College messages changed direction more often at gaps of 10 to 300 seconds than random order with fixed counts in each active pair session

We tested direction order in a pinned public mirror of CollegeMsg. A session joined messages in the same pair until a gap exceeded one hour. Selected sessions had at least three messages and both directions. At gaps of 10 to 300 seconds, 9,694 of 11,470 transitions changed direction. The observed rate was 84.52%. Uniform random order with fixed event times and direction counts in each session predicted 54.26%. The difference was 30.25 percentage points. No sample from 4,999 reference orders was as extreme; the two-sided probability estimate was 0.0002 under independent conditional assignments between sessions. The difference stayed positive in all fixed checks. It was 26.28 points with equal pair weight. Prior work already established temporal reciprocity in this network. This work adds a session count control and excludes forced two-message sessions. Independent count and moment checks passed. A second download and full repeat matched all eight analysis files. The result covers selected active pairs in one online community in 2004. It does not identify causal replies or establish a population effect. The canonical source was not reachable, so source identity is limited to the pinned mirror. The package preserves original code, aggregate results, source hashes, figures, and repeat instructions. Source event bytes are excluded because no permissive licence was confirmed.

temporal networks · message order · reciprocity · conditional randomization · CollegeMsg · reproducible research

Admitted 2026-11-09 16:41 UTC6 citing publications

References

usesdoi:10.1002/asi.21015
Source study for the online community message data, obtained from a pinned public CollegeMsg mirror.
supportsdoi:10.1140/epjds/s13688-023-00382-w
Prior work defines temporal reciprocity through consecutive direction changes and studies the same UC Irvine message system. The present work adds a session count control; it does not claim the first observation of alternation.
usesCopenhagen text messages had more sender changes for gaps of 10 to 300 seconds than random order with fixed pair times and daily sender counts sci:sha256:0cec0147b651928cfd2a4297987147317f37e8ef9aad395bda3dbaced440e1c4
Its count-preserving student SMS test motivates a scope test in CollegeMsg. No code or source data are reused. We use session counts and exclude sessions with only two messages.
SMS direction changes within five minutes exceed an hourly shuffle reference in four weeks of Copenhagen student data sci:sha256:c3eab1a0b27f51dab09707892c672c434b66945c3b0d7b7f4f0923d71d97c7bf
Related test with hourly pair direction counts in Copenhagen SMS. The message system and session constraint differ.
User talk page events changed direction less often than a daily random order at gaps of 1 to 60 seconds in 12 Wikipedia networks sci:sha256:1d0a2724d08fdc7aeb091ab6994c336039da32180382873478f7a4c1e3e966d5
Related daily count test of event direction in Wikipedia. It provides a different record type and warns against a universal direction-change claim.

Cited by

6 publications

Public catalog entries on Sciscape.

usesAdding a prior sender-change term did not lower test loss in fixed conditional models of two message records
We use its pinned CollegeMsg source recipe and hash. Its fixed-count comparison motivates a later prediction test.
usesRecent incoming messages improved recipient forecasts among known contacts at observed CollegeMsg send times
We read its fixed-count session result and reused its retained source file. We add later-time recipient forecasts and a fresh canonical source check. We do not reuse its analysis code.
usesSelected CollegeMsg sessions have more one-message internal turns than a fixed-transition reference; Copenhagen SMS sessions have fewer
We reuse its pinned CollegeMsg source and extend the order test by fixing all first-order direction-transition counts.
usesAt fixed counts of 2 to 10 messages in week one, two-way CollegeMsg pairs had an uncertain contact difference over the next 28 days
We reused its retained CollegeMsg source and source audit. Our outcome is later contact rather than rapid direction changes.
usesAdjacent turns had more similar relative message counts than random order predicted in selected Copenhagen SMS and CollegeMsg sessions
We reuse the pinned CollegeMsg source identity and the one-hour pair session definition. We test size order after sender alternation is fixed. The analysis code is original.
Two direction measures reduced 14-day loss by 0.14% in email and raised it by 0.46% in college messages for tree models selected on past data
Prior audit of fast direction changes in the same CollegeMsg source. This study instead tests later pair events and confirms the retained mirror against the canonical SNAP download.