Corruption, defined as the abuse of power for private gain at public expense, significantly undermines government performance and erodes public trust. While anti-corruption efforts aim to combat this, their impact on public perceptions is complex. While long-term effects are often positive, short-term revelations of corruption scandals can increase public perceptions of corruption. Existing research primarily focuses on investigation numbers over specific periods, lacking empirical evidence on sustained campaigns. This study addresses this gap by exploring how fluctuations in annual investigation numbers within a sustained anti-corruption campaign influence public perceptions in China, where such campaigns are common and information is publicly available. The central question is whether and how these fluctuations alter public perceptions of corruption, considering the potential interplay between past trends and recent events.

The literature review examines the measurement of corruption (objective indicators like the number of investigations, and subjective indicators like perceptions), and factors influencing corruption perceptions. These factors include individual characteristics (place of residence), information sources (personal experience, media coverage, and grapevine), and government anti-corruption efforts. The complex relationship between anti-corruption efforts and perceptions is highlighted; increased efforts can initially worsen perceptions by exposing corruption, but may improve them in the long run. The review introduces two models of evaluation: memory-based (judgments based on recalled information, with a recency effect) and online evaluation (continuous formation and revision of attitudes by integrating new information). The study hypothesizes that sustained campaigns will align with the online evaluation model, where both past trends and recent information shape perceptions.

An online survey experiment was conducted in China in early 2022, recruiting participants through Sina Weibo. A total of 2317 participants (after screening for inattentive responses) completed the questionnaire. The experiment used a fictitious country to control for pre-existing knowledge of China's specific anti-corruption campaign, but maintained similar corruption ratios and investigation numbers (scaled down for cognitive ease). Participants were presented with nine years' worth of fictional annual corruption investigation data, varying the past trend (upward, downward) and the most recent year's investigation number (high, low). A control group received no annual figures. Participants then rated the government's current corruption level on a 1-5 scale. A 2x2 design crossed past trends with recent investigation numbers, creating four experimental conditions (continuous upward, unexpected decrease, continuous downward, unexpected increase) and a control condition (undisclosed annual figures).

The results showed significant differences in perceived corruption levels across groups. The average perceived corruption was highest in the continuous upward trend condition (4.34) and lowest in the continuous downward trend condition (1.42). Unexpected changes in the most recent year also significantly affected perceptions. An unexpected decrease following an upward trend still resulted in high perceived corruption (4.06), while an unexpected increase after a downward trend led to a dramatic rise in perceived corruption (4.15), compared to the continuous downward condition (1.42). OLS regression analysis confirmed that both past trend and the recent number of investigations significantly and positively predicted corruption perceptions, supporting the online evaluation model. The interaction term between past trend and recent investigations was also significant, indicating that recent, unexpected information can reverse perceptions based on past trends, particularly when the past trend had been downward. This effect was more pronounced for an unexpected increase than an unexpected decrease.

The findings strongly support the online evaluation model, demonstrating that individuals continuously update their perceptions of corruption, integrating both past trends and the latest information. This contradicts the memory-based model, which would predict a stronger influence of recent data alone. The results highlight the importance of considering the entire sequence of information—not just the total number or recent numbers—in understanding how sustained anti-corruption campaigns shape public perceptions. The asymmetry in the effect of unexpected increases versus unexpected decreases suggests that positive signals might require sustained evidence to effectively alter deeply ingrained negative perceptions.

This study contributes to the understanding of corruption perception formation within sustained anti-corruption campaigns, showing the importance of both past trends and recent information. The online evaluation model provides a robust framework for interpreting the dynamic relationship between anti-corruption efforts and public perceptions. Future research should explore the long-term impact of sustained positive trends and investigate the influence of other factors like media framing and individual experiences.

The study's limitations include the lack of control for individual corruption experiences and educational backgrounds, and the use of a sample from Sina Weibo, potentially limiting the generalizability to the wider Chinese population. The use of a fictitious country might also have limited the external validity of findings. Future research could address these limitations by incorporating these factors and employing more representative sampling techniques.