Informal ESG Governance: Media Sentiment and Green Transition Impression Management

This study aims to examine the impact of ESG media sentiment on corporate green transition impression management. The core hypothesis is that ESG media sentiment has a significant inhibitory effect on corporate green transition impression management. Data are obtained from three sources: corporate financial data from the CSMAR database; the independent variable, ESG media sentiment, from the DATAGO commercial database; and the dependent variable, “green transition impression management”, which is self-constructed by the authors through textual analysis of corporate annual reports and manual coding from 2009 to 2022, ultimately forming a panel data structure with 21,852 firm-year observations. All continuous variables have been winsorized at the 1st and 99th percentiles, and missing values are handled consistently with the original data. Due to copyright and institutional subscription restrictions, the original real data cannot be shared directly. However, we provide a pseudo‑dataset (readme.dta) that has an identical structure, including the same variable names, variable definitions, number of observations, means, standard deviations, and distributional characteristics, but all variables are independently randomly generated. Therefore, running regressions with the pseudo‑dataset will not yield statistically significant results. In the real data, if the hypothesis holds, we should observe a significantly negative regression coefficient of ESG media sentiment on green transition impression management, indicating that more positive media sentiment is associated with less impression management behavior by firms. Furthermore, mechanism analysis, heterogeneity analysis, and a series of robustness and endogeneity tests (e.g., instrumental variable approaches, alternative model specifications) can further verify the reliability of this negative relationship. To use the data, researchers need to install the reghdfe and ftools packages in Stata 17.0 or higher, and then run the code files in the following order: descriptive statistics, main regression, robustness tests, endogeneity tests, mechanism analysis, and heterogeneity analysis. All results (regression tables, descriptive statistics, etc.) will be automatically saved in the Results folder. It should be noted that the pseudo‑dataset is provided solely for code verification; the coefficients and significance levels produced using it do not represent the true conclusions. To replicate the actual results of the original paper, researchers need to obtain permission for the original commercial databases and replace the pseudo‑dataset with the real data.