Manipulating Networks

Once a network is loaded, you’ll often need to reshape it before analysis: pull out a subset of nodes, merge nodes together, drop directionality, or remap tie values. This tutorial covers the five most common transformations.

Subsetting

nwsubset keeps only the nodes for which an if condition on an attribute is true — here, only the wealthiest Florentine families:

. nwwebuse florentine, nwclear

. nwsummarize flomarriage
--------------------------------------------------
   Network name:  flomarriage
   Network id:  2
   Directed: false
   Valued: false
   Two-mode: false
   Nodes: 16
   Selfloop: false
   Edges: 20
   Minimum value:  0
   Maximum value:  1
   Density:  .167
   Temporal: false

. nwsubset flomarriage if wealth > 50

. nwsummarize flomarriage_sub
--------------------------------------------------
   Network name:  flomarriage_sub
   Network id:  3
   Directed: false
   Valued: false
   Two-mode: false
   Nodes: 3
   Selfloop: false
   Edges: 1
   Minimum value:  0
   Maximum value:  1
   Density:  .333
   Temporal: false

By default the result is named netname_sub — 16 nodes down to 3. Unlike the similar-looking nwgen flo_sub = flomarriage if wealth > 50, nwsubset also copies over the original network’s node labels.

Collapsing

nwcollapse merges nodes together — the merged node inherits every tie either of the original nodes had. by() specifies which nodes get merged, and the new node is named after that variable’s value:

. nwrandom 8, prob(.3) name(mynet)

. nwsummarize mynet
--------------------------------------------------
   Network name:  mynet
   Network id:  4
   Directed: true
   Valued: false
   Two-mode: false
   Nodes: 8
   Selfloop: false
   Arcs: 24
   Minimum value:  0
   Maximum value:  1
   Density:  .429
   Temporal: false

. gen att = _n

. replace att = 1 in 2
(1 real change made)

. nwcollapse mynet, by(att) generate(mynet_collapsed)
(16 real changes made, 16 to missing)
(2 real changes made)
(8 real changes made)

. nwsummarize mynet_collapsed
--------------------------------------------------
   Network name:  mynet_collapsed
   Network id:  6
   Directed: true
   Valued: false
   Two-mode: false
   Nodes: 7
   Selfloop: false
   Arcs: 23
   Minimum value:  0
   Maximum value:  1
   Density:  .548
   Temporal: false

Nodes 1 and 2 collapse into one, taking mynet from 8 nodes down to 7.

Symmetrizing

nwsymmetrize (an alias for nwsym) turns a directed network into an undirected one. By default (mode(max)), a tie exists in either direction if it existed in either direction before:

. nwwebuse glasgow, nwclear

. nwsummarize glasgow1
--------------------------------------------------
   Network name:  glasgow1
   Network id:  1
   Directed: true
   Valued: false
   Two-mode: false
   Nodes: 50
   Selfloop: false
   Arcs: 113
   Minimum value:  0
   Maximum value:  1
   Density:  .046
   Temporal: false

. nwsymmetrize glasgow1, generate(glasgow1_sym)

. nwsummarize glasgow1_sym
--------------------------------------------------
   Network name:  glasgow1_sym
   Network id:  4
   Directed: false
   Valued: false
   Two-mode: false
   Nodes: 50
   Selfloop: false
   Edges: 74
   Minimum value:  0
   Maximum value:  1
   Density:  .06
   Temporal: false

. nwsym glasgow1_sym, check
--------------------------------------------------
   Network name:  glasgow1_sym
   Directed: false
   Symmetric: true

113 directed arcs become 74 undirected edges — reciprocated pairs collapse into a single tie. mode(min) instead requires a tie in both directions before keeping it, which typically keeps far fewer ties.

Recoding

nwrecode remaps dyad values according to arbitrary rules — think recode, but for tie values instead of a Stata variable:

. mata: M = (0,1,3 \ 2,0,1 \ 0,2,0)

. nwset, mat(M) name(scores) directed labs(A,B,C)

. nwsummarize scores, mat
--------------------------------------------------
   Network name:  scores
   Network id:  5
   Directed: true
   Valued: true
   Two-mode: false
   Nodes: 3
   Selfloop: false
   Arcs: 5
   Minimum value:  0
   Maximum value:  3
   Density:  .833
   Temporal: false
       1   2   3
    +-------------+
  1 |  .   1   3  |
  2 |  2   .   1  |
  3 |  0   2   .  |
    +-------------+

. nwrecode scores (1=10) (2=20) (3=30), generate(scores_recoded)
(3 real changes made)
(5 changes made to scores)

. nwsummarize scores_recoded, mat
--------------------------------------------------
   Network name:  scores_recoded
   Network id:  6
   Directed: true
   Valued: true
   Two-mode: false
   Nodes: 3
   Selfloop: false
   Arcs: 5
   Minimum value:  0
   Maximum value:  30
   Density:  .833
   Temporal: false
        1    2    3
    +----------------+
  1 |   .   10   30  |
  2 |  20    .   10  |
  3 |   0   20    .  |
    +----------------+

Rules can also cover ranges (1/5=1) and handle missing/other values (missing=9, else=44) — see nwrecode for the full rule syntax.

Dichotomizing

Turning a valued network binary at a single cutoff is common enough to have its own thin wrapper around nwrecode: nwdichotomize. Dyads at or above the threshold become 1, everything else becomes 0:

. mata: M = (0,150,40 \ 90,0,220 \ 60,30,0)

. nwset, mat(M) name(trade) directed labs(A,B,C)

. nwsummarize trade, mat
--------------------------------------------------
   Network name:  trade
   Network id:  7
   Directed: true
   Valued: true
   Two-mode: false
   Nodes: 3
   Selfloop: false
   Arcs: 6
   Minimum value:  30
   Maximum value:  220
   Density:  1
   Temporal: false
         1     2     3
    +-------------------+
  1 |    .   150    40  |
  2 |   90     .   220  |
  3 |   60    30     .  |
    +-------------------+

. nwdichotomize trade, threshold(100) generate(trade_binary)
(3 real changes made)
(6 changes made to trade)

. nwsummarize trade_binary, mat
--------------------------------------------------
   Network name:  trade_binary
   Network id:  8
   Directed: true
   Valued: true
   Two-mode: false
   Nodes: 3
   Selfloop: false
   Arcs: 2
   Minimum value:  0
   Maximum value:  1
   Density:  .333
   Temporal: false
       1   2   3
    +-------------+
  1 |  .   1   0  |
  2 |  0   .   1  |
  3 |  0   0   .  |
    +-------------+

Only the two ties of value ≥ 100 (A→B at 150, B→C at 220) survive as 1; everything else, including the two below-threshold ties, becomes 0.