An Analysis of Three Legal Citation Networks Derived from Austrian

Supreme Court Decisions

Markus Moser

and Mark Strembeck

1,2,3

Vienna University of Economics and Business (WU Vienna), Austria

Secure Business Austria Research Center (SBA), Austria

Complexity Science Hub Vienna (CSH), Austria

Keywords:

Austrian Supreme Court, Case Study, Court Decisions, Network Analysis.

Abstract:

In this paper, we present a case study on the structural properties of three citations networks derived from

Austrian Supreme Court decisions. In particular, we analyzed 250,984 Supreme Court decisions ranging from

1922 to 2017. As part of our case study, we analyzed the degree distributions, the structural properties of

prominent court decisions, as well as changes in the frequency of legal citations over time.

1 INTRODUCTION

Different studies analyze the structure of citation net-

works derived from court decisions. Most of those ex-

isting studies focused on common law court decisions

(esp. in the US) from legal systems where ”case law”

is signiﬁcantly more important than in the legal tradi-

tion of continental Europe

. In this study, we investi-

gate the citation network formed by the Austrian civil

law Supreme Court (i.e. from a legal system without

a case law tradition).

What sets this study apart from similar investiga-

tions is the large number of decisions available for

analysis (250,984) as well as the extensive time range

(from 1922 to 2017) that was considered for our anal-

ysis. Due to the vast number of court decisions, the

network generation procedure is fully automated and

does not rely on a few handpicked samples.

In Austria, Supreme Court decisions can either

reference/cite another court decision or a so called

legal proposition. Legal propositions are documents

which aggregate the major points of court decisions

that are similar in content. This leads to the ques-

In a legal system based on ”case law”, the law is ac-

tively developed through court decisions and courts base

their decisions on previous judicial verdicts. In this con-

text, the principle of ”stare decisis” demands that judges

must consider previous cases for their own decision. This is

fundamentally different from civil law jurisdiction as found

throughout continental Europe where case law and case ci-

tation are of lesser importance, and legislation remains the

ultimate source of the law.

tion of potential differences that arise between the ci-

tation networks that can be derived from these two

types of citations. Moreover, since legal propositions

must also refer to the court decisions they aggregate,

a third citation network can be constructed from the

outgoing links of legal propositions.

For this study, we investigated the following ques-

tions for the three citation networks mentioned above:

• Does the degree distribution of a network follow

a power-law distribution that often emerges in ci-

tation networks?

• Do the most prominent nodes of the network,

as established by in-degree centrality and Klein-

berg’s hub and authority score (Kleinberg, 1999),

also hold legal signiﬁcance?

• How do the citation networks evolve over time?

Which nodes receive the highest number of cita-

tions? How does the number of citations vary be-

tween time periods? How do structural properties

of the networks develop over time?

2 RELATED WORK

The structural properties of citation networks derived

from court decisions have been studied in several le-

gal domains, most prominently in the US legal sys-

tem. For example, (Chandler, 2005) studied the ci-

tation network of the United States Supreme Court.

Moreover, (Fowler and Jeon, 2008) and (Fowler et al.,

Moser, M. and Strembeck, M.

An Analysis of Three Legal Citation Networks Derived from Austrian Supreme Court Decisions.

DOI: 10.5220/0007749900850092

In Proceedings of the 4th International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2019), pages 85-92

ISBN: 978-989-758-366-7

2007) identiﬁed the most central decisions of the

United States Supreme Court. They point out the

value of Kleinberg’s hub and authority score for de-

termining the most relevant cases in the citation net-

work. It allows to identify key cases that are inﬂu-

encial (authority score) and well founded in law (hub

score). Therefore, Kleinberg’s score is assumed to be

superior for analyzing networks of court decisions in

comparison to other traditional measures of network

centrality. In another study (Smith, 2007), the citation

network resulting from US Supreme Court decisions

was found to follow a power-law degree distribution.

(Agnoloni and Pagallo, 2015) studied the citation

network of the Italian Constitutional Court where they

also found a power-law degree distribution and con-

ﬁrm the ﬁnding that high hub and authority scores

are often correlated to the most debated cases in Ital-

ian legal journals. Similar studies exist for other le-

gal systems. For example, (Winkels et al., 2011)

studied the citation network of the Dutch Supreme

Court, (Mazzega et al., 2009) investigate citations in

the French legal system, (Tarissan et al., 2016) and

(Lettieri et al., 2016) decisions of the European Union

Court of Justice. In addition, (Gelter and Siems,

2012) investigated cross-citations between ten of Eu-

rope’s highest courts.

(Koniaris et al., 2017) analyze EU legislation and

create a multi-relational network encompassing the

hierarchy between different sources of primary and

secondary EU law and different types of relationships

between legal documents. They also perform a re-

silience test in order to predict the behavior of the

network under malfunctions when legal documents

or connections between them are severed, simulating

that existing law can be amended or removed.

3 DATA EXTRACTION AND

NETWORK DERIVATION

Figure 1: Data extraction and analysis.

Figure 1 shows an overview of the data extraction and

analysis procedure that we applied for our case study.

In particular, Austrian Supreme Court decisions are

accessible via a publicly available database called ”le-

gal information system” (German: RechtsInforma-

tionsSystem, abbreviated RIS)

. For automated data

retrieval, the court decisions are part of the ”Open

Government Data” initiative and accessible via a va-

riety of interfaces, supporting formats such as JSON,

XML and HTML.

For our case study, we used the JSON-based in-

terface which returns a list of all available decisions

for a given time range. For each result, this data also

contains links to the full text of the decisions in XML

format. This way, we retrieved all Supreme Court de-

cisions from 1922 to the end of 2017, resulting in a

total of 250,984 ﬁles. Each ﬁle represents a distinct

decision from the ﬁeld of civil law or criminal law or

a legal proposition which aggregates and summarizes

a major point of two or more decisions.

The XML ﬁles retrieved in the previous step were

used for building and analyzing directed graphs of ci-

tations using the R language and the igraph package

For each XML ﬁle with at least one citation, a node is

added to the citation graph. For every outgoing cita-

tion an edge is created to the node that is being cited.

In addition, the respective decision dates are stored as

node attributes. The data retrieved from the RIS al-

low to distinguish between decisions and legal propo-

sitions, but do not allow to infer precisely to which

legal matter the ﬁle belongs. However, not all deci-

sions are available as full text. This especially applies

to court decisions which were published before the

RIS became available. Thus, if no XML ﬁle for a

court decision exists, the decision can only be cited,

but cannot have outgoing links and attributes in our

network. In particular, we derived the following net-

works from the data:

• Network A: court decisions citing other court de-

cisions by their unique reference number;

• Network B: decisions citing legal propositions by

their unique identiﬁer;

• Network C: legal propositions citing decisions by

their reference number.

Network A (Reference Number Citing Refer-

ence Number): every Supreme Court decision has

at least one unique reference number (German:

Gesch

aftszahl, abbreviated GZ). A reference num-

ber encodes basic information about the decision in

a systematic manner. For example, the reference

number 3 Ob 646/79 informs us about the decid-

ing senate (e.g. 3), about the deciding court (e.g. O

for Supreme Court), the type of legal matter (e.g. b

for civil law), an incremental consecutive numbering

(e.g. 646) and the year in which the case was received

https://www.ris.bka.gv.at/

igraph.org

COMPLEXIS 2019 - 4th International Conference on Complexity, Future Information Systems and Risk

Figure 2: in-degree (a) and out-degree (b) distribution of network A.

by the Supreme Court (e.g. 1979). Note that this is

not necessarily the same year as the decision date.

In our data-set, 123,222 of the 250,984 XML doc-

uments have a reference number and thus can be clas-

siﬁed as court decisions. However, only 60,758 of

these decisions have outgoing citations to other deci-

sions. Based on this data, we derived a directed graph

consisting of 96,444 nodes and 188.024 edges repre-

senting citations between the nodes. For 16,122 of the

nodes (16.7%) that have at least one incoming edge

(i.e. they are cited at least once) the RIS does not pro-

vide a corresponding XML ﬁle. Thus, the correspond-

ing nodes can only have incoming citations (incoming

edges) and no attributes.

The largest connected component in Network A

consists of 86,247 out of 96,444 nodes (89%). The

network is not fragmented into large disconnected

components, which indicates that citations occur be-

tween different legal areas such as criminal law and

civil law.

Network B (Reference Number Citing Legal

Proposition): So-called ”legal propositions” (Ger-

man: Rechtss

atze, RS) are artiﬁcially created hub

nodes pointing to several court decisions that are sim-

ilar in content. These documents are created and

maintained by the ofﬁce of records (German: Evi-

denzb

uro) of the Austrian Supreme Court

and made

available in the RIS. They are updated and amended

if new decisions appear that ﬁt the ofﬁce’s similarity

requirements. In our data-set, 30,242 of the nodes are

RS and 41,494 nodes are court decisions referencing

those RS. From those nodes, we derived a directed

bipartite citation network consisting of 71,736 nodes

and 167,584 edges.

http://www.ogh.gv.at/service/evidenzbuero/

Network C (Legal Proposition Citing Reference

Number): Legal propositions are not only cited in

court decisions, but are linking to decisions them-

selves. Thus, legal propositions have outgoing links

to at least one decision whose content they summa-

rize. Note, however, that legal propositions cannot

point to other legal propositions.

From our data-set, we derived a bipartite net-

work consisting of 300,480 nodes and 565,814 edges,

with 176,564 nodes representing court decisions and

123,916 nodes representing legal propositions. For

Network C, 76.186 (25% of all nodes) of the ref-

erence numbers that are cited by some other node

are not available as XML ﬁles. In all three net-

works, we found two major clusters consisting of

civil law decisions/propositions and criminal law de-

cisions/propositions respectively.

4 STRUCTURAL PROPERTIES

In order to investigate the structural properties of the

three networks, we will analyze the respective degree

distributions, identify the most prominent nodes, and

look at the temporal changes in the citation links.

Different studies found that citation networks de-

rived from scientiﬁc publications often show a power-

law degree distribution, see, e.g., (Price, 1965), (Sila-

gadze, 1999) and (Redner, 1998). However, since it is

not trivial to distinguish power-law distributions from

other types of (heavy-tailed) distributions (Stumpf

and Porter, 2012), we applied the procedure sug-

gested by (Clauset et al., 2009) in order to determine

the degree distributions of the three citation networks

we derived. In particular, the procedure includes a

An Analysis of Three Legal Citation Networks Derived from Austrian Supreme Court Decisions

Figure 3: incoming (a) and outgoing (b) citations of network A by year.

goodness-of-ﬁt test for different types of distributions

(exponential, lognormal, Poisson, power-law), as well

as a direct comparision of the different distributions

via Vuong’s test (Vuong, 1989). The corresponding

computations have been conducted with the poweR-

law package (Gillespie, 2015).

4.1 Network A

The goodness-of-ﬁt test for the in-degree distribu-

tion of Network A resulted in a p-value of 0.215 for

a discrete power-law distribution and p-values close

to zero for all other distributions. Direct compar-

isons between the distributions using Vuong’s Test

(Vuong, 1989) conﬁrm these ﬁndings and show that

the power-law distribution is clearly a better ﬁt than

exponential or Poisson distributions. However, a di-

rect comparison with the lognormal distribution is in-

decisive. Thus, the analysis indicates that there is

a good chance the true in-degree distribution has a

heavy-tail. For the out-degree distribution we found

similar results, with the goodness-of-ﬁt tests resulting

in p-values of 0.318 for a power-law distribution and

0.279 for a lognormal distribution. For the out-degree

distribution, the direct comparisons slightly favored

the lognormal distribution over the power-law distri-

bution (see also Figure 2).

The decision with the highest number of incom-

ing citations (130) is 5Ob93/97f from 1997 which is

a short decision addressing an issue related to proce-

dural law. Note that 129 of the citations are also from

1997 and only one citation is from 1999. The sec-

ond highest number of incoming citations is 102 for

5Ob115/97s (right to build) which is also from 1997,

followed by 5Ob190/97w (also right to build) with 96

incoming citations. Surprisingly, nodes with a high

degree of incoming citations seem to receive citations

only for a short period of time. Thus the in-degree of a

node may not be a good indicator of long-term signif-

icance. An attempt to explain these ﬁndings could be

a citation cascade triggered by a decision that affected

many follow-up decisions of that time (a pattern that

could be expected in procedural law). The results also

underscore the lower signiﬁcance of prior decisions

in civil law courts compared to common law courts.

Figure 3 shows a barplot for the temporal evolution of

the distribution for incoming and outgoing citiations

in Network A.

Calculating the authority scores of the network

gives very similar results. The ten nodes with the

highest in-degree and the ten nodes with the high-

est authority score have seven nodes in common. For

outgoing citations, we ﬁnd that decision 2Ob215/10x

(tenancy law) from 2012 has 60 outgoing citations,

followed by 13Os55/13g (criminal law) with 54 out-

going citations.

The ”decision date” attribute determines the date

when a court decision was made. For legal proposi-

tions, it refers to the decision date of the oldest de-

cision included in this legal proposition. Figure 3 a)

shows which years are cited most often. Intuitively

one could probably assume that comparatively old de-

cisions (before 1980) had more time to gather cita-

tions and are therefore cited more often than recent

decisions. However, Figure 3 a) shows that this is

not the case. Instead, citations peak around a speciﬁc

time period from the mid-1990s to the ﬁrst half of the

2000s with a gradual increase for the early 1980s and

a gradual decrease in the second half of the 2000s.

The number of citations for the years after 2010 re-

COMPLEXIS 2019 - 4th International Conference on Complexity, Future Information Systems and Risk

Figure 4: incoming (a) and outgoing (b) citations of network B by year.

mains high but is steeply declining, most likely be-

cause recent decisions have not been available long

enough to gather a high amount of citations.

It seems that the ability of a decision to attract new

citations declines with time, as they lose their rele-

vance and are superseded by newer decisions. This

interpretation seems meaningful since jurisdiction is

dynamically evolving and changes that occur in legis-

lation make citing older decisions less useful.

Figure 3 b) shows a sharp increase during the mid-

1980s, followed by another spike in the late-1990s.

Since then, the number of decisions remains high.

The low number of older decisions can to some extend

be explained by the incompleteness of the data. Older

decisions might not yet have been added to the RIS

database and might partially account for the 16.7% of

nodes that are cited but do not have XML ﬁles (see

above). Another hypothetical reason for the increas-

ing total number of citations might be the availability

of electronic systems for information retrieval. The

RIS has been available since 1998 and might have fa-

cilitated the research for similar decisions.

4.2 Network B

With regard to the degree distribution of Network B,

one would probably assume that the patterns and re-

lations discovered for network A also apply to the ci-

tations of legal propositions. However, neither the in-

degree nor out-degree distribution show evidence for

heavy-tailed distributions.

In total, 30,242 nodes in our data-set represent le-

gal propositions (42% of all nodes) and have incom-

ing citations. The goodness-of-ﬁt test (Clauset et al.,

2009) for different distributions (exponential, Pois-

son, power-law, lognormal) showed that neither of the

tested distributions provides a good ﬁt for our data-set

(the highest p-value being 0.048 for power-law dis-

tribution, while all other distributions yield p-values

close to 0). A direct comparison (Clauset et al., 2009)

between the power-law distribution and other distri-

butions are inconclusive, also indicating that neither

distribution provides a good ﬁt for the data.

For the out-degree distribution of network B we

analyzed 41,494 nodes. The goodness-of-ﬁt test re-

turned a p-value of 0.738 for a lognormal distribution

followed by a p-value of 0.539 for a Poisson distri-

bution, whereas the tests for the power-law and ex-

ponential distributions return values close to 0. How-

ever, with an x

min

value of 48 only the far-right tail of

the data provides a ﬁt for Poisson distribution. The

estimated lognormal distribution with an x

min

of 7

clearly provides a better ﬁt, which is also conﬁrmed

by direct comparisons.

In general, legal propositions receive a higher

amount of citations than an individual court decision.

The node with the highest amount of citations in Net-

work C is RS0099810 (criminal procedural law) and

has 1147 citations, followed by RS0042963 (civil pro-

cedural law) with 971 citations. Moreover, the most

cited RS have remained popular over a longer period

of time. For example, RS0099810 accumulated its

citations from 2004 to 2017 and RS0042963 is cited

from 1998 to 2017.

Nodes with a high in-degree also tend to achieve

a high authority score. The top ten nodes for both

scores have 4 nodes in common. However, the au-

thority score might still not be a very useful mea-

sure for Network B since it only connects nodes of

one type (RS) with nodes of another type (court de-

An Analysis of Three Legal Citation Networks Derived from Austrian Supreme Court Decisions

Figure 5: incoming (a) and outgoing (b) citations of network C by year.

cisions). The court decision with the highest number

of distinct references to RS is 13Os105/15p (criminal

law) which has 73 citations followed by 13Os55/13g

(criminal law) with 62 citations. The average number

of outgoing citations to RS is 4.

Even though RS are only in use since 1996, the

in-degree and authority score seem to be better pre-

dictors for node signiﬁcance for RS than for individ-

ual court decisions. The most prominent RS nodes

remain popular and continue to accumulate citations

over considerably longer time periods. In contrast, the

most prominent nodes for direct citations of decisions

are the result of brief spikes (see above).

Figure 4 shows the distributions citations in Net-

work B. Figure 4 a) covers the entire time period ana-

lyzed for this paper. This is because the date of an RS

is the date the oldest court decision the RS refers to.

The resulting plot is similar to the out-degree plot of

network C in Figure 5 and has similar characteristics.

Most notably, while some RS include very old deci-

sions, many new RS refer to recent decisions only.

Figure 4 b) shows the absolute numbers for deci-

sions citing RS per year. The plot shows a growing

tendency for RS to be cited. In absolute numbers, cit-

ing individual court decisions is more common than

citing RS though.

4.3 Network C

Legal propositions aggregate the results of similar de-

cisions into summary statements. Those summaries

are listed separately by reference number. However,

these references can directly be retrieved via the cor-

responding tag in the respective XML ﬁle. Network

C consists of 123,916 nodes (41% of all nodes) clas-

siﬁed as legal propositions and 176.564 nodes (59%

of all nodes) classiﬁed as individual court decisions.

For the in-degree distribution of incoming cita-

tions for court decisions, the goodness-of-ﬁt tests

(Clauset et al., 2009) found that both a power-law and

a lognormal distribution are valid hypotheses for the

data with p-values of 0.74 for power-law and 0.50 for

log-normal respectively (with the p-values for Pois-

son and exponential distributions being 0).

The direct comparison (Clauset et al., 2009) of the

power-law and lognormal distributions again indicate

that both models potentially ﬁt. However, the esti-

mated lognormal distribution with an x

min

value of

4 seems to be a slightly better ﬁt than the estimated

power-law distribution with an x

min

value of 21

For the out-degree distribution of legal proposi-

tions to individual court decisions, the results are dif-

ferent though. Here the goodness-of-ﬁt test clearly fa-

vors a lognormal distribution (with a p-value of 0.42

and an x

in value of 7, while all other distributions

have a p-value close to 0).

When investigating the court decisions with the

highest amount of incoming citations from legal

propositions, we ﬁnd 2Ob215/10x from 2012 (ten-

ancy law) is referenced in 96 RS and 2Ob1/09z

from 2010 (leasing agreements) in 77, followed by

15Os42/92 (criminal law). In total, 12 court decisions

have an in-degree greater than 50, belonging to a time

period from the 1980s to 2016. Thus, the in-degree of

an individual court decision seems to be a meaning-

ful predictor for node importance in Network C. On

average, court decisions are cited by 3.2 RS.

The legal proposition with the highest out-degree

is RS0042963 (civil procedural law), which refers to

583 decisions, followed by RS0043758 (civil proce-

COMPLEXIS 2019 - 4th International Conference on Complexity, Future Information Systems and Risk

dural law) referring to 446 decisions. On average, a

RS aggregates its content from 4.5 decisions. Note

that RS that are prominent in network B have a signif-

icantly higher amount of outgoing links than average

(between 69 and 583 links).

However, the ten most relevant nodes according

to the authority score are disjoint from the ten nodes

with the highest in-degree. This ﬁnding might be ex-

plained by the different nature of Network C, since

nodes are not strictly cited but summarized in RS with

similar content. Similar to network B, it is bipartite,

which might, again, impact the validity of the author-

ity score as a predictor for importance of a node.

The node with the highest authority score is

1Ob258/11i (civil procedural law, immission reduc-

tion), followed by 1Ob202/13g (civil procedural law),

two decisions (5Ob7/74, 5Ob8/74) that are not avail-

able as full text and 10ObS100/11w (social law).

Figure 5 shows the incoming and outgoing cita-

tions for Network C. The lower numbers for 2017

indicate that not all decisions of this year have been

processed and aggregated into RS yet. Also note that

our data-set contains about 25% of nodes that are not

available as XML ﬁles and thus cannot be part of this

analysis since they have no decision date.

The distribution in Figure 5 b) shows that RS

pointing to reference numbers are spread over the

whole time period under investigation. We also notice

that most RS do not include very old decisions (i.e.

court decisions that have been made before 1950).

5 LIMITATIONS AND OUTLOOK

The case study discussed in this paper, only takes into

account citations by reference number (i.e. individ-

ual court decisions) or by legal propositions. How-

ever, other types of citations are also possible, such

as references by collection number for example. Fur-

thermore, the data that we analyzed was limited to

the data provided by the RIS database which does not

(yet) offer the complete set of all decisions since 1922

and tends to omit older decisions from the time before

the RIS was established.

Moreover, so far we only analyzed structural prop-

erties of the corresponding citation networks. In our

future work, we plan to investigate the semantic pur-

pose of the different citations. In addition, we did not

make a distinction between the different ﬁelds of law.

This is because the metadata of the XML ﬁles do not

contain sufﬁcient information to determine the exact

ﬁeld of law a court decision belongs to. A very broad

distinction could be established in future research by

looking at the type of legal matter as encoded in the

reference number which could be used to determine if

a decision belongs to criminal law or civil law. For a

detailed analysis, a ﬁner level of distinction would be

required though.

The analyses of Network A indicate that the num-

ber of incoming citations alone is a poor predictor for

the (legal) signiﬁcance of a node. In the future, this

could be addressed by taking the time-span into ac-

count where nodes received citations.

6 CONCLUSION

In this paper, we presented a case study on a structural

analysis of three legal citation networks derived from

Austrian Supreme Court decisions.

The case study showed that the popularity of ci-

tations referring to individual court decisions decays

with time and very old nodes are hardly cited at all, in-

dicating a dynamically evolving jurisdiction that does

not rely on direct citations of landmark cases.

Moreover, court decisions do not tend to accu-

mulate many citations over a longer time period, the

most-cited nodes gained their citations in brief time

periods of two to three years, indicating a fast obsoles-

cence of popularity in decisions. In our future work,

the detection of node importance could be improved

by taking the time period into account in which a node

received citations.

In contrast, legal propositions (RS) that aggregate

court decisions are cited more often by subsequent

court decisions and tend to accumulate citations over

a longer period of time. The outgoing citations of RS

show that recent decisions are favored over older ones

and only a small fraction of RS include at least one

link to an old decision (before 1950).

The popularity of RS prospectively results from

the fact that citing a single RS allows referencing a

number of related court decisions at once. In addition,

it is easier to look up a single RS than a larger number

of individual decisions. Future research could address

the question if and under which conditions the popu-

larity of RS declines over time or how the popularity

of a RS changes if a new decision is added to it.

Furthermore, our case study found that prominent

RS aggregate a higher number of decisions than the

average RS. Referencing many decisions thus seems

to be an indicator of RS importance, even though the

RS with the highest number of references to decisions

are not necessarily the most-cited RS.

Moreover, it should be noted that the tempo-

ral evolution of citations is likely to be affected by

the availability of information technology and infor-

mation retrieval systems, which signiﬁcantly facili-

An Analysis of Three Legal Citation Networks Derived from Austrian Supreme Court Decisions

tate research for related court decisions, leading to a

greater overall amount of citations.

The results of this study might help to develop in-

teractive visualization tools or recommender systems

to help legal professionals navigating related material

for a particular legal matter.

REFERENCES

Agnoloni, T. and Pagallo, U. (2015). The case law of

the Italian constitutional court, its power laws, and

the web of scholarly opinions. In Proceedings of

the 15th International Conference on Artiﬁcial Intelli-

gence and Law, pages 151–155. ACM.

Chandler, S. (2005). The network structure of Supreme

Court jurisprudence. University of Houston Law Cen-

ter No. 2005-W-01.

Clauset, A., Shalizi, C., and Newman, M. (2009). Power-

law distributions in empirical data. SIAM Review.

Fowler, J. H. and Jeon, S. (2008). The authority of Supreme

Court precedent. Social networks, 30(1):16–30.

Fowler, J. H., Johnson, T. R., Spriggs, J. F., Jeon, S., and

Wahlbeck, P. J. (2007). Network analysis and the law:

Measuring the legal importance of precedents at the

US Supreme Court. Political Analysis, 15(3):324–

346.

Gelter, M. and Siems, M. (2012). Networks, dialogue

or one-way trafﬁc: An empirical analysis of cross-

citations between ten of europe’s highest courts.

Utrecht L. Rev., 8:88.

Gillespie, C. (2015). Fitting Heavy Tailed Distributions:

The poweRlaw Package. Journal of Statistical Soft-

ware, 64(2).

Kleinberg, J. M. (1999). Authoritative sources in a hy-

perlinked environment. Journal of the ACM (JACM),

46(5):604–632.

Koniaris, M., Anagnostopoulos, I., and Vassiliou, Y. (2017).

Network analysis in the legal domain: A complex

model for European Union legal sources. Journal of

Complex Networks, 6(2):243–268.

Lettieri, N., Altamura, A., Faggiano, A., and Malandrino,

D. (2016). A computational approach for the experi-

mental study of EU case law: analysis and implemen-

tation. Social Network Analysis and Mining, 6(1):56.

Mazzega, P., Bourcier, D., and Boulet, R. (2009). The net-

work of French legal codes. In Proceedings of the 12th

international conference on artiﬁcial intelligence and

law, pages 236–237. ACM.

Price, D. J. D. S. (1965). Networks of scientiﬁc papers.

Science, pages 510–515.

Redner, S. (1998). How popular is your paper? An empir-

ical study of the citation distribution. The European

Physical Journal B - Condensed Matter and Complex

Systems, 4(2):131–134.

Silagadze, Z. (1999). Citations and the Zipf-Mandelbrot’s

law. arXiv preprint physics/9901035.

Smith, T. A. (2007). The web of law. San Diego L. Rev.,

44:309.

Stumpf, M. P. and Porter, M. A. (2012). Critical truths about

power laws. Science, 335(6069):665–666.

Tarissan, F., Panagis, Y., and

Sadl, U. (2016). Selecting

the cases that deﬁned Europe: complementary metrics

for a network analysis. In Proceedings of the 2016

IEEE/ACM International Conference on Advances in

Social Networks Analysis and Mining, pages 661–668.

IEEE Press.

Vuong, Q. H. (1989). Likelihood ratio tests for model se-

lection and non-nested hypotheses. Econometrica,

57(2):307–33.

Winkels, R., Ruyter, J., and Kroese, H. (2011). Determining

authority of Dutch case law. Legal Knowledge and

Information Systems, Vol. 235, pp. 103-112, 2011.

COMPLEXIS 2019 - 4th International Conference on Complexity, Future Information Systems and Risk