When Metrics Backfire: How China’s Performance Incentives Left a Shadow of Research Misconduct

Retractions have climbed to unprecedented levels worldwide, and the search for explanations has often focused on individual ethical lapses. A new study published in Humanities and Social Sciences Communications shifts the lens to a different suspect: the institutional machinery that rewards researchers for their numbers. Analyzing three decades of misconduct-related retraction data from China, researchers Baosi Meng, Wenjing Huang, and Xun Wu of The Hong Kong University of Science and Technology (Guangzhou) document systematic temporal associations between the introduction of performance-based research evaluation reforms and subsequent shifts in misconduct trends.

The research team employed a multiple interrupted time-series design, a statistical approach that examines whether the timing of a policy intervention coincides with measurable changes in the level or slope of an outcome over time. Their outcome of interest was the misconduct-related retraction rate, calculated per 10,000 field-specific publications, drawing on retraction records from the Retraction Watch Database and publication denominators from OpenAlex. The analytic sample covered research articles published between 1992 and 2020, restricted to retraction notices in which a Chinese institution appeared in either the first or last author-affiliation position.

Three policy moments anchored the analysis: the 1995, 1998, and 2015 research evaluation reforms in China. These reforms progressively institutionalized metric-based evaluation, tying career advancement, funding, and institutional prestige to quantitative indicators such as publication counts and journal rankings. The interrupted time-series analysis revealed significant level and trend shifts in misconduct-related retraction rates around each of these reform years, suggesting that changes in the evaluation regime were temporally aligned with changes in integrity outcomes.

The disciplinary patterns proved particularly striking. The clearest differential increase in misconduct-related retractions occurred in STEM fields following the 1998 reform, a period when quantitative metrics became deeply embedded in the evaluation of scientific output in China. Meanwhile, humanities and social sciences fields, which had historically been evaluated with less emphasis on publication metrics, showed a pronounced acceleration in misconduct-related retraction rates after 2015, coinciding with the later extension of metric-based evaluation to those disciplines.

This staggered timing across disciplines is central to the study’s argument. If misconduct were purely a matter of individual ethics, one would not expect retraction trends to track the sequential rollout of evaluation reforms across different fields. Instead, the observed patterns align with the hypothesis that the institutionalization of metric-based incentives reshapes the risk environment for research integrity, creating pressures that can push some researchers toward questionable practices when the rewards for output outweigh the safeguards for quality.

The researchers were careful to constrain their claims. While more recent Chinese reforms have aimed to de-emphasize quantitative indicators, the study confines its analysis to the early phases of policy introduction and intensification, where sufficient post-reform temporal observations exist to identify stable misconduct patterns. This methodological discipline allows the team to document associations without overreaching into causal claims that the data cannot fully support. The temporal alignment they map is systematic, but the design captures correlation in time rather than direct experimental proof of causation.

The coding of misconduct-related retractions followed the Retraction Watch Database User Guide, with each retraction reason classified according to documented rationale. Records with multiple subject tags were assigned a single primary field using a priority rule favoring more specific disciplinary labels over broad or methodological ones. Publication denominators excluded reviews, editorials, letters, conference publications, proceedings articles, and preprints, ensuring that the retraction rates reflected genuine research articles rather than the full breadth of scholarly output.

The findings arrive at a moment of intense global discussion about research governance. Retraction rates have risen across many countries, and the pressure to publish has become a shared concern from North America to East Asia. China’s experience offers a particularly valuable case study because its evaluation reforms were introduced in distinct waves, creating a natural temporal structure that researchers can exploit to examine how incentive systems and integrity outcomes move together. The study’s authors frame their contribution as addressing common challenges in fostering ethical research amid increasing competition and quantified evaluation.

For policymakers, the message is sobering. Evaluation systems designed to accelerate productivity may carry hidden costs that surface years later in the form of retracted papers and damaged reputations. The temporal lag between reform and misconduct patterns suggests that the consequences of incentive design are not immediate but accumulate as researchers adapt to the new rules of the game. As institutions worldwide experiment with alternatives to pure metrics, including narrative portfolios and responsible research assessment initiatives, the Chinese experience documented by Meng, Huang, and Wu provides empirical grounding for the argument that how success is measured matters as much as success itself.

Beyond the headline findings, the study’s methodological architecture deserves attention because interrupted time-series designs occupy a distinctive niche in policy evaluation. Unlike randomized experiments, they cannot isolate causal effects cleanly, but they are well suited to situations where a policy rolls out at a known date and where long historical baselines exist. By comparing pre-reform and post-reform trajectories of misconduct-related retraction rates, the researchers could distinguish abrupt level changes from gradual trend accelerations, which matters because these two patterns imply different mechanisms. A level shift might reflect a burst of misconduct as researchers rush to meet new targets, while a sustained slope change suggests that the altered incentive environment persistently reshaped behavior across cohorts of researchers entering the system.

The choice of retraction rates as the outcome variable also reflects careful measurement reasoning. Raw retraction counts would conflate changes in misconduct with changes in publication volume, which grew explosively in China over the study period. Normalizing by field-specific publication counts per 10,000 articles controls for the sheer expansion of the research enterprise, making it possible to ask whether misconduct became more frequent relative to output, not merely more numerous in absolute terms. The field-level normalization is equally important because publication norms, retraction practices, and editorial scrutiny differ substantially across disciplines, so a single aggregate rate could mask divergent dynamics.

The continuity correction applied in the field-level comparisons, adding 0.5 retractions to numerators in both periods, addresses a practical statistical problem: many fields recorded zero misconduct-related retractions in the early 1990s baseline, which would make ratio calculations undefined or wildly unstable. This conservative adjustment tempers the magnitude of apparent increases in fields starting from near-zero baselines, and it is a reminder that some of the most dramatic percentage changes in the data occur in disciplines where the absolute numbers remain small. Readers interpreting field-level ratios should therefore weigh both the relative and absolute magnitudes of change.

The 1998 reform deserves particular emphasis in interpreting the STEM pattern. That policy moment coincided with the deep embedding of quantitative indicators into Chinese science policy, including lists of high-status journals and incentive structures that rewarded publications in indexed venues with cash bonuses and promotion advantages. The alignment of the clearest differential increase in STEM misconduct-related retractions with this period is consistent with a large body of scholarship on metric fixation, which argues that when evaluation systems reward countable outputs, actors optimize for the count rather than the underlying quality the count is meant to proxy. The humanities and social sciences pattern after 2015 mirrors this logic with a delay, consistent with the later diffusion of metric-based evaluation into fields where peer review of portfolios had traditionally dominated.

It is worth noting what the study does not claim. The authors explicitly confine their analysis to early phases of policy introduction and intensification, where enough post-reform observations exist to identify stable patterns. More recent Chinese policy initiatives, including directives that discourage the sole use of publication counts and journal tiers in evaluation, fall outside the observation window, so the study cannot say whether de-emphasizing metrics reverses the misconduct trends it documents. That question remains open and is arguably the most policy-relevant test of the incentive hypothesis: if retraction trajectories bend downward as quantitative indicators lose weight, the temporal association would gain considerable inferential strength.

The data infrastructure underpinning the analysis also illustrates how the modern scholarly record enables this kind of research. The Retraction Watch Database, which catalogues retraction notices with structured reasons, and OpenAlex, which provides publication metadata and field classifications, together allow researchers to construct outcome measures that were impractical to assemble a decade ago. The manual mapping of Retraction Watch subject tags to the OpenAlex Topics hierarchy, with priority given to specific disciplinary labels, reflects the messy reality of bibliographic data, where a single retracted article may carry multiple tags and where broad categories like technology would otherwise dilute disciplinary comparisons.

For the international research community, the study adds empirical weight to ongoing debates about responsible research assessment. Movements such as the Declaration on Research Assessment have urged funders and institutions to move beyond journal-based metrics, and China’s own ministries have issued guidance in a similar direction. What this study contributes is a documented temporal record showing that the consequences of evaluation design are observable, discipline-specific, and lagged. As other countries that adopted metric-heavy evaluation regimes earlier or later than China accumulate comparable retraction data, replication of this interrupted time-series approach across national contexts would help establish whether the patterns documented here are a Chinese phenomenon or a general property of quantified research governance.

Subject of Research: Temporal association between performance-based research evaluation reforms and research misconduct trends in China

Article Title: Shadow of success: the temporal association between performance incentive system and research misconduct

Article References: Shadow of success: the temporal association between performance incentive system and research misconduct. (n.d.). https://doi.org/10.1038/s41599-026-09307-0

Image Credits: AI Generated

DOI: 10.1038/s41599-026-09307-0

Keywords: research misconduct, retractions, performance incentives, research evaluation, China, STEM, humanities and social sciences, time-series analysis, research integrity, metric-based evaluation, research governance, Retraction Watch