Recent Development of Social Simulation as Reflected in JASSS between 2008 and 2014: A Citation and Co-Citation Analysis

The research field of social simulation comprises many topics and research directions. A previous study about the early years indicated that the community has evolved into a di erentiated discipline. This paper investigates the recent development of social simulation as reflected in Journal of Artificial Societies and Social Simulation (JASSS) publications from 2008 to 2014. By using citation analysis, we identify the most influential publications and study the characteristics of citations. Additionally, we analyze the development of the fieldwith respect to research topics and their structure in a co-citation analysis. The citation characteristics support the continuing highly multidisciplinary character of JASSS. Prominently cited are methodological papers and books, standards, and NetLogo as themain simulation tool. With respect to the focus of this research, we observe continuity in topics such as opinion dynamics and the evolution of cooperation. While some topics disappeared such as learning, new subjects emerged such as marriage formation models and tools and platforms. Overall, one can observe a maturing interand multidisciplinary scientific community in which both methodological issues and specific social science topics are discussed and standards have emerged.


Introduction
. Social simulation is considered to be a multidisciplinary and rapidly developing field (Meyer et al. ; Squazzoni et al. ).This is mirrored by increasing citations in di erent ISI-and Scopus-indexed sources (Squazzoni ) and by the fact that simulation methods have more recently gained a foothold in di erent social science publication outlets (Fioretti ; Leitner & Wall ; Secchi & Seri ).Nevertheless, the Journal of Artificial Societies and Social Simulation (JASSS) remains one of the major publication outlets for research in social simulation (Secchi & Seri ; Squazzoni ; Squazzoni & Casnici ).
. In , JASSS experienced a significant change that led to intensive discussions about the journal.A er years, Nigel Gilbert, the founder of the journal, handed over his responsibilities as editor, which resulted in a debate in the social simulation community about the future direction of JASSS (see e.g., SimSoc ).This discussion addressed the scope of the journal and its possible future direction.Some perceive JASSS as an interdisciplinary journal at the intersection of various fields such as the social, behavioral, and computational sciences, whereas others suggest that the journal's scope should be extended to include more fundamental questions of science that support simulation research in general.Another group points out that JASSS frequently publishes technical, epistemological, and methodological papers.This discussion shows the di erent perspectives on the journal and its role in the community.

.
This diversity might also be due to the multidisciplinary and dynamic character of JASSS, which makes it even more challenging to get an overview of the journal, the related fields of social simulation, and the recent development of both.Such an overview, however, would be beneficial for several reasons.First, it would allow for an empirical basis for the above-described discussions about the past and future direction of JASSS.Second, given the increasing interest from other disciplines, it would make access to the field much easier, as newcomers could inform themselves about the current state of the literature.Of particular relevance for this would be a summary of the most influential articles and main foci of research.Finally, such an overview would complement a previous study about the development of the field in the early years, from its beginning in until (Meyer et al. ).In combination, the two may provide an overview of the whole time span of the first years of JASSS under the editorship of Nigel Gilbert and the field during that time. .The objective of this paper is to provide such an overview.We map the recent developments in the field of social simulation as reflected in JASSS publications from to .This study focuses on developments regarding the most cited sources and on networks of frequently co-cited publications.Bibliometric methods such as citation and co-citation analysis are suitable to uncover hidden patterns in publication outlets.These patterns delineate historical developments, depict the current situation, and provide a foundation to discuss future developments (Van Raan ). .
In terms of the methods used, we closely link our work to a previous study of the development of JASSS in its first years (Meyer et al. ), which allows us to identify continuities as well as changes.We particularly want to investigate the following issues: ( ) What are the recent developments in view of cited publications, types of citation sources, or influences of certain fields?Does more recent development di er from that in the first years and in what respect? ( ) Which co-citation networks emerged, how strong are their relations, and to what research topics are these related?( ) Are trends observable, in the past seven years or the overall time span of years?Are there indicators for the future direction of the journal and social simulation as a discipline? .
The paper is organized as follows.In the next section, we describe our method and data set.A erwards, we present the most influential publications in JASSS and highlight some specific source characteristics regarding publication age, source type, and cited journals classified by discipline.Subsequently, we present our results of the co-citation analysis to identify research clusters in JASSS and investigate the relationships between these clusters.Further, we present a longitudinal analysis of the research topics in social simulation.Finally, we draw some conclusions and make suggestions for future research.

Method
. This paper investigates the development of the intellectual structure of JASSS from to .To this end, we apply the bibliometric methods of citation and co-citation analysis.Both methods are established for the analysis of scientific fields (Osareh a,b) and have been successfully applied to the analysis of di erent journals (Meyer et al. , ; Mustafee et al. a,b; Squazzoni & Casnici ).To maximize comparability with the previous study, we mainly follow the methods applied in Meyer et al. ( ), but extend them to explore additional questions. .Citation analysis investigates the occurrences of referenced publications.Via citations, an author shows the relation between the own work and the work of other scholars (Osareh a).Whilst the number of citations is generally considered to be an indicator of the degree of a study's perceived relevance and influence (Bornmann & Daniel ; Radicchi & Castellano ), citation counts also have weaknesses.Studies of citation behavior show that scientists not only cite other work to acknowledge the intellectual or cognitive influence of scientific peers, but also for other, probably less scientific reasons, which are individual and di erent (Bornmann & Daniel ).Furthermore, so-called "sleeping beauties", which are publications whose importance is not recognized for several years a er publication, may remain undiscovered (Ke et al. ).Still, citations are an indicator to determine the influence of publications and thus are commonly used evaluation measures. .
Co-citation analysis examines the relationships between cited publications.A co-citation means that two publications are cited in the same document.For example, citation A and citation B are co-cited if both publications are listed in the same reference list of article C. The number of co-citations among publications is regarded as an indicator of their proximity (Gipp & Beel ; Small ).The identified relationships between cited publications allow us to draw conclusions about the internal structure of research, based on the resulting clusters of publications. .
Using absolute citation values is not suitable to generate clearly defined clusters of publications.Sources with a high number of citations tend to appear more frequently in clusters than less cited sources due to their wide dissemination.To address this problem, several scaling methods have been developed.Gmür ( ) evaluated established methods and suggested a measure called a CoCit score, which sets the squared co-citation count in relation to the minimum and mean counts of two individual citations A and B (Gmür ). ).To reduce the complexity of analysis, we focus on the most cited publications, only including those with at least three co-citations.Further, we set a minimum CoCit score value of .as the threshold (both in line with Meyer et al. ).As a result, groups with distinguishable network topologies emerge such as isolated pairs, trees, mesh, and fully connected clusters.In this paper, we refer to a cluster if a network contains at least three sources linked by at least three co-citation relationships, with CoCit scores greater than or equal to .(as in Meyer et al. ).
. For the data set generation, we used the online index of JASSS articles.This index provides an open access database to all articles and their references .We gathered the data by parsing all journal articles published in JASSS between and , excluding book review articles.The parser retrieved the associated lists of references in a CSV file and assigned each source an ID.Identical sources were assigned the same ID.We corrected parser bugs and data inconsistencies manually .A erwards, we were able unambiguously to verify citations as duplicates.If necessary, we corrected the source IDs and frequency of citations by hand.We used the resulting data set for the citation analysis and proceeded with the calculation of symmetric reference-reference matrices for the co-citation analysis, in line with the description of Zhang et al. ( ).

Data Set
. The resulting data set forms the basis for the citation and co-citation analysis.First, we want to provide some descriptive statistics.This comprises the number of JASSS articles included as well as the publications referenced in these articles.Table shows the data set for -.The whole time span of years is included in the table to compare the results of our study with the results from the previous study (Meyer et al. ).
. The first study investigated two time periods, -and -.In this study, we address the subsequent years from to and divide these seven years into two periods of .years .This is driven by the motivation to create comparable time periods âĂŞ in terms of the number of analyzed articles and citations âĂŞ to the previous study.The number of JASSS articles increased over time per issue and year from articles in the first period of this study to articles in the second period.The overall increase in the number of articles and citations is also reflected in the average number of articles per issue.JASSS issues of later periods include more articles than issues in the early years (see, for example, an average number of .articles per issue in the last period compared with .articles per issue in the period before).Even though the recent periods are shorter, their number of citations is comparable with the two periods of the previous study.

.
To deal with the rise in scientific publications, we determine the growth rate for the domain of social simulation and incorporate that with regard to the split of the data sets.First, we consider the general rise in scientific publications.This trend was identified as corresponding to a doubling of the global number of references in scientific publications within years (Pan et al.
).Some scholars come up with e ective indices to discount exponential growth for selected domains (e.g., Parolo et al.
).Nevertheless, deflation indices are dependent on research domains and average indices neglect the heterogeneity within and across disciplines.Following our assumption that social simulation is at least partially reflected in JASSS, we determined its growth rate by analyzing the yearly number of citations in JASSS (see Figure ).Given our data set, we assumed an exponential function and identified an .% growth rate of references per year (R 2 = 0.76).This growth rate is above average in relation to the growth rate of .% for all domains, and remarkably higher than .% for the social sciences (Pan et al. ).

.
To understand this result, we consider the two parameters that determine the growth rate: the increasing number of publications and the increasing average number of references per publication.As previously mentioned, the social simulation growth rate is also driven by both parameters.We identify an increasing number of articles in JASSS as well as higher numbers of references.Overall, the comparable high growth rate indicates fast growth. .
The growth rate of references influences the comparability of citation metrics in di erent time spans.Average growth and deflation or inflation rates are metrics to reflect trends.Definite rates are per se not retrievable considering that references are commonly made in discrete time intervals of years.Given the calculated social simulation growth rate of .%, a six-year time span ( -) would be the most statistically comparable time span to the previous study ( -) to come to comparable data set sizes.Following our aim to cover the full time span of the editorship of Nigel Gilbert, we cover a seven-year time span in this study ( -) and accept a slight distortion.A division of the seven-year time span by growth rate approximately results in a four-year ( -) and a three-year ( -) period.Nevertheless, we slightly deviate from this and use equal periods of .years, on the grounds that an uneven split of the recent time span would dilute the tangibility of comparison.Finally, we account for the growth rate at a higher aggregation level.The previous study covers ten and our study a shorter time span of seven years.

.
Looking at the citation characteristics, we identify that the average source age increased from to .This e ect can be ascribed to the fact that some fundamental work is still cited in more recent publications.To visualize this, we plotted the publication years of the cited publications in JASSS per time period (see Figure ).The distribution graph shows a typical shape found in other fields as well, and could be approximated by a le -skewed distribution (see e.g., Schä er et al. ).
. Next, we investigate the frequency of citations.The repeated occurrence of citations in JASSS articles is the basis for our citation and co-citation analysis.The overview in Table shows the number of citations that occur once, twice, or at least three times.Most publications are only cited once, while only about % of the references occur multiple times.Compared with the previous study, the relative frequency of single citations increased from about % to more than % in the most recent period (Meyer et al. ).This indicates slightly increasing diversification in terms of sources.

.
Citations that occur at least three times represent the data set for the co-citation analysis.We identify on average .citations that occur at least three times per classified time period in JASSS.Regarding this measure, the two time spans ( years vs. years) of the studies are comparable.Within these time spans, however, only the two recent periods are similar in view of the number of citations that occur more than three times ( and ).We see an imbalanced distribution of citations vs.
citations within the first time span of the previous study.This also results in a denser co-citation network for the second period of the first study (Meyer et al. ).Previous study (Meyer et

Results of the Citation Analysis
. The citation analysis identifies the most cited sources and their characteristics, such as the external publication sources acknowledged in JASSS and corresponding disciplines.We extract the most common sources from the data set for both periods.Table ranks the most cited sources in descending order .In addition, the relative citation value is calculated as the number of citations divided by the number of JASSS articles published in the corresponding time period.Moreover, we classify the types of sources into books, journal articles, web pages, and proceeding papers.
. The results show three standard books that are cited frequently in both periods: Axelrod's The Evolution of Cooperation ( ), Epstein and Axtell's Growing Artificial Societies ( ), and Gilbert and Troitzsch's Simulation for the Social Scientist ( ) .In the second period, Nigel Gilbert's ( ) Agent-Based Models appeared.This book became the third most cited source with citations in the second period, and thereby superseded the highly cited books of Axelrod ( ) and Epstein & Axtell ( ).With the exception of the top three most cited sources, the majority of cited sources are journal articles.It is remarkable that in both periods, six articles were published in JASSS itself.).Thus, one can conclude that the simulation tool NetLogo is used in at least % of the studies recently published in JASSS .

.
In addition, there are indicators of another standard that may emerge for agent-based models.In particular, the Overview Design Details (ODD) protocol (Grimm et al. ) as well as its review and first update  .We further investigate the types of sources cited in the two periods (see Figure ).To this end, we classify the citations into nine categories: journal, book, book chapter, proceeding paper, lecture notes, working paper, thesis, web page, and miscellaneous .The trend reported in the previous study towards more journal article citations continues.In the recent period, .% of the total citations are journal publications, which reflects a continuous increase compared with .% for the first period.Correspondingly, the number of book and book chapter citations declined over time.This also indicates that more relevant journal publications exist that specialize in topics around social simulation, which again hints at a maturation of the field. .
In the previous study, web pages were part of the category "miscellaneous", but in the subsequent years they have become more cited.For this reason, we introduced "web page" as a new distinct category.Regarding the most cited sources, the increasing percentage of web citations basically results from the acknowledgement of simulation tools such as NetLogo, Repast, and Swarm, which are provided as open-source so ware via web pages.Articles consequently cite web pages that use these tools.In addition, a number of methodological papers compare simulation methods and tools by referencing the corresponding web pages.
. Finally, we investigate whether the multidisciplinary nature of social simulation observed by Meyer et al. ( ) is supported by our citation data.To this end, we identify the most frequently cited journals and their relative citation values based on the total number of , journal citations between and (see Table ).The ranking shows ten journals, which are represented in both time spans, which may indicates certain stability.Nevertheless, the impact of many journals changed.

.
JASSS itself is still by far the most cited journal ( .%).This result is in line with other self-citation rates of journals, which are about % (Thomson Reuters ).The second and third most cited journals are Nature ( .%) and Science ( .%).The list again provides evidence of the multidisciplinary nature of JASSS.The referenced journals cover a broad field of research disciplines among natural scientific journals (e.g., Physica A, Physical Review E, Journal of Theoretical Biology, and Ecological Modeling, social science and economic journals (e.g., American Journal of Sociology and American Economic Review), and a journal related to psychology (Journal of Personality and Social Psychology).

.
Against the background that the number of journal citations increased, we expect that journal citations are diverse.We applied the Herfindahl-Hirschman Index (HHI) to provide a concentration measure for journal citations (Schmalensee ), which has already been applied in other bibliometric studies (e.g., Chi ).Considering the number of journals (N ), the share of a journal based on citations (x i ), and the arithmetic average of the shares X, the HHI is calculated as follows: The result of the index is proportional to the average market share, and ranges from /N to .A higher index indicates a concentration of citations.Including the journal citations in the first time span ( -: N= ) and the journal citations in the second time span ( -: N= ), the HHIs result in .% and .%, respectively, which indicates a concentration of journal publications.This result is contractionary to our expectation of diversification as well as to the fact that N increases remarkably.The HHI is not invariant to N, as a greater N usually decreases the index.To examine this phenomenon further, we excluded the self-citations of JASSS and recalculated the index.This result shows diversification, with an HHI of .% for the first time span and .% for the second time span .Thus, the external environment of JASSS diversifies in terms of cited publications outside JASSS, while the high number of self-citations in JASSS points to concentration.

.
To investigate the multidisciplinary character of JASSS further, we classify all cited journals into subject fields.We conduct the classification by using the list "Essential Science Indicators Subject Areas" provided by Thomson Reuters ( ) .The list classifies , journals into a subject field, but some journals cited in JASSS are not covered.No category is assigned to .% of the journals cited in the first time period and .% in the second time period.JASSS is recorded in the category "social science, general".However, self-references of JASSS are excluded from the analysis in order to focus on the outgoing citations that unambiguously reflect the journal environment.

.
Based on journal citations, the impact of di erent subject fields on JASSS is shown in Figure .Recently, the category "social science, generalâĂİ occurred at the top of the list.In both time spans, most journals can be classified into the categories "social science, general" and "economics & business".The category "psychiatry/psychology" remains the third most influential category.The decreasing influence of computer sciencerelated journals is noticeable, while journals classified as "environment/ecology" gained more impact .
. The diversity of cited journals and subject fields supports JASSS's self-description as an "interdisciplinary journal for the exploration and understanding of social processes by means of computer simulation" (JASSS ).Given its orientation to the method of computer simulation, however, one may expect a stronger acknowledgement of journals in the discipline of computer science (Wellman ).However, this assumption is not reflected in the results.The methods of computer simulation are rather briefly referenced by authors.This can be ascribed to the fact that simulation concepts are also partially developed within social sciences (Davidsson ).This is di erent for the externally addressed research topics, mainly related to social science and economics and business.

.
Overall, the results of the citation analysis empirically support the dynamic and interdisciplinary character of social simulation.Moreover, the shi in publication outlets towards journals continues, which was considered to be an indicator for maturation in the previous study (Meyer et al. ).This indication is further supported in this study by the more frequent use of certain tools and standards.

Results of the Co-Citation Analysis
. While the citation analysis gives a first impression of the development in the academic community, the cocitation analysis uncovers the structure and interrelations within a discipline.By using a co-citation analysis, we identified clusters and distinct groups within them.All nodes and links that belong to a distinct group are numbered and colored , and these groups typically represent subfields and specific research streams in the discipline.Figure depicts the results of the co-citation analysis for the period from to / .The network has a density of .
and is composed of links and nodes.It consists of two clusters: cluster ( ) that comprises seven groups and a separated cluster ( ) that consists of a single group.

.
At the center of the first cluster is the group ( . ) Learning in Social Dilemmas .The hub of this cluster is the publication by Izquierdo et al. ( ) entitled "Reinforcement Learning Dynamics in Social Dilemmas", with nine links.The group is composed of nodes, links, and has a density of . .It is the most connected group within the surrounding groups and is centrally located in the first cluster. .
Next, we find two basic classes of topics in this cluster: Social science-related topics and methodologicaloriented topics.The groups ( . ) Norms and ( . ) Evolution of Cooperation represent social science topics such as group ( . ) Learning in Social Dilemmas.Topic ( . ) Environmental Aspects is rather separated.The other groups are related to methodological aspects in social simulation, namely ( . ) Modeling, ( . ) Validation, and ( . ) Replication.
. The second cluster represents the topic ( ) Opinion Dynamics, which can also be classified as related to social sciences.Five topics (Opinion Dynamics, Learning in Social Dilemmas, Norms, Modeling, and Environmental Aspects) have already been identified in the previous study (Meyer et al. ).We identify many connected groups in this period.General topics such as Learning, Validation, Replication, and Modeling are relevant for all simulation studies, and thus are acknowledged from many perspectives.

.
In the most recent period ( / to ), we observe a process of di erentiation (see Figure ), and identify six clearly separated clusters and eight groups.This network has a slightly higher density of ., including links and nodes in comparison with the earlier network ( to / ).Nevertheless, the groups in   .Further, new non-methodological topics emerged in the most recent period.The group ( ) Marriage Models shows a fully connected network pattern with a density of ., indicating a strong integration of publications.
In addition, the new group ( ) Simulation of Science fulfills the minimum criteria of three linked nodes to be depicted in the network.
. Given the co-citation results, social simulation scientists still vividly discuss methodological topics, as three out of eight identified groups deal with methodological aspects.Methodological topics are represented in the most recent period by the groups ( ) Standards, ( . ) Methodology, and ( . ) Tools and Platforms.The scientific discussion about tools and platforms is in line with our result from the citation analysis that NetLogo has become an established tool for social simulation researchers (see Section ).
. A er the identification of groups and their structure within the networks, we investigate how the identified groups are linked with each other.Due to the chosen CoCit score of ., not all links between groups are depicted.Thus, a further analysis investigates the aggregated strength of connections between groups.To analyze the strength of the connection between group X and Y with n possible links, we calculated the GroupCoCit score as follows (Meyer et al. ): . Subsequently, we use the same technique as before to visualize the resulting networks.The result for the period from to / is shown in Figure .The average GroupCoCit score is .and the median is . .Again, we depict only the value of the strongest GroupCoCit scores between the groups with a threshold of .(all scores are listed in Appendix B).

.
In line with the strongly interconnected co-citation network, we find strong relations among the methodological groups.The weakest link between the methodological groups ( . ) Validation and ( . ) Replication has a GroupCoCit score of . .The other links have relatively high scores, such as a score of .between ( . ) Validation and ( . ) Modeling and a score of .between ( . ) Modeling and ( . ) Replication.
. The group ( . ) Learning in Social Dilemmas is centrally located in the network and has six distinct connections to others.Looking more closely at this group, we can identify many publications around general topics such as reinforcement learning (Izquierdo et al.
), agent-based models (Epstein ), and the prominent publication Why agents?" by Axtell ( ).Thus, many articles in this group address general issues of agent-based modeling, which is relevant for many research perspectives in social simulation.The centrally located group ( . ) Learning in Social Dilemmas shows no link to the group ( ) Opinion Dynamics.This supports our observation that ( ) Opinion Dynamics is a rather separated group, while the others are connected.

.
The network and GroupCoCit scores for the period from / to are depicted in Figure .For the most recent period, we identify considerably low values of the GroupCoCit scores with a mean of .and a median of . .This supports our observation concerning the process of di erentiation in the most recent period. .
First, we see a triangle among ( ) Standards, ( . ) Methodology, and ( . ) Tools and Platforms.As in the previous period, we find strong connections among the methodological topics.The group ( . ) Tools and Platforms is linked with a score of .to ( . ) Methodology.Similarly strong is the connection between ( . ) Methodology and ( ) Standards.A weaker connection exists between ( ) Standards and ( . ) Tools and Platforms.) about the homophily concept.This publication is located in the group ( ) Opinion Dynamics, but is also co-cited with all eight publications in the marriage group .This connection between ( ) Opinion Dynamics and ( ) Marriage Models exemplifies the interdisciplinary use of common methods and concepts in the community, beyond the variety of topics.Thus, social simulation has interdisciplinary characteristics, represented here, for example, by the links between the groups, as well as multidisciplinary characteristics, as shown by distinguishable groups in the co-citation networks.

Longitudinal Analysis of Social Simulation
. The co-citation analysis allows us to identify developments in recent years as well as in comparison with the earlier years investigated in Meyer et al. ( ).Given our results of the two analyzed periods, a certain level of stability is observable, while at the same time some issues have been dropped and new topics have emerged.Methodological topics are strongly represented in both periods, as are the topics Opinion Dynamics and Evolution of Cooperation.On the other hand, some topics only emerged recently such as Marriage Models, Simulation of Science, and Tools and Platforms.In contrast, other topics disappeared such as environmental aspects and learning.This illustrates the dynamic processes in JASSS.

.
Overall, the results of this final analysis suggest that the research topics in the most recent period are more distinguishable and less strongly cross-linked.Methodological issues form their own clusters and groups, while several social science-related issues have evolved over time.The coexistence of methodological and social science-related subjects can be seen as a major pattern emerging from our co-citation analysis. .
To foster the interpretation of the longitudinal development, we provide a rank flow chart.Figure highlights the issues discussed in JASSS, as identified by co-citation analyses, along the -year editorship of Nigel Gilbert.
The topics are ranked according to the number of publications per topic-related group .For an overview of network metrics within the co-citation networks for all periods, see Appendix C.
. Clearly visible is the dominance of Opinion Dynamics as the biggest cluster in the recent three periods.This shows that Opinion Dynamics has developed into a long-term topic for social simulation researchers.Similar is the development of Evolution of Cooperation, which emerged in the last two periods.Alongside this, the development of the topic Reciprocity is closely connected to Evolution of Cooperation.
. Overall, three topics were prominent in three periods: Methodology, Opinion Dynamics, and Norms.These topics may constitute social simulation as a discipline.However, the fact that several non-methodological topics .Methodological issues have occurred in each analyzed period since the foundation of JASSS.Within this, processes of di erentiation and consolidation are observable.In the very first years from to , just a single group about the methodology of social simulation was identified.From to , two distinct methodological topics emerged: Modeling and Methodology.In the third period from to / , we identified the topics replication, validation, and modeling.Recently, in the fourth period from / to , three topics cover methodological aspects, which are Methodology, Standards, and Tools and Platforms.
. One can speculate whether this reflects the development of the research method simulation in the social sciences.First, general methodological aspects were discussed, informed initially by books such as Gilbert and Troitzsch's text .Next, issues such as the ghost in the model (Polhill et al. ), simulation model alignment (Epstein & Axtell ), and the value of replication (Edmonds & Hales ) became relevant for the first applications, discussed in the groups Modeling and Methodology in the second period.In the next period, Replication and Validation became such prominent topics that they emerged as groups of their own in the co-citation network.In recent years, the maturity of the method can be recognized by the development of tools, platforms, and standards as prominent topics and thus individual groups.

.
The topic Environmental Aspects was a big group in the co-citation network of the second period, yet it became a smaller group in the third period and disappeared in the last.There are many simulation studies of environmental aspects such as socio-ecological systems.Possibly, the discussions of environmental aspects shi ed to other journals outside JASSS.Here, we identified more referenced journals in environment and ecology in recent years.
. The theme Network and Di usion only emerged in the second period but disappeared therea er.Still, the citation analysis shows that the publication "Collective dynamics of 'small-world' networks" by Watts & Strogatz ( ) remains one of the most cited publications in JASSS.Thus, the topic Network and Di usion seems to merge with other fields of social simulation.

Conclusion
. This paper investigates the recent development of social simulation as reflected in articles published in JASSS from to by means of bibliometric methods.Therby, it o ers several theoretical and practical contributions.First, it provides an empirical basis from which to discuss the intellectual structure of JASSS and the related community.The results from the citation and co-citation analysis confirm the continuing multidisciplinary nature of JASSS, which is in line with its self-characterization.Further, the interdisciplinary exchange of knowledge extends beyond the boundaries of single disciplines and can be considered to be a distinctive characteristic of both the journal and the discipline.Furthermore, the citation analysis identifies NetLogo (Wilensky ) as the most cited source, which indicates that it has become an important modeling environment in the field of social simulation.The most cited sources as well as the co-citation networks indicate that general methodological issues are vividly discussed by the researchers publishing in JASSS.This result underpins the perceived role of JASSS as an interface for many researchers in the social sciences, who are linked with each other due to their common research interests in method simulation.
. Second, in combination with the previous study, changes and continuities can be identified over the longer time span of years.Looking at the most cited sources, several publications persist with high citation values, particularly Axelrod ( ); Epstein & Axtell ( ); Gilbert & Troitzsch ( ).This result highlights some of the well-established publications in the social simulation community.The trend to acknowledge more journal articles continues, as already identified in the previous study (Meyer et al. ).At the same time, some frequently cited book publications in the first years of JASSS lost their relevance.The citation results show that certain standards such as the ODD protocol (Grimm et al. , ) and tools such as NetLogo (Wilensky ) are now more frequently cited and thus are becoming increasingly established among researchers in the field.This observation indicates that the field might be evolving into a discipline with shared tools and standards.Along this line, the co-citation analysis depicts two long-term research topics in the field, which are Opinion Dynamics and Evolution.We conclude that these research fields are well-established in JASSS.On the other hand, new topics such as marriage formation and the simulation of science indicate a certain dynamic and openness concerning the main topics discussed in JASSS. .
Third, this study may provide an orientation for newcomers to the field.The list of most cited publications gives an overview of the basic literature and illuminates the standards and tools currently in use in the field, such as the ODD protocol and NetLogo.The co-citation networks display the current (and past) topics and their main publications.In combination with the publications citing them, these o er good starting points for individuals seeking to become familiar with the specific topics or the field in general.
. As with any study, this paper has limitations.First, we focused on a single journal, while also drawing some more general conclusions regarding social simulation.For a more comprehensive overview of the development of social simulation, more journals and proceedings should be included in the analysis.However, since JASSS is the main journal in the field, it is thus a good indicator for analyzing the general development.Moreover, citation studies su er from a certain time lag, as it takes some time for publications to appear and to be referenced by other authors.Further, we subdivided the seven-year time span into two equal .year periods to have a comparable picture about overall development with the previous study.Another subdivision may lead to slightly di erent results.The analysis also needs to be restricted to the most cited publications for reasons of complexity reduction.Still, we conducted robustness tests with varying CoCit scores, which overall showed similar qualitative results.
. Future research could, besides addressing these limitations, use di erent bibliometric methods such as author co-citation analysis or bibliographic coupling.Further, the set of investigated articles could be extended by using a key word search and related methods to identify relevant papers.Finally, additional insights could be gained from repeating this study in some years to map the next steps in the development of JASSS and social simulation.Still, we hope that this study currently fosters the understanding and acknowledgement of the development of the intellectual structure of JASSS and its related community of social simulation.
Appendix A: List of Publications Co-Citation Networks Here, an html parser implemented in JAVA was applied, which was also used by Meyer et al. ( ).Articles not available in html format were added manually.
First, we checked the CSV file with random samples for completeness to check whether all articles and references had been recorded by the parser.Subsequently, one of us did an exhaustive correction of inconsistencies in the CSV file, which resulted from di erent citation styles, italic letters, word wraps, and incorrect citations in the original html files.
We subdivided the total number of published articles in as follows.The period until / includes all articles of volumes ( ), ( ), and ( ).The remaining volume ( ) includes articles and thus a similar number of articles.Volume ( ) was published on -Jul-.
The publications included in the ranking are defined by the share of citations.We determined a share that results in a ranking of length that is comparable to the first study.There should be no publications le out of the ranking that have the same number of citations.We did not include more than publications, as there are several publications that follow the most cited sources in their share of citations, so that the list would be too long.
The book Simulation for the Social Scientist of Gilbert and Troitzsch was originally published in .A second edition was published in .In our study, we refer with references of to both published editions.
Given the NetLogo citations divided by the articles published in JASSS from / to .
The category "journal" also includes peer-reviewed e-journals.The category "working paper" includes all citations with the key words of working paper, mimeo, discussion paper, position paper, and research report.The category "proceeding paper" includes paper citations with attached pdf links and the key words of symposium, conference, and workshop.The category "miscellaneous" includes all citations that could not clearly be assigned to a category such as statistical reports, technical papers, technical reports, newspaper articles, and unpublished conference talks.
The sensitivity of HHI to N , allows only for a comparison with studies with a similar N .We use the index here mainly to analyze the diversity between the time periods.
A direct comparison of the results of this analysis with those of the previous study (Meyer et al. ) would be limited due to substantial changes in terms of journal classifications in the Thomson Reuters SCII/ISI.Therefore, we also re-categorized the data of the first study by using the list "Essential Science Indicators Subject Areas" to make the results comparable.
The result of the subject field analysis is limited by the validity of the journal classification by Thomson Reuters.Scientific subject fields are o en ill-defined and blurry.Journals can represent an intersection of articles, which can be related to di erent subjects.For this reason, the classification is ambiguous (Bensman & Leydesdor ).For a detailed analysis of inward and outward citations, their interrelations, and the bibliographic impact of JASSS publications on certain research domains, see Squazzoni & Casnici ( ).
The network mapping and network colors of nodes and links are based on the function "Newman Grouping" provided by the used tool Organizational Risk Analyser (ORA).This function is based on the Newman Algorithm, which is recommended for identifying distinct groups within clusters (Carley et al. ; Clauset et al. ).
To label the identified groups, we started with sources at the center of a cluster, which have the highest number of links in a cluster.To validate our decisions, we discussed the labels with a number of experts.For additional feedback, the results were presented to several international conference and seminar audiences.
Network density was calculated as the number of edges divided by the number of possible edges not including self-references as follows: * number of edges/(number of nodes * (number of nodes-)) ( Iacobucci).One visible link corresponds mathematically to two edges given the bidirectionality of links.All the measurements of the groups are listed in Appendix A.
A closer look at the publication provides further evidence that homophily is relevant for both research issues.For opinion dynamics, the article describes ". . .homophily e ects in who we consider to be the relevant others in our organizational environment: those to whom we compare ourselves, those whose opinions we attend to. . ." (McPherson et al. , p. ).Homophily also e ects marriage formation because ". . . the homophily principle structures network ties of every type, including marriage . . ." (McPherson et al. , p. ).
Looking at Figure , one has to consider that there is an unbalanced number of co-citations from period ( -) to period ( -).This limitation of the previous study was discussed in Section .The above-average data set size in period increases the probability that more groups emerge.The greater a group, the more robust is its emergence to di erent sample sizes.Hence, regarding the smaller groups in period , the longitudinal comparison is limited.This is incorporated into the interpretation, which focusses on the main groups.

Figure :
Figure : Social simulation growth rate reflected in JASSS for years ( -) in comparison to other domains.

Figure :
Figure : Source age of cited publications in JASSS from

.
The NetLogo website became the most referenced source in JASSS with citations.In comparison to the first period, the relative citation value of NetLogo doubled to .%.To understand the reason for citing this source better, we investigated the context in which these NetLogo citations occurred.Twenty articles use NetLogo as a simulation platform for their simulation studies, and the remaining five articles have a clear methodological focus on agent-based modeling and related tools (Bersini ; Le Page et al.;Schwarz et al.  ; Thiele  et al.  , frequently cited sources.Note: Sources represented in both periods are highlighted in bold.Sources in gray boxes were among the most influential sources in Meyer et al. ( ).

Figure :
Figure : Types of sources cited in JASSS from

Figure :
Figure : Thomson Reuters subject fields and their impact on JASSS based on journal citations normalized to -.

Figure :
Figure : JASSS co-citation network from

Figure :
Figure : Link strength between the identified groups from

Figure :
Figure : Link strength between the identified groups from /

Figure :
Figure : Rank changes of social simulation research topics discussed in JASSS from

Table :
Frequency of citations in JASSS from to