A bar chart of researcher counts makes it easy to see where more researchers are based. But would the countries appear in the same order if those counts were divided by population?

An R analysis of clinical medicine by Loess, published on November 1, 2026, provides a concrete example. It uses the highly cited researchers database commonly called the Stanford–Elsevier top 2% list. Stopping at raw counts, however, can hide assumptions embedded in denominators and missing values.

Following the path from importing the table to comparing countries helps distinguish three questions: how many listed researchers there are, how concentrated they are relative to population, and how a field fits into a country's listed research community.

Separate the 2026 edition from the citation period

Start by fixing the table being analyzed. The official 2026 dataset, Version 9, was published on October 7, 2026 and uses a Scopus snapshot from August 1, 2026. Its single-year table measures citations received during 2025. Publication year and citation year are different.

Career-long and single-year results are separate tables. This analysis uses the single-year table; substituting the career-long table can change who is counted, even if the output is still called a country comparison of clinical medicine researchers.

The top 2% label does not mean simply taking the highest 2% of one worldwide ranking. The 2020 database update paper describes including the top 100,000 by a composite citation indicator, plus researchers in the top 2% of their main subfield. This accommodates researchers in fields with lower citation densities who might otherwise miss the overall threshold.

Counting rows therefore measures listed researchers who meet the selection criteria, not every researcher affiliated with a country.

Preserve the distinction between observed and filled countries in R

Even the Excel import can change what gets counted. readxl defaults to the first sheet when no sheet is specified. The workbook used in the analysis starts with column definitions, so reading its records requires an explicit read_excel(file, sheet = "Data").

The cntry column represents the country of the most recent institution, not nationality. The analysis fills missing values from top_cntry_career, the country appearing most frequently on the researcher's papers.

dplyr's coalesce() selects the first non-missing value at each position. Its ability to fill a gap does not make the columns semantically identical. The resulting grouping mixes recent institutional location with a fallback based on publication history.

Keeping the original country and a flag for filled values makes comparisons with and without that fallback possible. After restricting the main field to Clinical Medicine, count() counts rows per group unless weights are supplied. Dropping missing countries before aggregation reduces the population being counted, so retain the excluded count alongside the results.

Dividing by population changes a count into a density

In the published analysis, the United States leads in raw counts, while Switzerland has a higher count relative to population. These are figures reported in the analysis, not a new full-data computation here. The comparison uses 2024 population and includes countries with at least one million residents.

Country	Listed clinical medicine researchers	Per million residents
United States	28,458	83.7
Switzerland	1,345	About 149

The calculation is listed researchers divided by population, multiplied by 1,000,000. Using the World Bank total population indicator, SP.POP.TOTL, answers a question about density relative to residents. A selection rate among researchers would require a separate, consistently defined count of all researchers.

Joining population data can also change the comparison set. Under dplyr's join semantics, inner_join() removes unmatched rows. Starting with left_join() preserves the country counts and lets you inspect missing population values before choosing the comparison set.

Duplicate country keys in the population table can multiply rows as well. Reduce that table to one row per country, then state how missing values and the population threshold are handled before calculating ratios. This makes the membership of the comparison traceable.

Choose counts, population ratios, or field shares to match the question

With those processing decisions visible, the purpose of changing denominators becomes clearer. A country's listed clinical medicine researchers can support several distinct comparisons.

Question	Calculation
How many listed researchers are there?	Listed clinical medicine researchers
How concentrated are they relative to residents?	Listed clinical medicine researchers ÷ population
How prominent is clinical medicine among that country's listed researchers?	Listed clinical medicine researchers ÷ all listed researchers in that country

For the final ratio, do not filter to clinical medicine before building the denominator. Using that filtered count on both sides produces 100% for every country, rather than a field comparison. Nor does the field distribution among listed researchers establish the distribution of research funding or of all researchers.

Citation indicators should not be equated with research quality. The official dataset description warns against interpreting inclusion as absolute evidence of excellence. Population normalization does not change the fact that the numerator comes from a citation-based list.

The opening question now has an answer: raw counts and population ratios can produce different rankings because they answer different questions. In R, fix the dataset edition and period, preserve the trail of filling, joining, and exclusion, and choose the denominator for the question being asked. That is what turns a ranking into an explainable comparison.

Source

• Title: Top-cited scientists in clinical medicine by country: the Stanford–Elsevier top 2% list in R
• URL: https://www.r-bloggers.com/2026/11/top-cited-scientists-in-clinical-medicine-by-country-the-stanford-elsevier-top-2-list-in-r/