Capital Connectivity in Integrated Reports: Datasets from International Companies

The panel dataset provides information on the degree of capital connectivity disclosed in 840 integrated reports from 80 Asia-Australian, 78 European, and 10 American firms over a five-year period from 2018 to 2022. The sample includes firms from eleven different sectors, as defined by the Global Industry Classification Standards (GICS). The dataset focuses on international firms that voluntarily adopt integrated reporting, with these reports available in the official IR database.

Data extraction is performed using a robotic software, which systematically codes the quantity and sequence of specific capitals in the reports. The software uses a list of keywords related to the six capitals outlined in the IR Framework: Financial, Manufactured, Intellectual, Human, Social & Relationship, and Natural. A self-calculated connectivity score is then generated to assess firms’ capital connectivity, by counting the average number of different types of capital coded within a specific context in each integrated report. This score ranges from a minimum of one (1) and a maximum of six (6), reflecting the average number of different types of capital codes. Additional scoring is calculated based on the percentage of capital disclosed directly, followed by another capital. The perfect score of 100% indicates that a different capital follows every other capital disclosure, while the lowest possible score of zero means none. Please refer to the excel file for detailed scoring measurement.

Experts from industries and academicians have validated the instrument used to measure capital connectivity. To assess the reliability of the connectivity score, ten (10) integrated reports were randomly selected and manually coded, and the score was calculated based on the manual coding. Weighted Cohen’s Kappa inter-rater reliability test was used to compare Connectivity Score (CS) 1 and 2 using manual and automated coding. A coefficient of .99 (CS1) and .94 (CS2) indicates almost perfect agreement between manual and automated coding. In addition, a Pearson correlation is used to compare the manual and automated coding for Connectivity Score 3. A coefficient of .90 (CS3) shows that the manual score is statistically related with automated score at a .01 level of significance. This suggests that the scores calculated using robotic software is reliable.

The dataset is a valuable indicator of how committed IR adopters are to achieving connectivity, reflecting the integrative thinking involved in producing the integrated report.

本面板数据集涵盖了2018至2022年五年间，来自亚澳地区80家、欧洲78家及美洲10家企业的840份整合报告（Integrated Reporting, IR）中披露的资本关联程度信息。样本涵盖依据全球行业分类标准（Global Industry Classification Standards, GICS）划分的11个不同行业的企业。本数据集聚焦于自愿采用整合报告的跨国企业，相关报告均收录于官方整合报告数据库中。

数据提取通过自动化编码软件完成，该软件可对报告中特定资本的数量与出现序列进行系统化编码。该软件基于整合报告框架（Integrated Reporting Framework, IR Framework）中规定的六大资本体系构建关键词列表，六大资本分别为：财务资本、制造资本、智力资本、人力资本、社会与关系资本及自然资本。随后将生成自主计算的资本关联得分，以评估企业的资本关联程度：该得分通过计算每份整合报告中特定语境下编码的不同类型资本的平均数量得出，得分区间为1至6分，具体数值对应不同类型资本编码的平均数量。

额外得分的计算依据为：某一资本被直接披露后，紧随其后披露另一资本的占比。满分100%表示每一项资本披露后均紧随另一不同资本的披露，而最低分0则代表无任何此类关联披露。详细的得分计量规则请参阅Excel文件。

来自产业界的专家与学术界学者已对本研究用于衡量资本关联程度的工具进行了有效性验证。为评估资本关联得分的可靠性，研究团队随机选取了10份整合报告进行人工编码，并基于人工编码结果计算得到关联得分。研究采用加权科恩kappa（Weighted Cohen’s Kappa）评分者间信度检验，对比了人工编码与自动化编码得到的关联得分1（CS1）与关联得分2（CS2）。其中CS1的系数为0.99，CS2的系数为0.94，表明人工编码与自动化编码结果之间存在近乎完美的一致性。此外，研究采用皮尔逊相关分析（Pearson Correlation）对比了人工编码与自动化编码得到的关联得分3（CS3）。CS3的相关系数为0.90，表明人工得分与自动化得分在0.01的显著性水平上存在统计学关联，这表明通过自动化编码软件计算得到的得分具备可靠性。

本数据集可有效反映整合报告采用者对实现资本关联的投入程度，同时体现了编制整合报告时所蕴含的整合式思维。