An independent multinomial correction gave negative mean monthly composition distance on two Portal plots in 1978–2025

We tested a finite-count correction for changes in rodent species proportions on two unmanipulated Portal Project plots. PortalData 6.117.0 supplied 419 pairs of adjacent censuses from 1978 to 2025. Under independent multinomial sampling, factorial moments give an unbiased squared distance estimate. The raw pooled mean distance was 0.06842. The corrected mean was −0.02046, with a nominal 95% two-year block interval of −0.02917 to −0.01164. Both plots also had negative corrected means when tested separately. A squared distance cannot be negative. The joint sample assumptions are therefore not suitable for this monthly record. The mean product of the two plot change vectors was positive, 0.01152, with an interval of 0.00309 to 0.02051. The median repeated-ID fraction at the smaller endpoint was 57.9%. Simulations showed that shared observations can give a negative corrected mean with no true change. They do not identify the cause in these field data. Twelve-moon pairs and a four-control check gave positive corrected means, which limits the result to the tested sample. The raw low-count excess distance was not clear under the other metrics. This does not establish equal ecological change or challenge the published latent community model. The package preserves exact source extracts, provenance, code, all results, figures, and an independent CSV reconstruction. The correction and plot agreement depend on sample assumptions; captures do not measure true population composition without detection and population models.

community composition · finite samples · rodent capture · Portal Project · temporal dependence · multinomial model · reproducibility