Temporal comparisons drawn from aggregated occurrence databases assume that the records compared were produced by a comparable community of collectors. We tested that assumption using 297 voucher-backed records of Lycoris radiata, L. aurea, and L. chinensis retrieved from the Chinese Virtual Herbarium, comparing 2010–2015 with 2016–2020. Recorded habitat composition changed; Forest/Woodland fell from 44.9% to 22.3% of coded records and Urban/Residential rose from 8.9% to 20.4% (χ² = 20.308, df = 6, p = 0.002; permutation p = 0.002; Cramér’s V = 0.279). The contributing institutions and provinces, however, also fluctuated significantly (institution V = 0.550; province V = 0.380; both p < 0.001), and both were themselves strongly associated with recorded habitat. Stratification separated the two components: the Forest/Woodland decline persisted (Mantel–Haenszel χ² = 8.574, p = 0.003 by province; χ² = 6.250, p = 0.012 by institution), whereas the Grassland change did not (p = 0.871; p = 0.409). Record totals fell from 189 to 108, which we attribute to digitization lag rather than population change. Before interpreting temporal differences in aggregated occurrence data, researchers should examine changes in contributor composition and use stratified analyses to assess potential confounding.