net_centrality() scores how central each node is in a network, for one
or more measures, and returns one row per node. Strength sums a node's
edge weights (incoming or outgoing); closeness and betweenness read the
shortest paths between nodes, where a heavier edge is a shorter step;
diffusion counts the weighted walks that start at a node; and clustering
measures how strongly a node's neighbours are tied to each other. It
takes a single network or a group of them (a netobject_group, the
clusters of an mcml), scoring each separately.
Centralities describe the estimated network, not the population it was
estimated from. Rankings built on a few weak edges change from sample to
sample: check them with centrality_stability() before reporting which
node is most central.
Usage
net_centrality(
x,
measures = NULL,
loops = FALSE,
normalize = FALSE,
invert = TRUE,
normalize_diffusion = TRUE,
centrality_fn = NULL,
...
)
# S3 method for class 'net_centrality'
plot(
x,
reorder = TRUE,
ncol = 3L,
type = c("bar", "line", "heatmap"),
scales = c("free_x", "fixed"),
profile_scale = c("measure", "none"),
labels = TRUE,
drop_zero = FALSE,
...
)
# S3 method for class 'net_centrality_group'
plot(
x,
reorder = TRUE,
ncol = 3L,
type = c("bar", "line", "delta"),
scales = c("free_x", "fixed"),
palette = "Set2",
profile_scale = c("measure", "none"),
labels = FALSE,
drop_zero = FALSE,
...
)Arguments
- x
The network: a
netobject,netobject_group,mcmlorcograph_network. For theplot()methods: the result ofnet_centrality().- measures
The measures to compute. The default is
"InStrength","Betweenness"and"Diffusion";"all"computes every built-in one:"OutStrength"and"InStrength"(sum of outgoing or incoming weights),"ClosenessOut","ClosenessIn"and"Closeness"(inverse of the summed shortest-path distances from, to, or ignoring the direction of edges),"Betweenness"(how many shortest paths pass through the node),"BetweennessRSP"(betweenness over randomised shortest paths, which also counts near-shortest routes),"Diffusion"(the summed weights of walks of length 1 to n from the node) and"Clustering"(weighted clustering coefficient)."InCloseness"and"OutCloseness"are accepted for"ClosenessIn"and"ClosenessOut". Withcentrality_fn, name its measures here to have them returned.- loops
Whether self-loops (staying in a state) count. With
FALSE(default) they are left out of every measure, and a message says so.- normalize
Whether to range-normalise every measure to 0–1, as
tna::centralities(normalize = TRUE)does. DefaultFALSE.- invert
Whether a heavier edge is a shorter step in the shortest-path measures (distance = 1 / weight), as in
tna. WithFALSE, the weight itself is the distance. DefaultTRUE.- normalize_diffusion
Whether to range-normalise
Diffusioneven whennormalize = FALSE. DefaultTRUE.- centrality_fn
A function of your own that takes the weight matrix and returns a named list of centrality vectors, one per measure; list their names in
measures. DefaultNULL.- ...
Not used.
- reorder
For the
plot()methods: whether to sort the nodes in each measure's panel by their value (for a group, by their mean value across networks). DefaultTRUE.- ncol
For the
plot()methods: the number of panel columns. Default3.- type
For the
plot()methods: the kind of plot."bar"(default) draws one panel of bars per measure, grouped bars for several networks."line"draws each node's profile across the measures (for several networks, one panel per node and one line per network);"profile"is accepted for it. A single network also takes"heatmap", a nodes-by-measures grid with each measure scaled to 0–1 for the colour and its raw value printed in the tile. Several networks also take"delta", bars of the differences between networks: the second minus the first with two, each network's deviation from the per-node mean with more.- scales
For the
plot()methods:"free_x"(default) gives each measure its own value axis,"fixed"one shared axis.- profile_scale
For
type = "line":"measure"(default) rescales each measure to 0–1 so the profiles compare across measures;"none"draws raw values.- labels
For the
plot()methods: whether to print the values on the plot. DefaultTRUEfor one network,FALSEfor several.- drop_zero
For the
plot()methods: whether to leave out measures that are zero for every node. DefaultFALSE.- palette
For
plot()of several networks: the ColorBrewer palette for the networks. Default"Set2". Ignored bytype = "delta"with two networks, whose bars are coloured by which network is higher.
Value
For a netobject or cograph_network: a net_centrality data
frame with one row per node, a state column and one column per
requested measure (the node names are also the row names). For a
netobject_group or an mcml: a net_centrality_group, a list of such
data frames, one per network. plot() draws the measures per node.
For the plot() methods: a ggplot object.
Details
Strength is read from the weight matrix directly, and closeness and
betweenness from all-pairs shortest paths (Floyd-Warshall), without
igraph; the values agree with tna::centralities(). Diffusion is
range-normalised by default, which tna does not do.
Range normalisation of a measure that is the same for every node is
undefined and gives NaN, with a warning of class
nestimate_constant_measure. Diffusion on a "relative" network with
loops = TRUE is such a case: each row sums to 1, so every node has the
same diffusion; use normalize_diffusion = FALSE there.
References
Freeman, L. C. (1978). Centrality in social networks: conceptual clarification. Social Networks, 1(3), 215–239. (betweenness, closeness)
Opsahl, T., Agneessens, F. & Skvoretz, J. (2010). Node centrality in weighted networks: generalizing degree and shortest paths. Social Networks, 32(3), 245–251. (weighted strength and geodesics)
Kivimaki, I., Lebichot, B., Saramaki, J. & Saerens, M. (2016). Two
betweenness centrality measures based on randomized shortest paths.
Scientific Reports, 6, 19668. (BetweennessRSP)
Banerjee, A., Chandrasekhar, A. G., Duflo, E. & Jackson, M. O. (2013). The
diffusion of microfinance. Science, 341(6144), 1236498.
(Diffusion)
Onnela, J.-P., Saramaki, J., Kertesz, J. & Kaski, K. (2005). Intensity and
coherence of motifs in weighted complex networks. Physical Review E,
71, 065103. (Clustering)
Examples
seqs <- data.frame(
V1 = c("A","B","A","C"), V2 = c("B","C","B","A"),
V3 = c("C","A","C","B"))
net <- build_network(seqs, method = "relative")
net_centrality(net)
#> centralities computed excluding loops (diagonal). Pass `loops = TRUE` to include self-transitions.
#> Warning: Diffusion is the same for every node, so it cannot be range-normalised and is NaN. Use `normalize_diffusion = FALSE` for the raw values.
#> state InStrength Betweenness Diffusion
#> A A 1 1 NaN
#> B B 1 1 NaN
#> C C 1 1 NaN