Clustering statistics journals

This document hopes to be a useful benchmark for a “toy” dataset, namely the network of 47 statistics journals studied in the paper Varin, C., Cattelan, M. and Firth, D. (2016), “Statistical modelling of citation exchange between statistics journals”. J. R. Stat. Soc. A, 179: 1–63. doi:10.1111/rssa.12124.

It isn’t always clear how the various implementations of community detection algorithms handle cases of directed and undirected graphs and the difference between weighted edges or multiple unweighted edges. Therefore, in this document we will try each of the four combinations just to see what happens. Where igraph algorithms are not implemented for weighted or directed graphs, they will throw an error, which we have reproduced under the respective tabs. Beware: just because the algorithm does not throw an error does not necessarily mean that directions or weights are actually taken into account.

Self-citations have not been explicitly removed before performing any of the procedures below (though some algorithms may implicitly or explicitly ignore them). I might try removing self-citations later to see if it makes any noticeable difference to the results.

Also to do: nice minimal plots so we don’t have to read lists/tables all day. Update: scroll down to see some visualisations.

Infomap algorithm

Directed multi-edge

1. ANZS, Bcs, CSSC, CSTM, EES, Envr, JAS, JBS, JRSS-B, JSCS, JSS, StMed, StNee, StPap, SPL
2. AmS, Bka, Biost, CSDA, JCGS, JRSS-A, JSPI, Mtka, SJS, StataJ, StSci
3. BioJ, CJS, JASA, JNS, JRSS-C, LDA, StCmp, Stats, SMMR, StSin, Tech, Test
4. AISM, AoS, Bern, CmpSt, ISR, JABES, JMA, JTSA, StMod

Directed weighted

1. AmS, AISM, AoS, ANZS, Bern, BioJ, Bcs, Bka, Biost, CJS, CSSC, CSTM, CmpSt, CSDA, EES, Envr, ISR, JABES, JASA, JAS, JBS, JCGS, JMA, JNS, JRSS-A, JRSS-B, JRSS-C, JSCS, JSPI, JSS, JTSA, LDA, Mtka, SJS, StataJ, StCmp, Stats, StMed, SMMR, StMod, StNee, StPap, SPL, StSci, StSin, Tech, Test

Undirected multi-edge

1. AoS, StMed
2. JASA, StSci
3. Bka, JMA
4. CSDA, Test
5. JNS, JSPI
6. Bern, Bcs
7. Biost, CSTM
8. JAS, SPL
9. JRSS-A, JRSS-B
10. JRSS-C, StSin
11. BioJ, JSCS
12. LDA, SJS
13. AISM, Mtka, StNee
14. CmpSt, JCGS
15. CSSC, JTSA
16. CJS, SMMR
17. JBS, Tech
18. StCmp, StMod
19. Envr, JSS
20. Stats, StPap
21. AmS, EES
22. ANZS, JABES
23. ISR, StataJ

Undirected weighted

1. AmS, AISM, AoS, ANZS, Bern, BioJ, Bcs, Bka, Biost, CJS, CSSC, CSTM, CmpSt, CSDA, EES, Envr, ISR, JABES, JASA, JAS, JBS, JCGS, JMA, JNS, JRSS-A, JRSS-B, JRSS-C, JSCS, JSPI, JSS, JTSA, LDA, Mtka, SJS, StataJ, StCmp, Stats, StMed, SMMR, StMod, StNee, StPap, SPL, StSci, StSin, Tech, Test

Agglomerative hierarchical clustering

Directed, complete-linkage

1. AmS
2. AISM, JTSA, Stats, SPL
3. AoS, Bern, CJS, JMA, JNS
4. ANZS, JSPI, Mtka, Tech, Test
5. BioJ, JBS
6. Bcs, Biost, JRSS-A, JRSS-C, LDA, StMed, SMMR
7. Bka, JASA, JRSS-B, SJS, StSin
8. CSSC, CSTM, JAS, JSCS, StPap
9. CmpSt, CSDA, JCGS, StCmp
10. EES
11. Envr, JABES
12. ISR
13. JSS
14. StataJ
15. StMod, StSci
16. StNee

Undirected, complete-linkage

1. AmS, ISR
2. AISM, ANZS, CSTM, JSPI, JTSA, Mtka, Stats, StPap, SPL
3. AoS, Bern, Bka, CJS, JASA, JCGS, JMA, JNS, JRSS-B, SJS, StCmp, StNee, StSin, Test
4. BioJ, Bcs, Biost, JBS, JRSS-A, JRSS-C, LDA, StMed, SMMR, StMod, StSci
5. CSSC, CmpSt, CSDA, JAS, JSCS, Tech
6. EES, Envr, JABES
7. JSS
8. StataJ

Hierarchical clustering with complete linkage, applied to the undirected Pearson correlation matrix, should exactly reproduce the results of Varin et al (2016).

Edge betweenness

Directed multi-edge

1. AmS
2. AISM
3. AoS, JMA, StSin
4. ANZS
5. Bern
6. BioJ
7. Bcs, StMed
8. Bka
9. Biost
10. CJS
11. CSSC
12. CSTM
13. CmpSt
14. CSDA, JASA
15. EES
16. Envr
17. ISR
18. JABES
19. JAS
20. JBS
21. JCGS
22. JNS
23. JRSS-A
24. JRSS-B
25. JRSS-C
26. JSCS
27. JSPI
28. JSS
29. JTSA
30. LDA
31. Mtka
32. SJS
33. StataJ
34. StCmp
35. Stats
36. SMMR
37. StMod
38. StNee
39. StPap
40. SPL
41. StSci
42. Tech
43. Test

Directed weighted

1. AmS
2. AISM
3. AoS
4. ANZS
5. Bern
6. BioJ
7. Bcs
8. Bka
9. Biost
10. CJS
11. CSSC
12. CSTM
13. CmpSt
14. CSDA
15. EES
16. Envr
17. ISR
18. JABES
19. JASA
20. JAS
21. JBS
22. JCGS
23. JMA, JNS, JRSS-B, JSPI, SJS, StCmp, SPL, StSin, Tech, Test
24. JRSS-A
25. JRSS-C
26. JSCS
27. JSS
28. JTSA
29. LDA
30. Mtka
31. StataJ
32. Stats
33. StMed
34. SMMR
35. StMod
36. StNee
37. StPap
38. StSci

Undirected multi-edge

1. AmS
2. AISM
3. AoS, Bka, CSDA, JASA, JMA, JRSS-B, JSPI, StSin
4. ANZS
5. Bern
6. BioJ
7. Bcs, StMed
8. Biost
9. CJS
10. CSSC
11. CSTM
12. CmpSt
13. EES
14. Envr
15. ISR
16. JABES
17. JAS
18. JBS
19. JCGS
20. JNS
21. JRSS-A
22. JRSS-C
23. JSCS
24. JSS
25. JTSA
26. LDA
27. Mtka
28. SJS
29. StataJ
30. StCmp
31. Stats
32. SMMR
33. StMod
34. StNee
35. StPap
36. SPL
37. StSci
38. Tech
39. Test

Undirected weighted

1. AmS
2. AISM
3. AoS
4. ANZS
5. Bern
6. BioJ
7. Bcs
8. Bka
9. Biost
10. CJS
11. CSSC
12. CSTM
13. CmpSt
14. CSDA
15. EES
16. Envr
17. ISR
18. JABES
19. JASA
20. JAS
21. JBS
22. JCGS, StCmp, StSci
23. JMA
24. JNS
25. JRSS-A
26. JRSS-B, SMMR
27. JRSS-C
28. JSCS, Mtka, Stats, StPap, SPL
29. JSPI
30. JSS
31. JTSA
32. LDA, StSin
33. SJS
34. StataJ
35. StMed
36. StMod
37. StNee
38. Tech
39. Test

Modularity optimisation: fast and greedy

Directed multi-edge

Looks like an error! Here is the message:
At fast_community.c:538 : fast greedy community detection works for undirected graphs only, Unimplemented function call

Directed weighted

Looks like an error! Here is the message:
At fast_community.c:538 : fast greedy community detection works for undirected graphs only, Unimplemented function call

Undirected multi-edge

Looks like an error! Here is the message:
At fast_community.c:553 : fast-greedy community finding works only on graphs without multiple edges, Invalid value

Undirected weighted

1. AISM, ANZS, CSSC, CSTM, JAS, JMA, JNS, JSCS, JSPI, JTSA, Mtka, Stats, StNee, StPap, SPL, Tech
2. CmpSt, CSDA, JSS
3. AmS, ISR
4. EES, Envr, JABES
5. AoS, Bern, Bka, CJS, JASA, JCGS, JRSS-B, SJS, StCmp, StSin, Test
6. BioJ, Bcs, Biost, JBS, JRSS-A, JRSS-C, LDA, StMed, SMMR, StMod, StSci
7. StataJ

Modularity optimisation: Louvain method (Blondel et al)

Directed multi-edge

Looks like an error! Here is the message:
At community.c:2672 : multi-level community detection works for undirected graphs only, Unimplemented function call

Directed weighted

Looks like an error! Here is the message:
At community.c:2672 : multi-level community detection works for undirected graphs only, Unimplemented function call

Undirected multi-edge

1. EES, Envr, JABES
2. CmpSt, CSDA, JSS
3. AISM, ANZS, CSSC, CSTM, JAS, JMA, JNS, JSCS, JSPI, JTSA, Mtka, Stats, StNee, StPap, SPL, Tech
4. StataJ
5. BioJ, Bcs, Biost, JBS, JRSS-A, JRSS-C, LDA, StMed, SMMR, StMod, StSci
6. AmS, ISR
7. AoS, Bern, Bka, CJS, JASA, JCGS, JRSS-B, SJS, StCmp, StSin, Test

Undirected weighted

1. EES, Envr, JABES
2. CmpSt, CSDA, JSS
3. AISM, ANZS, CSSC, CSTM, JAS, JMA, JNS, JSCS, JSPI, JTSA, Mtka, Stats, StNee, StPap, SPL, Tech
4. StataJ
5. BioJ, Bcs, Biost, JBS, JRSS-A, JRSS-C, LDA, StMed, SMMR, StMod, StSci
6. AmS, ISR
7. AoS, Bern, Bka, CJS, JASA, JCGS, JRSS-B, SJS, StCmp, StSin, Test

Spinglass method

Directed multi-edge

1. ANZS, BioJ, CmpSt, Envr, JABES, JAS, JCGS, JSS, StCmp, StSci, Tech
2. AISM, Bern, CJS, CSSC, CSTM, JASA, JNS, JSCS, JTSA, Mtka, SJS, Stats, StPap, SPL, StSin, Test
3. AmS, AoS, Bcs, Bka, Biost, CSDA, EES, ISR, JBS, JMA, JRSS-A, JRSS-B, JRSS-C, JSPI, LDA, StataJ, StMed, SMMR, StMod, StNee

Directed weighted

1. AmS, BioJ, EES, ISR, JAS, JBS, JRSS-A, JRSS-C, JSS, LDA, SJS, StataJ, StMed, SMMR, StMod, StSci
2. AISM, CSSC, CSTM, CmpSt, CSDA, JMA, JNS, JSCS, JSPI, JTSA, Mtka, Stats, StPap, SPL, StSin, Tech, Test
3. AoS, ANZS, Bern, Bcs, Bka, Biost, CJS, Envr, JABES, JASA, JCGS, JRSS-B, StCmp, StNee

Undirected multi-edge

1. ANZS, BioJ, CmpSt, Envr, JABES, JAS, JCGS, JSS, StCmp, StSci, Tech
2. AISM, Bern, CJS, CSSC, CSTM, JASA, JNS, JSCS, JTSA, Mtka, SJS, Stats, StPap, SPL, StSin, Test
3. AmS, AoS, Bcs, Bka, Biost, CSDA, EES, ISR, JBS, JMA, JRSS-A, JRSS-B, JRSS-C, JSPI, LDA, StataJ, StMed, SMMR, StMod, StNee

Undirected weighted

1. AISM, ANZS, CSSC, CSTM, CmpSt, CSDA, ISR, JAS, JMA, JNS, JSCS, JSPI, JTSA, Mtka, Stats, StNee, StPap, SPL, Tech
2. AoS, Bern, Bka, CJS, Envr, JASA, JCGS, JRSS-B, SJS, StCmp, StSin, Test
3. AmS, BioJ, Bcs, Biost, EES, JABES, JBS, JRSS-A, JRSS-C, JSS, LDA, StataJ, StMed, SMMR, StMod, StSci

Random walks (walktrap)

Based on random walks of 6 steps.

See Pascal Pons, Matthieu Latapy: Computing communities in large networks using random walks, http://arxiv.org/abs/physics/0512106

Directed multi-edge

1. AmS, BioJ, Bcs, Biost, EES, JABES, JBS, JRSS-A, JRSS-C, JSS, LDA, StMed, SMMR, StMod, StSci
2. AISM, AoS, ANZS, Bern, Bka, CJS, CSSC, CSTM, CmpSt, CSDA, Envr, ISR, JASA, JAS, JCGS, JMA, JNS, JRSS-B, JSCS, JSPI, JTSA, Mtka, SJS, StCmp, Stats, StNee, StPap, SPL, StSin, Tech, Test
3. StataJ

Directed weighted

1. AmS, BioJ, Bcs, Biost, EES, JABES, JBS, JRSS-A, JRSS-C, JSS, LDA, StMed, SMMR, StMod, StSci
2. AISM, AoS, ANZS, Bern, Bka, CJS, CSSC, CSTM, CmpSt, CSDA, Envr, ISR, JASA, JAS, JCGS, JMA, JNS, JRSS-B, JSCS, JSPI, JTSA, Mtka, SJS, StCmp, Stats, StNee, StPap, SPL, StSin, Tech, Test
3. StataJ

Undirected multi-edge

1. AmS, BioJ, Bcs, Biost, EES, JABES, JBS, JRSS-A, JRSS-C, JSS, LDA, StMed, SMMR, StMod, StSci
2. AISM, AoS, ANZS, Bern, Bka, CJS, CSSC, CSTM, CmpSt, CSDA, Envr, ISR, JASA, JAS, JCGS, JMA, JNS, JRSS-B, JSCS, JSPI, JTSA, Mtka, SJS, StCmp, Stats, StNee, StPap, SPL, StSin, Tech, Test
3. StataJ

Undirected weighted

1. AmS, BioJ, Bcs, Biost, EES, JABES, JBS, JRSS-A, JRSS-C, JSS, LDA, StMed, SMMR, StMod, StSci
2. AISM, AoS, ANZS, Bern, Bka, CJS, CSSC, CSTM, CmpSt, CSDA, Envr, ISR, JASA, JAS, JCGS, JMA, JNS, JRSS-B, JSCS, JSPI, JTSA, Mtka, SJS, StCmp, Stats, StNee, StPap, SPL, StSin, Tech, Test
3. StataJ

Visualisation

Here I am just trying out the ggraph package on the statistical journals datatset. The indicated clusters are those from agglomerative hierarchical clustering by Varin et al.

Asymmetric MDS

Symmetric MDS

Force-directed algorithm

Spring-based algorithm

Plotting communities

Following advice given in this StackOverflow answer we will aggregate communities into nodes representing each one (i.e. “super-journals”) and then plot them again to see if anything is worth looking at.

The visualisations above are definitely not scaleable to large networks (using R and ggraph, anyway), but the ones below can be applied to very large networks, if the number of clusters within them is only around 50–100 or so.

For now, we are only looking at the clustering given by Varin et al (2016).

Circular

Asymmetric MDS

Symmetric MDS

Force-directed

Spring-based