Can first or last name uniqueness help to identify diaspora researchers from any country?

The data used was a copy of all journal articles in Scopus with a publication date from 2001 to 2021 as downloaded in January-February 2022 using the Scopus Applications Programming Interface (API). The 21 years of the current century 2001-2021 at the time of data collection were chosen to give recent data with a large enough period for robust name frequency information. Scopus was used in preference to the Web of Science for its greater coverage. Dimensions probably has even greater coverage but is less controlled in content. Google Scholar has far greater coverage but no API (Martín-Martín et al., 2021). The Scopus API was used to download records for all journal articles for 2001 to 2021 in January-February 2022.

Author first and last names were extracted from the downloaded Scopus journal article records, when present, as well as Scopus author identifiers and the first country affiliation listed, assuming this to be the main one. All names were in the Latin alphabet, so non-Latin names (e.g., Arabic, Chinese, Greek, Thai) were recorded in Latin versions. No attempt was made to correct any name errors and it is likely that given/first and family/last names were occasionally reversed in countries, such as China and Malaysia, where it is common to write the family name first. Only the first part of each last name was extracted in cases where there was a last name containing a space, unless it was one of a few common prepositions (including nobiliary particles): al, de, del, von, von/van der, von/van den, de la. The initial name list was found to be incomplete for The Netherlands and Belgium, so the data for these two countries was recalculated with additional prepositions (op de, van ‘t, ten). Last names in some countries, including Spain are often doubled (and not hyphenated), but only the first part of any last name was used because it seems likely that researchers moving to countries where this is unusual would often shorten their last names. Article titles, abstracts, and keywords were also recorded from the API at the same time (for RQ4).

Unique authors were identified by Scopus author ID, ignoring authors without an ID. Each author’s national affiliation was recorded from the first journal article in the data set. This was repeated for last names, and again for both considering only first authors. For the year range chosen, there were 249 country or region affiliations in Scopus, but data was gathered only for the 200 with the most researchers due to software array size limitations. The largest omitted territory was the British Overseas Territory of Gibraltar (167 journal articles) and the smallest was Saint Martin (French Part) (4 journal articles).

Four datasets were produced, each of which was a frequency table with a list of 200 countries/regions (as recorded in Scopus) as columns and a large set of rows, one for each name (first or last), recording the number of authors with that name in each country/region. Two tables analyzed first authors only (for RQ4) and two tables analyzed all authors (for all other RQs). From these tables, the proportion of each name occurring in each country/region could be calculated. The results report this information in various different ways, as described in the figure and table captions.

The results refer only to the declared affiliations of authors of Scopus-indexed journal articles 2001-2021. There are huge variations in the extent to which the first and last names of researchers in a country/region are never found in other countries/regions (Table 1, Table 2: columns 3 and 7) during the period 2001-2021. In terms of (purely) “national” names (defined here to be names only used by the researchers of one country), whilst 81% of Thai-affiliated authors of Scopus journal articles 2001-2021 have (purely) national last names, the same is true for only 1% of authors in Macao, China. First names have smaller differences: whilst 60% of authors from Laos have national first names, none from the Cayman Islands do. Overall, however, it is uncommon for a last name to be national, and very uncommon for a first name to be national.

Some names may occur predominantly in one country, even if they occasionally occur in other countries due to migration or other reasons. These might be termed partly national names. The “percentage national” or “percentage [country]” of a name is defined here to be the percentage of authors with that name who are affiliated with a given country (not reported in the tables). Thus, Babatunde is 57% national for Nigeria (or 57% Nigerian) since 150 of the 263 Babatundes in Scopus have a Nigerian affiliation.

There are substantial international differences in the extent to which first and last names tend to occur predominantly in one country/region (Table 1, Table 2: columns 4 and 8). On average, the last name of a researcher in Thailand will be used 88% of the time in Thailand, but the last name of a researcher in Macao, China will be used only 2% of the time in Macao, China (i.e., 98% of researchers sharing a last name with a Macao researcher are outside Macao, China). Similarly, the first name of a researcher in Japan will be used 87% of the time in Japan, but the first name of a researcher in the Falkland Islands (Malvinas) will be used only 0% (to 0 decimal places) in the Falkland Islands (i.e., virtually all researchers sharing a first name with a Falkland Islands researcher are outside of the Falkland Islands).

At the country level, there is a trend for countries that have a higher percentage of (purely) national first names also to have a larger percentage of (purely) national last names. There are exceptions, however, and the relationship is not close to a straight line (Figure 1). For example, a high proportion of researchers’ last names in Lithuania are national whereas few first names are. In most cases, last names are more likely to be national than first names, but a few countries (e.g., Cambodia) break this trend. On average (averaging across countries rather than names), 9% of first names are national and 19% of last names are national.

Figure 1.

There is a stronger trend for countries with a higher average first name percentage in the country to also have a higher average last name percentage in that country (Figure 2) and the pattern is closer to a straight line. Again, there are outliers (e.g., last names are far more national than first names in Uganda) and last names are generally more national than first names. On average (averaging across countries rather than names), the first name percentage national is 23% and for last names it is 38%.

Figure 2.

This section reviews factors that may lead to international differences in first and last names, extending a previous list (Mateos, 2007). Tables 1 and 2 suggest a size effect on name distinctiveness and there is indeed a trend for countries/regions with greater numbers of authors to have a higher percentage of national first and last name shares (Figure 3, 4), although the pattern is not linear. The trend is a natural side effect of size. If names were distributed randomly around the world, then larger countries/regions would naturally have higher shares of each name so Figures 3 and 4 would be monotonically increasing lines. The substantial deviations from monotonically increasing lines together with the differing distributions shown above confirm that other factors are at work. These include the following.

Figure 3.

Figure 4.

To test the assumption that names occurring more frequently in a country were more likely to be from people with a national origin in that country, additional datasets were constructed as above, but specifying that the country name also had to occur in the title, abstract or keywords. For example, the Nigeria sets summarised the national frequency of first (or last) names of authors of articles mentioning Nigeria. An author was recorded when any of their articles mentioned the country name. Assuming that (a) a name that occurs in a higher proportion in one country is more likely to be from someone with a national origin in that country and (b) people with a national origin in a country are more likely to research that country, then the relationship between the proportion of authors with a name from a country and the proportion of authors with that name researching the country should be linear with a positive slope, even considering only researchers not based in the country.

Consistent with the above assumptions, and therefore partially validating them, there were positive Pearson correlations between national percentage and percentage of non-national authors mentioning the country for the 50 countries with the most authors. The average Pearson correlation was 0.50 for first names (min: 0.14, max: 0.86) and 0.52 for last names (min: 0.09, max: 0.89). Broadly linear shapes can also be seen in the graphs (e.g., Figure 5, others available in the online supplement: https://doi.org/10.6084/m9.figshare.21956795). For example, 0% of the world’s first authors called Rajesh are in Greece and authors outside Greece with a first name similarly rare in Greece have a 0.1% chance of mentioning Greece in any of their articles (abstract, keywords or titles). In contrast, 82% of the world’s authors called Vassiliki have a Greek affiliation but the Vassilikis outside Greece (and the 5 other first names that are 81% in Greece) have an 8.4% chance of mentioning Greece in any of their Scopus articles 2001-2021 (abstract, keywords or titles). This is consistent with the Vassilikis outside Greece tending to have a far above average interest in Greece, such as through a family connection or move. The figure of 8.4% is not low given that the vast majority of research does not mention a country because it is general, theoretical, or mentions a narrower location.

Figure 5.

The countries with first or last names that tend to be a high percentage national (Table 1) are the main candidates for supporting an analysis of the international spread of diaspora researchers (mobile or embedded), although special care should be taken with larger countries (mainly China) because names that coincidentally appear in small and large countries at similar rates would appear to be national to the larger country only. More generally, for all pairs of countries, a popular name might coincidentally appear in a few other countries (e.g., Lee in South Korea is also an Anglophone last name), creating a spurious impression of national variation. An attempt to track international spread of ethnicities through first or last names would also be insensitive to historical factors. For example, many US names originated from the UK centuries ago, and it has little meaning to think of US culture as having “spread from” the UK (and elsewhere). The same is true for long-established Indian and Chinese communities in other countries, such as Singapore, Malaysia, and Indonesia.

This section focuses on the distribution of national name percentages for a country as a key factor suggesting whether a sample of predominantly national names can be extracted from it.

Countries with few international names and a unimodal distribution of name frequency against national percentage are straightforward to use for name-based identification of diaspora researchers, at least from a surface examination. In such cases, a percentage point can be selected from the graph, above which names are likely to be associated with that country. For example, in the case of Turkey, 60% seems to be a reasonable percentage, at the start of the large peak for both first and last authors, but above the more international (but still majority Turkish) Mustafa, which would otherwise contribute a substantial number of false matches (Figure 6). A higher percentage point could be selected to increase the purity of the results if an examination of names associated with, say 60-70%, found them to be often not Turkish but from a culture within Turkey that overlapped with other countries (e.g., Kurdish names, international Islamic names). Other countries with similarly national unimodal distributions include Iran, Russia (last names only), Thailand (first names only), India, Japan, and China.

Figure 6.

An understanding of historical context and cultural overlaps can influence the choice of percentage cut-off. For the unimodal distribution of China (Figure 7), a cut-off of 85% might be chosen rather than about 60% because names in the small peak in the range of 60% to 85% might represent a partial overlap with Taiwan or Hong Kong, China, with their overlapping names (e.g., Cantonese language/culture in Hong Kong, China also appears in Macao, China, South China and diaspora communities). Manual checking of a sample of names would be needed to verify this.

Figure 7.

Countries with bimodal first or last name distributions – many international names and many national names, appearing as separate peaks on a name frequency graph – have the potential for name-based identification of diaspora researchers. The name distribution of each such country needs to be examined separately to detect the appropriate range of percentages to classify as national names. For example, in the case of Greece, it seems that the Greek first and last names tend to occur above about 45% (Figure 8), reflecting wide international mobility of Greek researchers. Thus, a subset of apparently Greek first names could be generated from this set, perhaps with manual checks of a random sample to test this assumption, or with comprehensive manual checks of these names to remove obviously non-Greek occurrences.

Figure 8.

Other countries with bimodal distributions include Finland, Brazil (although fairly flat), Italy (different peaks for first and last names), France (but fairly flat), Germany (weak second peak), Hong Kong, China (but the “national” peak is low and weak, presumably due to overlaps with China), Ukraine (first names only), Saudi Arabia (but very weak national peak), Romania, Indonesia, Israel, Malaysia, Nigeria (with stronger last names), Denmark (but weak flat national peak), Hungary, and the Czech Republic. In some cases, the bimodal distribution may hide other countries within the national peak.

Some countries have unimodal distributions strongly dominated by international names, and these hold little prospect for name-based diaspora researcher identification. If the few mainly national names of countries like the UK (Figure 9) with mainly international names are used to find UK diaspora researchers then the results would be extremely partial due to the small numbers of results as well as overrepresentation of distinctive ethnic groups (e.g., the Welsh for the UK, Māori for New Zealand). Other countries with unimodal distributions and few national names include Argentina, Australia, Austria, Canada, Chile, Columbia, Ireland, New Zealand, Portugal, and South Africa. In the case of South Africa, there are also a few African national names, such as Mthiyane and Tshepo.

Figure 9.

The USA has a slightly bimodal distribution but it is relatively flat, with no obvious cut-off point for US names (Figure 10). Names with high percentages seem to be partly last names that have been anglicized from a European name or that were Central European Jewish names before the Holocaust, perhaps from extended families migrating to the USA where those left behind didn’t survive. There are also first names that are relatively unique to the USA, such as Elwood and Seth. Choosing names above a high percentage (e.g., 80%) might give the potential to identify a small minority of US diaspora researchers internationally, although this would need to be validated by checking a sample of non-US researchers with a high percentage of US names (e.g., Seth, Vogelstein).

Figure 10.

Pakistan’s similarly flat name distribution (Figure 11) also makes identifying national names difficult. This is particularly the case because there are few names that are more than 80% from Pakistan. The flat distribution may be due to the large cultural overlap with India, some Islamic names, and large diaspora populations in the UK and elsewhere. Spain also has a relatively flat distribution.

Figure 11.