Methodological heterogeneity outweighs geography in shaping reported riverine microplastic concentrations: a global meta-analysis and implications for water-quality monitoring

The workbook Microplastics_Rivers_Meta_Analysis_DATA.xlsx contains eight sheets: - Analysis_dataset (509 rows x 57 fields) — every extracted observation, harmonised, with methodological provenance and composition percentages. - Pooled_283 (283 rows x 61 fields) — the subset carrying enough unit metadata to enter the pooled concentration analyses, with extraction-confidence flags and the source URL used. - Excluded_226 (226 rows x 62 fields) — every observation excluded from pooling, each with an explicit reason in the first column. - Unit_conversions (174 rows) — the complete conversion table. 170 distinct unit strings appear in the source literature; four of them occur in both matrices and take a different factor in each, hence 174 rules. Each row gives the raw string, its frequency, the factor, the harmonised unit and, where applicable, the exclusion reason. - Paired_66 (66 rows) — study-level median water and sediment concentrations for the studies reporting both matrices in convertible units, with the water-to-sediment ratio. - Composition (367 rows x 27 fields) — morphology and polymer percentages as reported, before renormalisation. - Method_groups (20 rows) — the mapping from verbatim method descriptions to the analysis groupings, with example source strings. - Study_list (309 rows) — all included studies with DOI, country, rivers, observation count and whether each entered the pooled analysis. Concentrations are harmonised to items m-3 for water and items kg-1 dry weight for sediment. Values that could not be converted to a particle concentration are retained in the workbook and flagged rather than deleted, so the exclusion decisions are auditable. Microplastics_Rivers_Meta_Analysis_CODE.zip contains the Python code that reproduces every statistic and figure in the associated article from this workbook alone: a shared module of loading, harmonisation and statistical helpers, eight numbered analysis scripts, a runner that executes them in order and then compares each recomputed value against the value printed in the manuscript, a README and a pinned requirements file. There is no network access, no external data and no hidden state. Known gaps, stated plainly. Two groups of fields used in the published analysis are not columns of this workbook. First, the standard deviation reported by original authors alongside their mean is absent, so the random-effects branch of the analysis (DerSimonian-Laird pooled estimates, Cochran Q, tau-squared, I-squared, prediction intervals and the 55-water / 44-sediment study counts) cannot be regenerated from this deposit; nine of the ninety-eight verified statistics are affected and are reported as "missing input" rather than silently omitted. Second, catchment drainage area and river length were joined from an external gazetteer and are likewise not included. Everything else reproduces exactly.