Empirical analysis and classification of database errors in Scopus and Web of Science

In the last decade, a growing number of studies focused on the qualitative/quantitative analysis of bibliometric-database errors. Most of these studies relied on the identification and (manual) examination of relatively limited samples of errors. Using an automated procedure, we collected a large corpus of more than 10,000 errors in the two multidisciplinary databases Scopus and Web of Science (WoS), mainly including articles in the Engineering-Manufacturing field. Based on the manual examination of a portion (of about 10%) of these errors, this paper provides a preliminary analysis and classification, identifying similarities and differences between Scopus and WoS. The analysis reveals interesting results, such as: (i) although Scopus seems more accurate than WoS, it tends to forget to index more papers, causing the loss of the relevant citations given/obtained, (ii) both databases have relatively serious problems in managing the so-called Online-First articles, and (iii) lack of correlation between databases, regarding the distribution of the errors in several error categories. The description is supported by practical examples concerning a variety of errors in the Scopus and WoS databases.