万尧
This project synthesizes quasi-experimental evidence on the effect of China's emissions trading scheme (ETS) on green innovation and asks how much of the wide variation in reported effects is attributable to study design rather than to genuine differences in policy impact. Scope. Literature is identified through two primary databases — CNKI (Academic Journals and Doctoral/Masters Dissertations) for Chinese-language work and the Web of Science Core Collection for English-language work — supplemented by Wanfang Data and by Google Scholar, the latter used to catch working papers, preprints and studies the three databases miss rather than as a systematic index; both English- and Chinese-language studies are eligible, published and unpublished. Outcomes and effect size. All estimates reported in an eligible study are coded, including dynamic (lead/lag/cumulative) and subsample results. The primary estimand is the absolute level of green patents; the confirmatory sample is restricted to contemporaneous, full-sample estimates in a main or mechanism (a-path) role. Green total factor productivity and the green-patent share are coded as separate estimands and analysed only as secondary outcomes. Because outcomes enter primary regressions in incompatible units (counts, logs, ratios), the primary effect-size metric is the partial correlation coefficient, computed from reported t-statistics and degrees of freedom; semi-elasticities from the log/count subset serve as a robustness path. All effect sizes are signed so that positive values indicate that the ETS promotes green innovation. Analysis. We fit three-level random-effects models with sampling variance at level 1, estimates nested within studies at level 2, and between-study variance at level 3, with cluster-robust inference at the study level. Multivariate meta-regression then tests pre-specified moderators, including estimator family (in particular heterogeneity-robust staggered estimators), treatment definition and coded treatment year, treatment scope (single pilot, multiple pilots, or all seven), functional form of the outcome, patent type (invention versus utility model, granted versus applied), green-patent classification standard and data source, fixed-effects and control-variable specification, sample level and period, and publication characteristics. For studies covering a single pilot, external context variables (carbon price, allowance tightness, number of covered sectors) are matched and tested as moderators. Selective reporting and publication bias are assessed using funnel-based tests, FAT-PET/PEESE, and selection-model approaches; outliers and influential estimates are retained at coding and addressed only at the analysis stage. Transparency. Literature searching and coding against a pre-specified codebook began before this registration was created; no effect sizes have been synthesized and no meta-analytic or meta-regression models have been estimated. The codebook and the search log as of the registration date are included in the registration.