Compare Network Change Across Repeated Snapshots
Align repeated network observations, distinguish real tie change from boundary and response changes, and compare structure across time without flattening the sequence into one graph.
Method sources
By the end of this tutorial
- 1Create comparable network snapshots with aligned nodes, tie definitions, and observation windows.
- 2Measure tie persistence, entry, exit, and structural change without hiding missing data.
- 3Visualize and interpret longitudinal patterns while preserving time order and uncertainty.
SNA
Network specification
Analysis scenario
A professional learning program surveys the same 24 educators at four quarterly meetings about instructional-advice seeking. The boundary is the fixed enrollment roster from the first meeting; departures, new arrivals, and nonresponse are recorded explicitly rather than silently changing who counts as part of the network.
Nodes
The 24 educators on the program's fixed enrollment roster, with presence and response status recorded at every quarterly wave.
Ties
A directed binary tie records that educator A named educator B as someone from whom A sought instructional advice during that quarter.
Network type
One-mode, directed, binary panel network with four equal observation windows, one fixed roster boundary, and wave-specific missingness indicators.
Step-by-step tutorial
Align the measurement design
Create a wave table with one row per quarter and document the roster, nomination prompt, recall period, data-collection date, and response status. Resolve identifier changes before building any graph, and keep a reason code for absence, departure, entry, or nonresponse.
Checkpoint
Every person has one stable identifier, every wave uses the same tie meaning, and missing data are distinguishable from an observed absence of a tie.
Build comparable snapshots
Construct one adjacency matrix per wave using the same ordered node roster. Keep inactive or unobserved nodes flagged rather than deleting them, then calculate response coverage and the number of observable dyads before comparing network statistics.
Checkpoint
All matrices use the same node order and dimensions, and each reported statistic is paired with its wave-specific coverage and eligible-dyad count.
Separate tie turnover from structure
For adjacent waves, count persistent, newly observed, and dissolved ties. Compare density, reciprocity, components, and node degree only after checking boundary and coverage stability. Use a similarity measure such as Jaccard overlap to summarize tie-set continuity.
Checkpoint
You can show which changes come from tie entry or exit and which could instead reflect roster movement, nonresponse, or a changed observation window.
Visualize time and test sensitivity
Use fixed node coordinates across small multiples so movement on the page does not masquerade as relational change. Repeat comparisons after excluding poorly observed waves or restricting to consistently observed actors, and report patterns that do not survive these checks.
Checkpoint
The display preserves wave order and positional comparability, while the written conclusion states how missingness and sensitivity choices affect the pattern.
Interpret with care
A higher density or degree at a later wave can indicate more reported advice ties, but it can also reflect better response coverage or a smaller set of eligible dyads. Interpret change only after placing the statistic beside the boundary, response rate, and observation window for each wave.
Stable aggregate statistics can hide substantial tie turnover, while visible layout movement can exaggerate change if coordinates are recomputed independently. Pair network-level measures with tie persistence and fixed-coordinate views before describing a network as stable, fragmented, or increasingly connected.