In 295 fish stocks with successive assessments in RAM Legacy 2018–2024, last-year recruitment estimates changed 2.5 times more than older years, with no mean bias

Question: how much do the newest recruitment values in the RAM Legacy Stock Assessment Database change when the next assessment of the same stock comes out? Approach: we pooled 14 public RAM versions (v4.40, 2018, to v4.66, 2024; 2,791 assessments). For each stock, we paired each assessment with the next assessment that ends later. We centred each log series on years at least 10 years before the last year of the older assessment, and calculated the revision (newer minus older) for each year before that last year (lag). Intervals are 95% bootstrap intervals over stocks. Results for recruitment (1,027 pairs, 295 stocks): at lag 0 the root mean square revision was 0.75 log units (0.64–0.87), 2.5 times (2.1–3.0) the value at lags 5–9. The mean revision was +0.003 (−0.054 to +0.066), so there was no mean bias. The slope of the newer on the older value was 0.73 (0.62–0.86) at lag 0 and 0.94 at lags 5–9: extreme last-year values moved back toward the mean. Strong last-year year classes were revised down by 0.16 log units, and 38% lost their strong status. 15.5% of last-year values were at the reference mean, against about 7% in older years. Ten-year recruitment trends changed little (slope 0.89, same sign in 83% of pairs). Spawning biomass (1,037 pairs, 317 stocks) had smaller revisions (0.30 at lag 0), and older values for the last 1–4 years were 2–3% too high, the same direction as Edgar et al. (2024). Results were similar without the alignment filter, with same-unit pairs only, and for assessments that end in 2010 or later. Main limitation: the newer assessment is not the true value, so these are differences between assessments and a lower limit for the error. Studies that use the newest RAM recruitment values should shrink or omit the last 2–3 years; ten-year trends are much less affected.

stock assessment · recruitment · RAM Legacy · retrospective pattern · assessment revision · spawning stock biomass · fisheries · data quality · meta-analysis · terminal year