Built, not observed: construction rules and the comparability of register-based population networks
Statistics offices and research groups in the Netherlands, Denmark and Sweden have built social networks of whole populations from administrative registers. A tie in these networks is not observed. It is produced by a rule: link an employee to the 100 colleagues who live nearest, to a random 100, or to a number that depends on the size of the workplace; link a household to the ten closest households within 50 metres, or with no limit. The rules differ between countries and between releases. We ask how much of a difference between two groups or two populations is produced by the rules. We wire the same synthetic populations with the published rule-sets, in pre-registered experiments, and repeat the central comparisons on a synthetic population built by others from census data. Under the Dutch yearly-series and the Swedish-like rules, construction accounts for 47% to 98% of a naive difference in total degree and clustering between two populations, and the naive sign of a clustering difference is wrong in 65% to 90% of populations. Under the Danish-like rules the share depends on how large workplaces are sampled, which is not published. Gaps between income groups rarely change sign, but they change in size by factors of up to four, appear under one rule-set and not another, and reorder the groups. A correction that needs only the other network's published rules works when the simulator behind it is right about the feature the rules act on, and can give the wrong sign otherwise. No network measure we tested is both stable across rule-sets and sensitive to a real difference. Comparisons across such networks need the rules to be aligned first.
Authors
- S. M. Abdulla Al Mamun
Institutions
- University of Manchester (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24374074
- Primary Topic
- Complex Network Analysis Techniques
- Type
- preprint