Changing U.S. Output of Scientific Articles: 1988-2003
Trends in Influential Articles and Journals
A publishing center's authorship of prominent S&E articles can be an indicator of the quality or influence of the research results it produces. Two citation measures are used to gauge quality and influence: (1) the frequency with which an article is cited and (2) the frequency of citation to the journal in which the article appears. Better and more influential articles tend to generate more citations. Because articles in frequently cited journals tend to be more widely noticed, these journals are able to attract better submissions and to publish better articles. These citation measures provide an indication of capability and competitiveness at the frontiers of research.
Shares of highly cited articles are used to analyze trends because worldwide counts of highly cited articles fluctuate for reasons unrelated to influence (see sidebar "Methodology for Determining Influential Article Counts by Percentile"). Unless otherwise noted, the top 5% share of cited articles and journals is used as the measure of highly influential research. Results were similar when analyses were conducted using other percentile shares (see sidebar "Trends in Various Percentiles of Cited Articles and Journals").
The U.S. share of the world's highly influential S&E articles slipped between 1992 and 1996 (appendix table 8). It then dropped more substantially, from 56% in 1996 to 50% in 2003 (figure 7 ). The drop paralleled the decline in the U.S. share of published articles that occurred when overall U.S. article output flattened beginning in 1992 (figure 6 ). For the three other major S&E publishing centers, shares of the world's most influential S&E articles increased during this period.
By field, the change in the U.S. share of highly influential articles was largely similar to the change in the U.S. share of total article output (figures 5 and 8 ). In physics and engineering/technology, the United States experienced the largest declines during this period; by 2003 the U.S. shares of highly influential articles were lower in these fields than in all others. Shares in the earth/space sciences and mathematics declined by nearly as much as those in physics and engineering/technology. The decline in the U.S. share of highly influential articles in clinical medicine and biomedical research was less than the overall decline in the U.S. share of highly influential articles. Similarly, those fields experienced less of a decline in their shares of total world article output than did the United States as a whole.
Whole-count trends in the U.S. share of the world's highly influential articles between 1996 and 2003 were similar to fractional-count trends: the U.S. lost share while the shares of the three other major S&E publishing centers increased. The U.S. whole-count share during this period fell from 62% to 58% (figure 8 and appendix table 9). The whole-count decline was less than the fractional-count decline, however. Consistent with the overall trend for all publications, the United States increasingly received credit for highly influential articles by participating in international collaborations.
The U.S. whole-count share declined in all fields between 1996 and 2003, but the size of the decline varied by field (figure 8 and appendix table 9). In three fields—clinical medicine, biomedical research, and the earth/space sciences—more than half of the U.S. share decline disappeared when publications were measured using whole counts rather than fractional counts. This finding suggests that the United States largely maintained its worldwide position in these fields, although it did so by more frequently engaging in international collaboration. The same trend occurred in all U.S. articles in these fields (figure 5 ). In all other fields, however, whole counting produced only marginal differences in the size of the U.S. decline, suggesting that other nations with growing article shares were displacing some of the U.S. share in highly influential publications.
Between 1996 and 2003, the overall decline in the U.S. share of articles published in highly influential journals was similar to the decline in highly influential articles (figure 9 and appendix table 10). The magnitude of the decline differed among fields, however, and these differences in magnitude did not correspond to differences for highly cited articles. For chemistry, clinical medicine, and engineering/technology, the decline in the share of articles in highly influential journals was greater than in the decline in the share of highly influential articles. In contrast, for the earth/space sciences and social sciences, the decline in share of articles in highly influential journals was smaller than the decline in share of highly influential articles.
The percentage of total U.S. article output classified as highly cited is a somewhat different indicator of quality and influence. It indicates whether prominent articles are becoming more or less common in the output that the United States generates. This percentage remained steady between 1996 and 2003 and was higher than would be expected by chance (i.e., above 5%) (table 4 ). Likewise, this indicator remained steady in most fields (appendix table 11). The exceptions were engineering/technology and mathematics, where the percentage of U.S. articles classified as highly cited increased between 1996 and 2003.
The EU-15, Japan, and the East Asia-4
The EU-15's share of the world's highly influential articles rose slightly from 28% in 1996 to 31% in 2003 ( figure 7 and appendix table 8). By field, the EU-15 share of highly influential articles surpassed that of the United States in physics in 2003, and its gap with the United States in mathematics and engineering/technology narrowed considerably (appendix table 8). For articles in influential journals, share trends by field were somewhat different. However, the EU-15's gap with the United States narrowed considerably in engineering/technology and physics, two fields that exhibited similar trends in highly influential articles (appendix table 10).
Japan's share of the world's highly influential articles remained steady, rising slightly in all fields except the social sciences and psychology during the post-1996 period (figure 7 and appendix table 8). Share trends for articles in influential journals were similar in all fields except physics, where the Japanese share increased at a smaller rate between 1996 and 2003 (appendix table 10).
The East Asia-4's share of the world's highly influential articles more than quadrupled but remained very low overall, rising from less than 1% in 1996 to 2% in 2003 (figure 7 and appendix table 8). However, the East Asia-4 share increased considerably in physics, chemistry, and, especially, engineering/technology. Its share of the world's highly influential engineering/technology articles rose from 3% in 1996 to 10% in 2003, nearly equal to Japan's share. Trends were similar for articles in influential journals (appendix table 10).
Among U.S., EU-15, and Japanese articles, the percentage classified as highly influential remained roughly constant between 1996 and 2003. In contrast, the percentage of East Asia-4 articles in this category doubled during this period (table 5 ). Despite this trend, the East Asia-4 ended the period with the lowest percentage of its articles (1.6) in the highly cited group, well behind the United States (8.1), the EU-15 (4.7), and Japan (3.4).
 Citations are not a straightforward measure of either quality or influence. Other factors affect what authors choose to cite; e.g., they are likely to cite their own previous articles and the work of colleagues, mentors, friends, or scientists in their own country. Different fields have different citation patterns. In addition, authors who are members of relatively large disciplinary or national communities are cited more frequently than those in smaller communities.
 Journals are ranked by the average number of citations to articles published in each journal. Because the journal citation ranking is an average, articles appearing in highly cited journals may not necessarily be highly cited.
 Citation counts are based on a 3-year period with a 2-year lag. In 2003, for example, a highly influential article is an article published between 1999 and 2001 that received more citations in 2003 than 95% of the articles published in these years. Because of the lag between publication and citation, the measure of highly influential articles begins with 1992, using data on articles cited in that year that were published between 1988 and 1990.
 In analyses of the entire set of S&E articles, the number of articles in the database as a whole is used to calculate the top 5% considered to be highly influential. Because citation patterns vary from field to field, this method of determining the number of highly influential articles causes articles in some fields to be overrepresented and articles in other fields to be underrepresented. However, when citation counts are analyzed by field, articles are classified as highly influential if they are among the top 5% cited in their field, which partially controls for field differences in citation patterns. Thus, the same article may be classified as highly influential in some analyses and not highly influential in others.