In 32 BioTIME 2.0 multi-site studies, new sites had no excess abundance decline in their first three years, so peak-start site selection was not detected

Question: monitoring sites are often selected when a species or assemblage is abundant. If the start year is a temporary high, the series shows an artificial decline in its first years (site-selection bias). We tested for this signature in BioTIME 2.0 (708 studies). Approach: a site is one study at one coordinate pair. (1) Within-study test: we compared annual change at new sites with change at old sites of the same study in the same calendar year (study x year fixed effects, per-study fits, random-effects pooling). (2) Series-level test: for 12,579 site series in 278 studies (8 or more years), we measured the deviation of the first year from the series trend. (3) Calibration: in 2,047 long series, we moved the start to the higher of 2 to 6 adjacent years to get the signature of peak-start selection in the same data. Results: in 32 studies with 6,323 new sites, the cumulative excess change of log abundance per sample over the first three intervals was +0.005 (95% CI -0.023 to +0.033); without the largest-weight study it was -0.016 (-0.059 to +0.027). A start at the higher of only two adjacent years gives -0.35 (-0.48 to -0.24). First-year abundance was not above the trend (-0.067; -0.161 to +0.025), against +0.40 under the same weak selection. No realm or taxon group showed a positive first-year deviation. Removal of the first year made abundance trends slightly more negative on average, so the first year does not inflate declines. The only start signal was an increase in recorded richness at new sites (+0.025 to +0.063 log units over three intervals), which agrees with first-year under-recording. Limitations: the tests detect selection on temporary highs, not on permanently better sites; the start is the first year in BioTIME, not always the real start; one amphibian scheme has 59% of the weight in the most precise estimate; the Living Planet Database and the Breeding Bird Survey were not tested.

site-selection bias · regression to the mean · BioTIME · biodiversity time series · monitoring · population trends · species richness · observer effects · meta-analysis · fixed effects