oxford, entitled “Ritual Modes: Divergent modes of ritual, social cohesion, prosociality, and conflict" (HW, PF); a grant from the Institute of Economics and Peace to develop a Historical Peace Index (HW, PF, PT, DH); and the program “Complexity Science,” which is supported by the Austrian Research Promotion Agency FFG under grant #873927 (PT).

• 1. McNeill WH. The pursuit of power. Chicago, IL: University of Chicago Press; 1982. [ Google Scholar ]
• 2. Kremer M. Technological Change: One Million B.C. to 1990. The Quarterly Journal of Economics. 1993;108(3):681–716. [ Google Scholar ]
• 3. McNeill JR, McNeill WH. The Human Web: A Bird’s-Eye View of World History. New York: W. W. Norton; 2003. [ Google Scholar ]
• 4. McNeill WH. The Rise of the West. New York: New American Library; 1963. [ Google Scholar ]
• 5. Duffy M, editor. The military revolution and the state, 1500–1800. Exeter: University of Exeter Publications; 1980. [ Google Scholar ]
• 6. Downing B. The military revolution and political change. Princeton, NJ: Princeton University Press; 1992. [ Google Scholar ]
• 7. Parker G. The Military Revolution, 1500–1800: Military Innovation and the Rise of the West (2nd ed.). Cambridge: Cambridge University Press; 1996. [ Google Scholar ]
• 8. Turchin P. Ultrasociety: How 10,000 Years of War Made Humans the Greatest Cooperators on Earth. Chaplin, CT: Beresta Books; 2016. [ Google Scholar ]
• 9. Wittrock B. The Meaning of the Axial Age. Axial Civilizations and World History. 2005:51–85. [ Google Scholar ]
• 10. Bellah RN. Religion in Human Evolution: From the Paleolithic to the Axial Age. Cambridge, MA: Harvard University Press; 2011. [ Google Scholar ]
• 11. Morris I. Why the West Rules—For Now: The Patterns of History, and What They Reveal About the Future. New York: Farrar, Straus and Giroux; 2010. [ Google Scholar ]
• 12. Morris I. The Measure of Civilization: How Social Developments Decides the Fate of Nations. Princeton: Princeton University Press; 2013. [ Google Scholar ]
• 13. Pomeranz KL. How Big Should Historians Think? A Review Essay on Why the West Rules—For Now by Ian Morris. Cliodynamics. 2011;2:304–29. [ Google Scholar ]
• 14. Schumpeter JA. Business Cycles. New York: McGraw-Hill; 1939. [ Google Scholar ]
• 15. Arthur WB. The Nature of Technology: What It Is and How It Evolves New York: Free Press; 2011. [ Google Scholar ]
• 16. Mokyr J. The Lever of Riches: Technological Creativity and Economic Progress. New York: Oxford University Press; 1992. doi: 10.1002/ijc.2910510117 [ DOI ] [ Google Scholar ]
• 17. Mokyr J. The Gifts of Athena: Historical Origins of the Knowledge Economy. Princeton, NJ: Princeton University Press; 2002. doi: 10.1007/s00262-002-0271-9 [ DOI ] [ Google Scholar ]
• 18. White LA. Energy and the Evolution of Culture. American Anthropologist. 1943;45(3):335–56. [ Google Scholar ]
• 19. White LA. The Evolution of Culture. New York: McGraw-Hill; 1959. [ Google Scholar ]
• 20. Cipolla CM. Guns, Sails, and Empires: Technological Innovation and the Early Phases of European Expansion, 1400–1700. New York: Pantheon Books; 1965. [ Google Scholar ]
• 21. Modelski G, Thompson WR. Leading sectors and world powers: the coevolution of global politics and economics. Columbia, SC: University of South Carolina Press; 1996. [ Google Scholar ]
• 22. Devezas T, Corredine J. The Nonlinear Dynamics of Technoeconomic Systems. An Informational Interpretation. Technological Forecasting and Social Change. 2002;69:317–57. [ Google Scholar ]
• 23. Turchin P, Gavrilets S. Evolution of Complex Hierarchical Societies. Social History and Evolution. 2009;8(2):167–98. [ Google Scholar ]
• 24. Turchin P, Currie TE, Turner EAL, Gavrilets S. War, Space, and the Evolution of Old World Complex Societies. PNAS. 2013;110:16384–9. doi: 10.1073/pnas.1308825110 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 25. Richerson P, Baldini R, Bell AV, Demps K, Frost K, Hillis V, et al. Cultural group selection plays an essential role in explaining human cooperation: A sketch of the evidence. Behavioral and Brain Sciences. 2016;39(e30). [ DOI ] [ PubMed ] [ Google Scholar ]
• 26. Currie TE, Turchin P, Turner E, Gavrilets S. Duration of agriculture and distance from the steppe predict the evolution of large-scale human societies in Afro-Eurasia. Humanities and Social Sciences Communications. 2020;7(1):34. [ Google Scholar ]
• 27. Bennett JS. Retrodicting the Rise, Spread, and Fall of Large-scale States in the Old World. SocArXiv Preprint. 2020. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 28. Taagepera R. Crisis around 2005 AD? A technology-population interaction model. General Systems. 1976;21:137–8. [ Google Scholar ]
• 29. Taagepera R. People, skills, and resources: An interaction model for world population growth. Technological Forecasting and Social Change. 1979;13(1):13–30. doi: 10.1016/0040-1625(79)90003-9 [ DOI ] [ PubMed ] [ Google Scholar ]
• 30. Komlos J, Nefedov S. A compact macromodel of pre-industrial population growth. Historical Methods. 2002;35:92–4. doi: 10.1080/01615440209604133 [ DOI ] [ PubMed ] [ Google Scholar ]
• 31. Tsirel S. On the Possible Reasons for the Hyperexponential Growth of the Earth Population. In: Dmitriev MG, Petrov AP, editors. Mathematical Modeling of Social and Economic Dynamics. Moscow: Russian State Social University; 2004. p. 367–9. doi: 10.1107/S010876730401339X [ DOI ] [ Google Scholar ]
• 32. Podlazov AV. A theory of the global demographic process. Herald of the Russian Academy of Sciences. 2017;87(3):256–66. [ Google Scholar ]
• 33. Korotayev A. A compact macromodel of World System evolution. Journal of World-Systems Research. 2005;11(1):79–93. [ Google Scholar ]
• 34. Korotayev A, Malkov A, Khaltourina D. Introduction to Social Macrodynamics: Secular Cycles and Millennial Trends. Moscow: URSS; 2006. doi: 10.1074/jbc.M601758200 [ DOI ] [ Google Scholar ]
• 35. Korotayev A, Malkov A. A compact mathematical model of the World System economic and demographic growth, 1 CE– 1973 CE. International Journal of Mathematical Models and Methods in Applied Sciences. 2016;10:200–9. [ Google Scholar ]
• 36. Grinin L, Markov A, Korotayev A. On similarities between biological and social evolutionary mechanisms: Mathematical modeling. Cliodynamics. 2013;4:185–228. [ Google Scholar ]
• 37. Salisbury RF. From stone to steel: economic consequences of a technological change in New Guinea. Cambridge: Cambridge University Press; 1962. [ Google Scholar ]
• 38. Styring A, Rösch M, Stephan E, Stika H-P, Fischer E, Sillmann M, et al. Centralisation and long-term change in farming regimes: Comparing agricultural practices in Neolithic and Iron Age south-west Germany. Proceedings of the Prehistoric Society. 2017;83:357–81. [ Google Scholar ]
• 39. Christian D. Maps of Time: An Introduction to Big History. Berkeley: University of California Press; 2004. [ Google Scholar ]
• 40. Korotayev AV, LePoire DJ, editors. The 21st Century Singularity and Global Futures: A Big History Perspective. Cham, Switzerland: Springer; 2020. [ Google Scholar ]
• 41. Chase-Dunn CK, Hall TD. Rise and demise: comparing world-systems. Boulder, CO: Westview Press; 1997. xi, 322 p. [ Google Scholar ]
• 42. Korotayev A. The World System Urbanization Urbanization Dynamics: A quantitative analysis. In: Turchin P, Grinin L, de Munck VC, Korotayev A, editors. History and Mathematics: Historical Dynamics and Development of Complex Societies. Moscow: URSS; 2006. p. 44–63. doi: 10.1074/jbc.M601758200 [ DOI ] [ Google Scholar ]
• 43. Korotayev A. Compact mathematical models of world system development, and how they can help us to clarify our understanding of globalization processes. In: Modelski G, Devezas T, Thompson WR, editors. Globalization as Evolutionary Process: Modeling Global Change. London: Routledge; 2008. p. 133–60. [ Google Scholar ]
• 44. Korotayev A, Zinkina J. Systemic boundary issues in the light of mathematical modeling of world-system evolution. Journal of Globalization Studies. 2017;8(1):78–96. [ Google Scholar ]
• 45. Barfield TJ. The perilous frontier: nomadic empires and China. Oxford, UK: Blackwell; 1989. [ Google Scholar ]
• 46. Kradin NN. Nomads, world-empires, and social evolution. In: Kradin NN, Korotayev AV, Bondarenko DM, Lynshi VA, editors. Alternative routes to civilization (in Russian). Moscow: Logos; 2000. p. 314–36. [ Google Scholar ]
• 47. Turchin P. A Theory for Formation of Large Empires. Journal of Global History. 2009;4:191–207. [ Google Scholar ]
• 48. Scheidel W. The Xiongnu and the comparative study of empire. In: Brosseder U, Miller BK, editors. iongnu archaeology–multidisciplinary perspectives on the first steppe empire in Inner Asia. Bonn: Rheinische Friedrich-Wilhelms-Universität Bonn; 2010. p. 111–20. [ Google Scholar ]
• 49. Collins R. Macrohistory: Essays in Sociology of the Long Run. Palo Alto, CA: Stanford University Press; 1999. [ Google Scholar ]
• 50. Hui VT-b. War and State Formation in Ancient China and Early Modern Europe [Hardcover]. Cambridge: Cambridge University Press; 2005. [ Google Scholar ]
• 51. Turchin P, Brennan R, Currie T, Feeney K, Francois P, Hoyer D, et al. Seshat: The Global History Databank. Cliodynamics. 2015;6:77–107. [ Google Scholar ]
• 52. François P, Manning JG, Whitehouse H, Brennan R, Currie T, Feeney K, et al. A Macroscope for Global History: Seshat Global History Databank, a methodological overview. Digital Humanities Quarterly. 2016;10(4). [ Google Scholar ]
• 53. Turchin P, Currie TE, Whitehouse H, Francois P, Feeney K, Mullins D, et al. Quantitative historical analysis uncovers a single dimension of complexity that structures global variation in human social organization. Proc Natl Acad Sci U S A. 2018;115(2):e144–e51. doi: 10.1073/pnas.1708800115 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 54. Whitehouse H, François P, Savage PE, Hoyer D, Feeney KC, Cioni E, et al. Big Gods did not drive the rise of big societies throughout world history. SocArXiv Preprint. 2021. [ Google Scholar ]
• 55. Kaplan JO, Krumhardt KM, Ellis EC, Ruddiman WF, Lemmen C, Goldewijk KK. Holocene carbon emissions as a result of anthropogenic land cover change. Holocene. 2011;21:775–91. [ Google Scholar ]
• 56. Zinkina J, Christian D, Grinin L, Ilyin I, Andreev A, Aleshkovski I, et al. A Big History of Globalization: The Emergence of a Global World System. Cham, Switzerland: Springer Nature; 2019. [ Google Scholar ]
• 57. Christian D. Silk Roads or Steppe Roads? The Silk Roads in World History. Journal of World History. 2000;11(1):1–26. [ Google Scholar ]
• 58. Beckwith CI. Empires of the Silk Road: A History of Cenrtal Eurasia from the Bronze Age to the Present. Princeton: Princeton University Press; 2009. [ Google Scholar ]
• 59. Williams T. The Silk Roads: An Icomos Thematic Study. International Council of Monuments and Sites. 2014. [ Google Scholar ]
• 60. McLaughlin R. The Roman Empire and the Silk Routes: The Ancient World Economy & the Empires of Parthia, Central Asia & Han China: Pen and Sword; 2016. [ Google Scholar ]
• 61. Turchin P. War and Peace and War: The Life Cycles of Imperial Nations. NY: Pi Press; 2006. [ Google Scholar ]
• 62. Nefedov SA. War and Society (In Russian: Voyna i obschestvo. Faktornyi analiz istoricheskogo protsessa). Moscow: Publishing House "Territoriya buduschego"; 2009. [ Google Scholar ]
• 63. Turchin P, Currie TE, Turner EAL. Mapping the Spread of Mounted Warfare. Cliodynamics. 2016;7:217–27. [ Google Scholar ]
• 64. Drews R. The End of the Bronze Age: Changes in Warfare and the Catastrophe ca. 1200 BC. Princeton: Princeton University Press; 1993. [ Google Scholar ]
• 65. Kim J. Elite Strategies and the Spread of Technological Innovation: The Spread of Iron in the Bronze Age Societies of Denmark and Southern Korea. Journal of Anthropological Archaeology. 2001;20(4):442–78. [ Google Scholar ]
• 66. Chirikure S. Metals in society: Iron production and its position in Iron Age communities of southern Africa. Journal of Social Archaeology. 2007;7(1):72–100. [ Google Scholar ]
• 67. Turner EAL. Discovering the Anvil Age: Mapping the spread of Iron metallurgy across Afro-Eurasia. Cliodynamics. 2020. [ Google Scholar ]
• 68. Turchin P, Hoyer D, Bennett J, Basava K, Cioni E, Francois P, et al. The Equinox2020 Seshat Data Release. Cliodynamics. 2020;11(1):41–50. [ Google Scholar ]
• 69. Eff EA, Routon PW. Farming and Fighting: An Empirical Analysis of the Ecological-Evolutionary Theory of the Incidence of Warfare. Structure and Dynamics. 2012;5(2):1–33. [ Google Scholar ]
• 70. Turchin P. Fitting Dynamic Regression Models to Seshat Data. Cliodynamics. 2018;9:25–58. [ Google Scholar ]
• 71. Johnson AW, Earle T. The evolution of human societies: from foraging group to agrarian state, 2nd edition. Stanford, CA: Stanford University Press; 2000. [ Google Scholar ]
• 72. Sanderson SK. Social Transformations: a General Theory for Historical Development. Lanham, MD: Rowman and Littlefield; 1999. [ Google Scholar ]
• 73. Turchin P, Currie TE, Collins C, Levine J, Oyebamiji O, Edwards NR, et al. An integrative approach to estimating productivity in past societies using Seshat: Global History Databank. Holocene. 2021;31(6):1055–65. [ Google Scholar ]
• 74. Wiener N. The theory of prediction. In: Beckenbath E, editor. ModernMathematics for Engineers. New York: McGraw-Hill; 1956. [ Google Scholar ]
• 75. Granger CWJ. Investigating Causal Relations by Econometric Models and Cross-spectral Methods. Econometrica. 1969;37(3):424–38. [ Google Scholar ]
• 76. Eff EA, Dow MM. How to deal with missing data and Galton’s problem in cross-cultural survey research: a primer for R. Structure and Dynamics. 2009;3(3):Article 1. [ Google Scholar ]
• 77. Dryer MS, Haspelmath M, editors. The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology; 2013. [ Google Scholar ]
• 78. Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York: Wiley; 1987. [ Google Scholar ]
• 79. Yuan YC. Multiple Imputation for Missing Data: Concepts and New Development (Version 9.0). Rockville, MD: SAS Institute; 2011. [ Google Scholar ]
• 80. Cioffi-Revilla C, Luke S, Parker DC, Rogers JD, Fitzhugh WW, Honeychurch W, et al. Computational Modeling Frontiers in International Politics: Agent-based Modeling and Simulation of Adaptive Behavior and Long-Term Change in Inner Asia. Proceedings of First World Congress on Social Simulation2006.
• 81. Baumard N, Hyafil A, Morris I, Boyer P. Increased Affluence Explains the Emergence of Ascetic Wisdoms and Moralizing Religions. Current Biology. 2015;25:1–6. doi: 10.1016/j.cub.2014.10.064 [ DOI ] [ PubMed ] [ Google Scholar ]
• 82. Frank AG. The Centrality of Central Asia. Studies in History. 1992;8(1):43–97. [ Google Scholar ]
• 83. Spolaore E, Wacziarg R. Long-Term Barriers to the International Diffusion of Innovations. NBER International Seminar on Macroeconomics. 2011;8(1):11–46. [ Google Scholar ]

Decision Letter 0

Olivier morin.

25 Jun 2021

PONE-D-21-17737

Rise of the War Machines: Charting the Evolution of Military Technologies from the Neolithic to the Industrial Revolution

Dear Daniel Hoyer,

I have received two reviewers' report on your manuscript and decided I would not wait for the third reviewer that I had invited. The two reviewers, I'm happy to say, concur that your manuscript is an important and valuable contribution to the cultural evolution literature. I share their point of view. Each reviewer offers specific recommendations that I would ask you to follow closely when revising the paper.

Reviewer 1 remarks that reducing the history of military technology to "hard" technologies (e.g. weapons or transportations) neglects important aspects of war tactics: "military doctrines, logistics, tactical skill, organizational structures, and ability to learn". In other words, your study captures the evolution of tactical "hardware", but not that of tactical "software", so to speak. They note that a more qualitative approach could capture this dimension. Please discuss this possible limitation of your study in the revision. Reviewer 1 also asks you to specify how your analysis can distinguish between technological change due to innovation vs. diffusion.

Reviewer 2 makes two specific and useful recommendations regarding your statistical analysis. The first is to make sure that the way you bundle military-technology related variables is justified. In other words, you should rule out the possibility that the effects you document are driven by one or a few outlier variables. Please implement the analysis they recommend, i.e., analysing random subsets of the 46 variables. Re-running your models with a subset of military technology variables is something you already do when you substitute "Core MilTech" for "MilTech". Reviewer 1 would like you to generalise this method to multiple, randomly selected subsamples. Reviewer 2 also suggests an alternative analysis for the impact of existing stock of technologies on the progress of military technologies.

In addition to these remarks, I have a number of editorial recommendations of my own, detailed below. Provided the revision appears to fulfil the reviewers' requirements and mine, I may not be sending it back for a new round of reviews.

My recommendations are either stylistic or related to the way statistical results are reported.

Concerning the reporting of results, my main concern is related to the repeated causality claims made in the manuscript. Given the data and methods, the results establish (at best) predictive causality in the sense of Granger. This type of causality cannot be equated with causality simpliciter , among other things because it does not rule our latent confounding effects. Please qualify all claims related to causality or causal inference.

I suspect there is a significant mistake in Table 1, which lists both MilTech and MilTech sq. as variables in the best-fitting model. This seems to contradict the description given on l. 301, but more importantly including two versions of the same variable in the same model raises obvious issues (multicollinearity, etc.). I suspect that you in fact tested 2 versions of the model, one with MilTech and the other with MilTech sq.

In general the manuscript in its current version assumes too much familiarity with the Seshat database and project, its organisation and its acronyms. Key constructs like "NGA" or "IN" are not explained or even glossed. Other constructs, like the "Scale" variable, will be familiar to readers who already read Seshat-based studies, but are only cursorily explained here (the characterisation given on l. 279 is insufficiently clear). For yet other variables like "SPC1" an explanation is promised but never provided (l. 349). There is also a tendency to use your own abbreviations for standard statistical constructs (e.g. using "delAIC" for ∆AIC or delta AIC). What is missing is not only a basic explanation of the Seshat lingo, but a sense of why the constructs were designed in the way that they are — for instance, why the Scale variable is a better measure of a polity's size than other possible measures.

Please submit your revised manuscript by Aug 09 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at [email protected] . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see:  http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at  https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols .

We look forward to receiving your revised manuscript.

Kind regards,

Academic Editor

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

3.  Thank you for stating the following in the Competing Interests section:

"The authors have declared that no competing interests exist."

We note that one or more of the authors are employed by a commercial company: "Complexity Science Hub."

a) Please provide an amended Funding Statement declaring this commercial affiliation, as well as a statement regarding the Role of Funders in your study. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. You can update author roles in the Author Contributions section of the online submission form.

Please also include the following statement within your amended Funding Statement.

“The funder provided support in the form of salaries for authors [insert relevant initials], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.”

If your commercial affiliation did play a role in your study, please state and explain this role within your updated Funding Statement.

b) Please also provide an updated Competing Interests Statement declaring this commercial affiliation along with any other relevant declarations relating to employment, consultancy, patents, products in development, or marketed products, etc. 

Within your Competing Interests Statement, please confirm that this commercial affiliation does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests ) . If this adherence statement is not accurate and  there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include both an updated Funding Statement and Competing Interests Statement in your cover letter. We will change the online submission form on your behalf.

Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: http://journals.plos.org/plosone/s/competing-interests

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This is a very valuable paper providing the most systematic large-n analysis to date of the evolution of military technology. It should acknowledge more clearly that large-n studies are just one of the tools needed for a proper explanation, but should definitely be published.

(Editor's note: I am here  pasting Reviewer 1's attached comments since they may not be automatically added to this action letter for word limit reasons. OM)

This is a valuable paper, making a real contribution to a significant debate. My one reservation is that I think this kind of large-n survey usually has to be combined with more detailed work to explain something as complicated as military technologies. Military historians typically find that technology is only one part of effectiveness, and not always the most important part. The most interesting result in this paper is the lack of fit between SocSoph and technological gains (lines 455-56, 532-36), but I wonder how much of this is because technologies are little use without appropriate doctrine (the central point in Stephen Biddle’s book Military Power [2004]). Fig. 5 illustrates this—looking at the scores for Latium around 2kya, they’re not very different from other societies, because Roman military technology wasn’t actually that different from other Mediterranean, Middle Eastern, and European societies (as measured in the categories at lines 287-98 in the supporting material); however, the Roman army’s ability to apply these technologies transformed what they meant in practical terms, and that was probably determined primarily by Roman SocSoph. I also immediately think of the classical Greeks, whose military technologies were very ordinary but their application of them was extraordinary.

One difficulty for large-n surveys is that while technological categories are relatively easy to identify in the archaeological record, military doctrines, logistics, tactical skill, organizational structures, and ability to learn aren’t; and even when ancient and medieval writers describe such doctrines, there’s no obvious way to score them. This paper should absolutely be published, because there’ve been few (if any) large-n surveys of this level of sophistication, but the main point it illustrates is perhaps that large-n surveys are only the beginning of the study of warfare, and we have to follow up on their results with analytic narratives and case studies.

I also thought that the paper should distinguish more between technological innovation and diffusion. I’m currently reading two case studies of the Comanches, who probably had the best light cavalry in the world in the 18th century. They didn’t invent light cavalry, because there were no horses in North America to invent it with before the Spaniards brought them, but they did adopt and then adapt horse-herding and perfect styles of mounted fighting much better than the Apaches or even the Spaniards and Mexicans. They didn’t adopt or adapt firearms because their light-cavalry tactics made muzzle-loading muskets irrelevant. The Comanches were eventually defeated because Texans adopted and adapted Comanche tactics in the 1830s and combined them with revolvers and, later, breech-loading rifles that had been invented on the US East Coast. Neither the Comanche nor the Texans could have invented revolvers and rifles themselves, but the US Army failed to figure out how to use the new technologies until Texans put them together with tactics in new ways; but the final Comanche defeat in the 1870s depended on the US Army learning from the Texans and then exploiting the scale of US federal infrastructure. 

These points aren’t just details that can be subsumed within a large-n model; we’re missing what really happened if we only see the coarse-grained technological/geographical narrative. The paper should be clearer that while a large-n survey is a necessary condition for a good explanation, it’s not sufficient, and should be treated as a starting point for other kinds of analysis.

One final detail: the throwaway comment on naval warfare on line 69 isn’t adequate. 

Reviewer #2: This paper examines the long-term evolution of military technologies using the Seshat: Global History Databank. Creating a series of composite variables, which use simpler variables as a proxy for more complex ones such as Information Complexity or Military Technologies, the authors then test several theoretical claims in the literature. Several factors seem to significantly predict the advancement of Military Technology (MilTech), e.g., global population size, cultural similarity, and the spread of iron and cavalry. They also find variables that do not appear to dramatically influence of the level of military technology. Specifically, with the notable exception of phylogeny, characteristics such as the scale and sophistication of a polity appear to be non-significant predictors. This leads them to conclude that military technology evolves, for the most part, as an exogenous variable.

Overall, I applaud the authors for doing a very thorough and ambitious investigation in the evolution of military technology. I believe the paper will provide a catalyst for many follow up studies and serve a vital role of stimulating scientific debate. I do, however, have two general points that I’d like to make.

For my first point, I want to highlight a potential limitation of this approach, in that aggregation might mask a simpler explanation for the results of your model. That is, your predictors might be predicting only a subset of your 46 binary variables used to create your MilTech measure. One way to deal with this is to create alternative MilTech measures by randomly sampling from your set of 46 variables. What you do here is create multiple MilTech measures composed from different, randomly sampled subsets of your 46 variables. How much does this change your overall results in this study? It might be the case that what your model is predicting is a specific subset of technologies that disproportionately influence the overall result. If the result is not robust to these random subsets, then I believe it diminishes your claim that the variables are predicting military technology per se. Instead, it might be the case that your models are predicting a specific subset of the technologies, as opposed to an aggregate measure of overall military technology.

My second point relates to the variable for the existing stock of technologies. Here, you use the existing stock of technology as a proxy for the influence of current technological level on military technological evolution. This is done in two ways: as a temporal autoregressive process and focusing on horse riding and iron smelting. For the temporal autoregressive measure, you could also look at a related approach such as transfer entropy. The advantage of this is that it does not rely on using a single timeseries to predict its future state. Instead, you can use separate timeseries (X and Y), as predictors of one another. This tells you how much uncertainty is reduced in the future values of Y by knowing the past values of X given the past values of Y. It is a way of measuring the influence of one timeseries process (X) on another timeseries process (Y). I feel that this would help you disentangle the directionality of the relationship between some of your variables (as the measure is non-symmetric and X|Y does not equal Y|X). So, for instance, you could have the existing stock of technology as one timeseries and the miltech variable as another timeseries. You would predict that the information flow goes from the existing stock to miltech, but not necessarily the other way around.

Finally, a minor point is that for the results in Table 1, you should probably use scientific notation for extremely small p-values (especially for the MilTech and MilTech.sq p-values where you just have 0). In fact, after writing this comment, I noticed you already did this in your supplementary materials (so it should be easy to address).

6. PLOS authors have the option to publish the peer review history of their article ( what does this mean? ). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy .

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool,  https://pacev2.apexcovantage.com/ . PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at  [email protected] . Please note that Supporting Information files do not need this step.

Submitted filename: PONE-D-21-17737 comments.docx

Author response to Decision Letter 0

Collection date 2021.

Our response to Reviewer and Editor comments are included in the uploaded file 'Response to Reviewers.docx'. We copy our responses below as well

Response to Editor

We thank both peer reviewers for their thoughtful and helpful comments as well as the Editor’s helpful summary and additional comments, and for allowing us the chance to revise our manuscript. Both reviewers had very favorable things to say about our analytical approach and description of results, each concluding that the paper will make a valuable contribution to the field.

The main concerns seemed to revolve around ensuring that we properly motivate our analytical approach and test the robustness of results to alternate methods, as well as properly qualifying our interpretations in light of the limitations of our methods. This later point is in line with the Editor’s main concern about our overly-ambitious causal inferences. We feel these are valid and quite useful critiques. We have made numerous edits and performed additional supplementary analyses to clarify our approach and interpretations and provide fuller support for our claims, while also highlighting the need for future work and alternative methods to further explore the evolution of military technology. Overall, our manuscript is much improved as a result of these comments, and we are grateful to the Reviewers for pointing out shortcomings in how we described and justified some of our central arguments.

Lastly, in the response to the original submission the PLOS editors requested that we add a Competing Interests Statement and amend the Funding Statement based on the idea that the Complexity Science Hub, Vienna (CSH), with which some of our coauthors are affiliated, is a commercial company. This is incorrect, as the CSH is a non-profit research institute based in Vienna. It is registered as a Verein (‘public association’) with the Austrian Vereinsbehoerden (you can find it listed as well with the Research Organization Registry, id: ror.org/023dz9m50 ). We have attached a file from the Austrian government regarding the CSH’s status. If you require further information about the CSH and its non-profit status, please contact Philipp Marxgut ( [email protected] ), CSH Secretary General. I have reaffirmed the lack of competing interests with all coauthors.

Responses to Editor Comments

See below for responses to R1’s comments

Reviewer 2 makes two specific and useful recommendations regarding your statistical analysis. The first is to make sure that the way you bundle military-technology related variables is justified. In other words, you should rule out the possibility that the effects you document are driven by one or a few outlier variables. Please implement the analysis they recommend, i.e., analysing random subsets of the 46 variables. Re-running your models with a subset of military technology variables is something you already do when you substitute "Core MilTech" for "MilTech". Reviewer 1 would like you to generalise this method to multiple, randomly selected subsamples.

This is a very good suggestion. We implemented this by using a kind of bootstrap, generating multiple response variables by randomly resampling Warfare Components that make up our primary MilTech measure. We repeated the procedure 100 times to obtain and analyze the 100 best models by AIC. This analysis revealed that the most strongly supported terms found in the primary analysis remain supported, appearing in the best models in all or all-but-one instance. The only significant difference is the Agri term, which appeared in only 64 of the 100 cases using this bootstrap method.

This analysis suggests that randomly choosing WCs has a small effect on our results, with the exception of Agri. The implication is that different WCs can substitute for each other with little loss of information, something that was already hinted at with the comparison between MilTech and Core MilTech.

We have included this supplementary analysis in the SI (under the subheading ‘Testing bias in aggregation of MilTech variables’) with a description of results. We also added a discussion of this robustness check in the main text (in the Dynamic Regression Results section) and a paragraph in the Discussion detailing the more limited support that Agri receives as a key driver of MilTech compared to the other factors.

Reviewer 2 also suggests an alternative analysis for the impact of existing stock of technologies on the progress of military technologies.

See our response below.

Concerning the reporting of results, my main concern is related to the repeated causality claims made in the manuscript. Given the data and methods, the results establish (at best) predictive causality in the sense of Granger. This type of causality cannot be equated with causality simpliciter, among other things because it does not rule our latent confounding effects. Please qualify all claims related to causality or causal inference.

This is a very good point and we agree that our original language did not adequately convey the appropriate qualifications to our interpretation of results. We have edited text throughout the MS to make this more clear.

Specifically, we added an explanation in the Statistical Analysis section explaining how we use the dynamic analyses to support Granger-type causal inference, but stressing also that these inferences are partial and provisional and need further study to support. We have also added text at the end of the Results section clarifying that our interpretations are based on the predictive relationships between our (theoretically-motivated) predictor and response variables. We stress as well that our interpretations are provisional and subject to additional analyses and alternate approaches. Finally, we have modified text throughout the Discussion section to qualify the conclusiveness of our claims and highlight the suggestive nature of these results, requiring future work to explore further.

This is not a mistake. We routinely perform checks for nonlinearity in how predictors affect the response variable. This check showed that the relationship between MilTech(t+1) and MilTech(t) (that is, lagged value of the response variable) is curvilinear. We capture this nonlinearity with a quadratic term, MilTech (t) squared. There is a strong statistical support for this form, because both the linear and quadratic MilTech have high t-values, and P << 0.0001. For this reason, we retain both terms in the best-fitting model. The interpretation of this statistical result is that the evolution of MilTech is characterized by stabilizing selection, with the equilibrium trajectory set by the predictors. If a random perturbation moves MilTech level away from the equilibrium, the negative quadratic term ensures that MilTech returns to it.

We have added text in the Statistical Analysis section clarifying the inclusion of this square term. Detailing the statistical model, we now note: "We also include quadratic versions of these terms at a time lag (not shown) in order to explore non-linear responses to response and predictor factors. "

We are grateful for bringing to our attention this deficiency in our descriptions. We have modified text throughout to provide additional details about our methods and terms used as part of the Seshat project. Firstly, we explain our data coding procedure and define key terms early on in the main text (in A General Approach to Quantifying the Evolution of Pre-Industrial Societies subsection), and have added more detail to the descriptions of the factors used in analyses in Materials and Methods (‘Hypotheses to be Tested: Defining Predictor Variables’ subsection), particularly the Scale and SocSoph factors. We also clarified our NGA-based sampling procedure and edited text in the SI to clarify what the different factors were (e.g. the relationship between Scale and SocSoph to each other and as different dimensions of social complexity, as well as further detail on how and why we constructed the alternate response measure CoreMil).

Responses to Reviewer 1 Comments

This is a valuable paper, making a real contribution to a significant debate. My one reservation is that I think this kind of large-n survey usually has to be combined with more detailed work to explain something as complicated as military technologies. Military historians typically find that technology is only one part of effectiveness, and not always the most important part. The most interesting result in this paper is the lack of fit between SocSoph and technological gains (lines 455-56, 532-36), but I wonder how much of this is because technologies are little use without appropriate doctrine (the central point in Stephen Biddle’s book Military Power [2004]). Fig. 5 illustrates this—looking at the scores for Latium around 2kya, they’re not very different from other societies, because Roman military technology wasn’t actually that different from other Mediterranean, Middle Eastern, and European societies (as measured in the categories at lines 287-98 in the supporting material); however, the Roman army’s ability to apply these technologies transformed what they meant in practical terms, and that was probably determined primarily by Roman SocSoph. I also immediately think of the classical Greeks, whose military technologies were very ordinary but their application of them was extraordinary. One difficulty for large-n surveys is that while technological categories are relatively easy to identify in the archaeological record, military doctrines, logistics, tactical skill, organizational structures, and ability to learn aren’t; and even when ancient and medieval writers describe such doctrines, there’s no obvious way to score them. This paper should absolutely be published, because there’ve been few (if any) large-n surveys of this level of sophistication, but the main point it illustrates is perhaps that large-n surveys are only the beginning of the study of warfare, and we have to follow up on their results with analytic narratives and case studies.

This is an excellent point and we agree that the material components of technological development are only one part of the story. The tactical, organizational, and managerial aspects that go into how these technologies are wielded certainly play key roles in their effect. Here we are primarily interested in the evolution of the technologies themselves rather than the other aspects, both because, as R1 notes, coding these more ephemeral aspects is very difficult (though certainly possible using our approach to similar doctrinal ideological categories; e.g. Mullins et al. 2018, Whitehouse et al. forthcoming) and beyond the scope of what we could include here, but also because part of the motivation for the present study is to model an approach that could be extended to such follow-up questions.

We have added text in various places in the main text to make this explicit. In particular, we added to the Discussion section text noting the limitations of our approach and the relatively coarse detail of our data, remarking how more fine-grained study and qualitative investigation are important areas for future research.

It is worth noting too that we entirely agree that the lack of significant effect seen between SocSoph and our response measure is an important (and somewhat surprising) result. It may very well have been expected that the SocSoph measure would mediate the ability of societies to effectively wield technologies as R1 notes; namely, that societies with more sophisticated information storage and exchange systems would be better equipped to develop or adapt the tactical and organizational mechanisms that accompany material technologies. Yet, as noted we do not see a significant relationship between SocSoph and MilTech or any other polity-specific factors. As we discuss in the main text, we interpret this as suggesting that MilTech is driven largely by extra-polity processes. The accompanying tactics and other ephemeral aspects that accompany technological change may develop after the technologies themselves have been widely adopted, or evolve through separate processes; again, future work exploring tactical and organizational aspects of technological evolution would be needed to provide insight into this open question, which is beyond the scope of the present study but we feel is ripe for investigation using the Seshat approach evidenced here; we have added text at the end of the Discussion section remarking on this.

Finally, the tactical knowledge aspect of technological change may be reflected to some degree in our measure of cultural similarity (Phylogeny). Indeed, it may be that the tactical knowledge accompanying major technological developments (or standing on their own merits) is driven by external competition or other forces, and is then spread more effectively to culturally similar societies than dissimilar ones, since closeness in both language and socio-cultural systems could facilitate the adoption and adaption of these tactics. This may help to explain the result showing that Phylogeny is a significant predictor of subsequent MilTech, in addition to the benefits of exchanging information about the material and mechanical components of new technologies. We have added discussions to this effect in the Discussion section where we treat this finding. We feel that our overall findings are strengthened by this additional dimension and thank R1 for raising this very important and interesting aspect of our results that we had not fully considered before.

Mullins, Daniel Austin, Daniel Hoyer, Christina Collins, Thomas Currie, Kevin Feeney, Pieter François, Patrick E. Savage, Harvey Whitehouse, and Peter Turchin. “A Systematic Assessment of ‘Axial Age’ Proposals Using Global Comparative Historical Evidence.” American Sociological Review 83, no. 3 (May 8, 2018): 596–626. https://doi.org/10.1177/0003122418772567 .

Whitehouse, Harvey, Pieter François, Daniel Hoyer, Kevin Chekov Feeney, Enrico Cioni, Rosalind Purcell, Robert M. Ross, et al. “Big Gods Did Not Drive the Rise of Big Societies throughout World History.” OSF Preprints, April 1, 2021. https://doi.org/10.31219/osf.io/mbnvg .

I also thought that the paper should distinguish more between technological innovation and diffusion. I’m currently reading two case studies of the Comanches, who probably had the best light cavalry in the world in the 18th century. They didn’t invent light cavalry, because there were no horses in North America to invent it with before the Spaniards brought them, but they did adopt and then adapt horse-herding and perfect styles of mounted fighting much better than the Apaches or even the Spaniards and Mexicans. They didn’t adopt or adapt firearms because their light-cavalry tactics made muzzle-loading muskets irrelevant. The Comanches were eventually defeated because Texans adopted and adapted Comanche tactics in the 1830s and combined them with revolvers and, later, breech-loading rifles that had been invented on the US East Coast. Neither the Comanche nor the Texans could have invented revolvers and rifles themselves, but the US Army failed to figure out how to use the new technologies until Texans put them together with tactics in new ways; but the final Comanche defeat in the 1870s depended on the US Army learning from the Texans and then exploiting the scale of US federal infrastructure.

We entirely agree that the distinction between invention and later diffusion is a critical one to make to understand the histories of technological spread and use. However, as noted above we intend for this piece to be only the opening salvo in a line of work investigating the various aspects of technological evolution, military and more generally. We feel it is important to lay the foundation and model an approach to these such future work by focusing on technological evolution more broadly – which, as we define here, includes innovation as well as adoption and adaptation. This is why we conflate these different processes in this piece.

We have added text to make this more clear to readers. Specifically, in the Introduction we clarify how we define ‘technological evolution’ for the purposes of this study. We also clarify in the Results and Discussion sections that change in MilTech can occur through multiple mechanisms, e.g. when describing regional patterns in the pace of technological change (distinguishing early-adopters from late-joiners to the growing global information network) we explain that we do not distinguish innovation from later adoption. We also highlight that disentangling these process will be an important focus of future studies.

The example R1 cites is very interesting. From the perspective of the evolution of military technology broadly, this case actually illustrates well the processes identified by our analyses; various technologies (and tactics) not invented in North America are introduced as the area becomes entangled in the information exchange networks of western Europe (through conquest, settlement, and expansion) and then these become widespread in use within a context of fierce interstate competition. While our coarse-grained macro-level model can not predict which group would win out in this competition, the key point from this level of analysis is that information exchange and inter-group competition seem to have facilitated the spread of key military technologies (herding, mounted archery, hand-held firearms) into an area that did not have them previously

We, indeed, agree that more fine-grained and detailed study is a necessary component of reconstructing social dynamics, as noted above. We would argue though that neither the fine-grained nor the broader, macro-level approach reveals what ‘really happened’, rather that both are needed together to untangle these complicated patterns; these are complimentary, rather than competing approaches (we discuss this in for example in Turchin 2008 as well as Francois et al 2016). For the reasons detailed above, here our focus is on the macro-level patterns. We note this in the Materials and Methods section and have provided additional text in the Discussion section emphasizing that future work exploring more fine-grained data and qualitative case studies is needed to delve deeper into the broader patterns and processes suggested by the analyses here.

Turchin, Peter. “Arise ‘Cliodynamics.’” Nature 454, no. 7200 (2008): 34–35. https://doi.org/10.1038/454034a .

François, Pieter, J. G. Manning, Harvey Whitehouse, Rob Brennan, T. E. Currie, Kevin Feeney, and Peter Turchin. “A Macroscope for Global History. Seshat Global History Databank: A Methodological Overview.” Digital Humanities Quarterly 10, no. 4 (2016). http://www.digitalhumanities.org/dhq/vol/10/4/000272/000272.html .

One final detail: the throwaway comment on naval warfare on line 69 isn’t adequate.

This is a good point and we agree that this was a poorly articulated point. We have removed this ‘throwaway’ comment from the main text, which we believe reads more clearly now. We discuss our inclusion of primarily land-based, rather than specifically maritime, technologies in the SI (the Defining the Response Variable: coding military technologies in past societies section) which we feel is adequate to explain our approach to readers. We would however be willing to add a discussion on this in the main text, if that would be useful to readers.

Responses to Reviewer 2 Comments

This paper examines the long-term evolution of military technologies using the Seshat: Global History Databank. Creating a series of composite variables, which use simpler variables as a proxy for more complex ones such as Information Complexity or Military Technologies, the authors then test several theoretical claims in the literature. Several factors seem to significantly predict the advancement of Military Technology (MilTech), e.g., global population size, cultural similarity, and the spread of iron and cavalry. They also find variables that do not appear to dramatically influence of the level of military technology. Specifically, with the notable exception of phylogeny, characteristics such as the scale and sophistication of a polity appear to be non-significant predictors. This leads them to conclude that military technology evolves, for the most part, as an exogenous variable.

We thank R2 for this positive and encouraging assessment.

We addressed this point by the analysis and results described above (in our responses to the Editor). Note that randomly sampling from the 46 variables is not a feasible approach because of the way we construct our Warfare Characteristics (WCs). For example, if a society has iron technology, than we assume that it also has the knowhow to produce “simpler” technologies (copper, bronze). Similarly, if a society has crossbows, then it doesn’t matter whether they know atlatls. Thus, sampling randomly form the raw binary variables will not yield the desired test. Instead we implemented this suggestion by randomly sampling from the six WCs – please see the results above and the discussion of these results now included in the SI.

This is a good point and we have improved our explanation of the logic underlying our dynamic regression methodology. The approach we use is known among the econometricians as ‘Granger Causality’, as was pointed out by the Editor above. Transfer Entropy (TE) is a closely related approach; in fact, it was shown to be the equivalent to Granger Causality for Gaussian variables (Barnett et al. 2009). There is some discussion in the literature about which approach is better to take. TE is a nonparametric method, and so it doesn’t depend on the assumption of linearity. However, we test for nonlinearities and include nonlinear effects, where detected, using polynomials. On the downside, nonparametric methods require more data and have lower statistical power when variables are well-behaved. In our case, both inspection of scatter plots and formal regression results suggest that in most cases linear forms capture well the relationships between the response and predictors. And where this is not the case (as with the functional relationship between MilTech and its lag), the quadratic relationship appears to capture the curvature well (which is not huge, as visual inspection suggests).

To summarize, we believe that our approach is doing precisely what the reviewer would like us to do – exploring the full range of possible cross-relationships between our data – and we feel that using TE would not add any additional information, but would reduce the statistical power of results. For these reasons, we have not included a TE analysis or discussion of this decision in the main text or as a supplementary analysis

Barnett, L., A. B. Barrett and A. K. Seth (2009). "Granger Causality and Transfer Entropy Are Equivalent for Gaussian Variables." Physical Review Letters 103(23): 238701.

We thank R2 for pointing out this oversight. We have corrected this in Table 1 of the main text.

Submitted filename: Response to Reviewers.docx

Decision Letter 1

20 Sep 2021

PONE-D-21-17737R1

Dear Dr. Hoyer,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/ , click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at [email protected] .

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact [email protected] .

Additional Editor Comments (optional):

Acceptance letter

27 Sep 2021

Dear Dr. Hoyer:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact [email protected] .

If we can help with anything else, please email us at [email protected] .

Thank you for submitting your work to PLOS ONE and supporting open access.

PLOS ONE Editorial Office Staff

on behalf of

Dr. Olivier Morin

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Data availability statement.

• View on publisher site
• PDF (2.7 MB)
• Collections

Similar articles

Cited by other articles, links to ncbi databases.

• Download .nbib .nbib
• Format: AMA APA MLA NLM

Add to Collections

The Society for Military History

Online Publications

Air & Space Power Journal Published quarterly, this is the professional flagship publication of the United States Air Force. It is designed to serve as an open forum for the presentation and stimulation of innovative thinking on military doctrine, strategy, tactics, force structure, readiness, and other matters of national defense. The online journal can be viewed in English, Spanish, and Portuguese editions. Armchair General Armchair General is a new history magazine that is designed to be an interactive experience in military history and strategy. Arms and Armour A peer-reviewed academic journal which aims to encourage and publish serious research in the field from scholars, both professional and amateur, around the world. Its areas of study, which are multi-disciplinary in approach and content, include, but are not limited to, the history, development, use, decoration and display of arms and armour throughout history. Baltic Security and Defence Review (BSDR) Baltic Security and Defence Review is the annual publication of the Baltic Defence College Baltic Defence College Joint Course MA Theses These theses have been placed on line as a research resource for scholars doing research in current security issues. These theses are also a good source of English language information on security issues relating to the Baltic States – Estonia, Latvia and Lithuania. Mark Calhoun, General Lesley J. McNair: Unsung Architect of the US Arm y - https://www.facebook.com/lesley.mcnair.unsung.architect Canadian Military Biography Canadian Military Journal Available in both French and English versions. From Colony to Country : A Reader's Guide to Canadian Military History From the National Library of Canada. John M. Gates, The U.S. Army and Irregular Warfare Joint Electronic Library Maintained by by the Defense Technical Information Center. Joint Force Quarterly Published for the Chairman, Joint Chiefs of Staff, by the Institute for National Strategic Studies, National Defense University, to promote understanding of the integrated employment of land, sea, air, space, and special operations forces. The journal focuses on joint doctrine, coalition warfare, contingency planning, combat operations conducted by the unified commands, and joint force development. Journal for Maritime Research The Journal for Maritime Research is a subscription-only, international journal targeted at maritime historians, commercial maritime organisations and maritime museum staff. MHQ: The Quarterly Journal of America's Military Past Military History Monthly Military Times is a monthly magazine dedicated to the history of conflict on Land, at Sea and in the Air. MilitaryHistoryOnline.com A webzine of community-submitted articles. Military Review The professional journal of the United States Army, prepared by US Army Command and General Staff College. Military Thought A Russian journal of military theory and strategy. East View Press holds the English translation rights to the key journal of the Russian Defense Ministry (titled Voennaia mysl), from which they have been translating and publishing selections since 1992 as the journal Military Thought . Marine Corps Gazette The professional journal of the United States Marine Corps Naval War College Review Parameters: The US Army War College Quarterly The United States Army's Senior Professional Journal. Scientia Militaria Academic journal from the Military Academy of South Africa The US Naval Institute Proceedings War and Society Published by the School of Humanities and Social Sciences of the University of New South Wales, Canberra, Australia. Paul Walsh has created an index covering the first thirty-nine volumes of the War & Society journal. The index is divided into three sections: articles by volume/issue, articles by subject, and articles alphabetically by author. The document can be downloaded at https://www.smh-hq.org/docs/WARandSOCIETYindex1983-2020.pdf .

Terms of Use and Privacy The Society for Military History's Website Terms of Use governs all use of the smh-hq.org website and all content, services and products available at of through the website. Additional information may be found in the SMH Privacy Policy .