Percentile and stochastic-based approach to the comparison of the number of citations of articles indexed in different bibliographic databases

Recent studies have shown that the coverage of Scopus and Web of Science (WoS) databases differs substantially. Consequently, the citation counts of a paper are different depending on the database used, making it difficult to apply both together. To address this problem, this paper aims to examine whether the percentile- and stochastic-based approach is effective for converting citation counts between two databases while guaranteeing its time-normalization. For this analysis, we collected a dataset of 326,345 papers, published in 1987–2017 in the top 10% source titles of the following fields: Industrial and Manufacturing Engineering, Aquatic Science, Social Psychology and Archaeology. First, we applied the linear regression model to the citation percentiles of indexed papers in both databases. Secondly, we used the predicted results of this linear dependence, combined with the Monte Carlo simulations, to obtain the probability density function of a percentile from papers in the database in which they are missing. The results indicate that, with the method proposed in this paper, it is possible to convert the citation counts of articles between Scopus and WoS. In addition, it also predicts the citation impact of a missing paper on one of those databases, based on the citation impact on the other database. Tests on subsamples, using Lin’s concordance coefficient, suggest substantial agreement between estimated and real citation values. This allows the combined use of the citation counts of two databases, improving the coverage and accuracy of both bibliometric studies and bibliometric indicators.