Correct attribution of scientific publications with the countries of authors is vital for many scientometric tasks. Despite this, the Scopus and Web of Science databases contain a significant number of publications without affiliation with any country, which leads to their loss in the analysis results. This includes scientific types of documentsarticles, reviews, conference proceedings, etc. It has been shown that the practice of excluding such publications from consideration in scientometric studies can lead to a significant distortion of their results. We propose a method that allows predicting, with reasonable accuracy, the connection between a publication and a country based on the affiliation history of the authors. This article analyses the significance of losses due to unaffiliated articles for the top 20 countries in terms of the number of publications indexed in Scopus over the past 20яyears. The importance of this problem has been declining in recent years, but is still noticeable especially for developed countries.