Fransson's SID: for each time bin, how much more tied a network's communities are internally than to each other, both normalised by the number of pairs available. Positive means segregation exceeds integration.
Arguments
- dn
A temporal network from
dynet().- communities
Either a vector named by vertex name, or a single string naming a column of the node table, as
mixing()takes. A positional vector is refused: matching a partition to vertices by position is exactly the index-based addressing this package avoids.- scope
"pertime", the default, for one row per bin, or"overall"for the mean over bins.- sessions
How to treat sessions, as in
centrality_series().- start, end, step, window
The measurement grid, as in
snapshots().- plot
Whether to draw the result as well as return it. Drawing is a side effect in the manner of
graphics::hist(): the verb still returns its tidy table, invisibly when it has drawn.
Value
A dynet_metric at level = "graph" with columns time (under
scope = "pertime"), measure, which is the constant "sid", and
value. A leading session column is present under
sessions = "separate". The value is signed and unbounded: it is a
difference of two normalised sums, not a proportion, so there is no range
to check it against. Print it, summary() it, plot() it, or take the
plain frame with as.data.frame().
Details
For communities \(a = 1, \dots, C\) of sizes \(N_a\), with
\(W_a\) the ties inside community \(a\) and \(B_{ab}\) the ties
between \(a\) and \(b\) in a bin, the value is
$$SID = \sum_a \left( \frac{2 W_a}{N_a (N_a - 1)} -
\sum_{b \ne a} \frac{B_{ab}}{N_a N_b} \right).$$
Each community's within-density enters once. With two communities this
equals teneto.networkmeasures.sid(calc = "overtime") exactly (checked
against teneto 0.5.3). With three or more it does not: teneto sums the
pairwise contrast over every ordered pair of communities, so its
within-density term enters \(C - 1\) times and its value is
\(\sum_a (C - 1) 2 W_a / (N_a (N_a - 1)) - \sum_a \sum_{b \ne a}
B_{ab} / (N_a N_b)\). Values from the two are comparable only for
\(C = 2\). A bin with no tie scores zero.
A community of one vertex makes the within-community normaliser
2 / (N_a (N_a - 1)) a division by zero. Rather than drop the vertex,
which would change every other community's denominator, the bin returns
NA_real_ and a dynet_singleton_community warning names the offending
community.
The measure is defined on undirected networks, and a directed one is read
by folding each layer onto its transpose, as persistence() does. Weights
and spell counts are ignored: a pair is tied in a bin or it is not.
Conditions
Errors: dynet_bad_communities when the partition is unnamed, names a
vertex the network does not have, or leaves a vertex unassigned;
dynet_no_sessions under sessions = "separate" without a session column;
dynet_bad_input when dn is not a dynet. Warns with
dynet_singleton_community when a community has one member.
References
Fransson, P., Thompson, W. H., Skiold, B., et al. (2018). Brain network segregation and integration during an epoch-related working memory fMRI experiment. NeuroImage, 178, 147-161. doi:10.1016/j.neuroimage.2018.05.040
Thompson, W. H., Brantefors, P., & Fransson, P. (2017). From static to temporal network theory. Network Neuroscience, 1(2), 69-99. doi:10.1162/NETN_a_00011
See also
mixing() for who ties to whom by attribute, and persistence()
for whether those ties survive.
Examples
dn <- dynet(school_contacts)
groups <- stats::setNames(
rep(c("a", "b"), length.out = nrow(dn$nodes)), dn$nodes$name
)
sid <- segregation(dn, communities = groups)
sid
#> # Segregation-integration difference (graph-level)
#> # 22 time points, 1 per bin | time in step
#> # positive is more segregated than integrated; signed and unbounded
#> time measure value
#> 0 sid -0.054421769
#> 1 sid 0.244897959
#> 2 sid -0.142857143
#> 3 sid -0.224489796
#> 4 sid -0.047619048
#> 5 sid 0.231292517
#> 6 sid 0.047619048
#> 7 sid 0.054421769
#> 8 sid -0.074829932
#> 9 sid -0.258503401
#> 10 sid -0.163265306
#> 11 sid 0.006802721
#> # 10 more rows. summary() aggregates them; plot() draws them.