Spatial bibliometrics on the city level

Spatial bibliometrics has attracted increasing attention in recent years [1,2]. An overview of studies using bibliometric data for spatial analyses has been published by Frenken and Hoekman [3]. A good indication for the recent popularity of spatial bibliometrics is the tendency in university rankings to present the results not only as tabular lists but also on maps. Bornmann et al. [4] visualise institutional ranking data at www.excellencemapping.net: for each institution, the estimated probabilities of publishing highly cited papers and publishing in the most influential journals are shown on maps. The Leiden Ranking (see www.leidenranking.com) provides the map view in which the ranking positions of universities are presented on maps. The users of these and similar Internet-based visualisations receive a regional overview of performance distributions, which is not possible on the base of institutional lists. For example, the visualisation of the institutional data by Bornmann et al. [4] points out a west-east gradient of scientific performance in several subject areas. However, the problem with these visualisations is that institutions are frequently spatially clustered in larger cities whereby institutions are positioned one above the other. Thus, the user is not able to inspect the results in important areas which are characterised by many institutions on the map.

In this Brief Communication, we show two example maps with the object of improving the perception of spatial results. The underlying data are from the Scimago Institutions Ranking (see https://www.scimagoir.com), which is based on Scopus data (Elsevier). As with the Leiden Ranking, the ranking results are visualised on maps. To avoid the problems described above, the publication data for the maps, which are presented in this Brief Communication, have not been aggregated on the institutional level, but on the city level. The city level has the advantage that corresponding shape files (popular geospatial vector data formats for geographic information system software) are available for spatial visualisations. By visualising the data on the city level, the data can be presented on maps without overlaps. Maps on the city level are also interesting in terms of content because ‘science thrives in cities. With two-thirds of the global population projected to live in urban areas by 2050, cities will become even more distinctly the domain of knowledge and innovation’ (see natureindex.com/supplements/nature-index-2017-science-cities/index).

We use data for German cities to demonstrate the city approach for science mapping. Table 1 shows those cities in Germany with the most papers belonging to the 1% most frequently cited papers within their subject area and publication year. The use of the top-x% field-normalised indicators for performance analyses is recommended in the Leiden Manifesto including 10 principles to guide research evaluation [5]. By considering the top-1% papers, the study focuses on the most highly cited papers. Since we base the study with the choice of the city level on larger units, this focus might be justified. However, considering all papers – and not only the top-papers – might lead to different results. The papers (including all subject areas) in this study refer to the period from 2011 until 2015. Only substantial papers have been considered (articles and reviews); thus, papers with document types such as editorial material have been excluded. The 16 cities in Table 1 are a selection from 200 German cities which published at least one paper (considering all papers, not only highly cited papers) in the publication period.

OPEN IN VIEWER

Table 1.

Number of papers as well as number and proportion of papers belonging to the 1% most frequently cited papers for German cities with more than 500 top-1% papers (sorted by number of top-1% papers in decreasing order).

City	Number of papers	Number of top-1% papers	Proportion of top-1% papers
Munich	46,917	1987	4.24
Berlin	56,898	1923	3.38
Heidelberg	32,163	1468	4.56
Hamburg	24,375	1004	4.12
Bonn	20,694	849	4.1
Dresden	20,692	725	3.5
Frankfurt am Main	16,703	653	3.91
Leipzig	16,870	644	3.82
Aachen	17,543	643	3.67
Cologne	18,231	612	3.36
Freiburg im Breisgau	16,765	605	3.61
Mainz	14,234	603	4.24
Tübingen	16,019	591	3.69
Göttingen	17,423	575	3.3
Potsdam	12,995	568	4.37
Erlangen	15,601	560	3.59

In this study, we used the full counting approach to assign papers to cities. This means that a paper was fully assigned to a city if at least one author was from this city. An alternative would have been the fractional counting approach, whereby each paper is fractionally assigned to corresponding cities. For example, if authors of a paper are from three cites, the paper would be assigned one-third to each city. The advantage of the fractional counting approach is that multiple assignments of single papers are considered in the calculations. The disadvantage of the approach is that there exist not only one but also numerous methods of how papers can be fractionally assigned to units [6].

Figure 1 shows the number of papers belonging to the 1% most frequently cited papers on the city (and municipality) level. The cities have been assigned to four performance classes by using the Characteristic Scores and Scales (CSS) method – ‘a parameter-free solution for the assessment of outstanding performance’ [7]. This method was introduced by Glänzel and Schubert [8] for assigning papers in a field and publication year to meaningful impact classes (as a rule four classes). ‘Characteristic scores are obtained from iteratively truncating samples at their mean value and recalculating the mean of the truncated sample until the procedure is stopped or no new scores are generated’ [9].

OPEN IN VIEWER

Figure 1.

Number of papers belonging to the 1% most frequently cited papers for German cities (see the list of cities in Table 2 in Appendix 1). White areas indicate cities without any paper. Some cities are so small that they do not appear as coloured (white) areas, but are masked by black borders.

In the first step, the mean number of top 1% papers has been calculated for all German cities. Cities with numbers below the mean have been classified as ‘below average number’ (light-pink cities in Figure 1; n = 157). Cities with paper numbers above the mean have been used for further calculations in the second step. For these cities, the mean number has been calculated again and the cities with numbers above the mean have been assigned to the category ‘above average number’ (dark pink in Figure 1; n = 27). In the third step, the procedure of mean calculation and separation of two classes has been repeated, which resulted in two further city classes labelled ‘far above average’ (light-red cities in Figure 1; n = 12) and ‘outstanding’ (dark-red cities in Figure 1; n = 4).

The results in Figure 1 reveal that only a few regions in Germany are involved in the production of (highly cited) papers. Out of more than 11,000 German cities (and municipalities), only 150 are associated with the publication of highly cited papers (200 cites have at least one paper published). The largest concentration of cities with scientific activities (on the top-cited level) can be found in western Germany. The cities with the highest numbers of papers belonging to the 1% most frequently cited are especially big German cities: Berlin, Hamburg and Munich. However, this result changes if the size-independent perspective in Figure 2 is used instead of the size-dependent perspective in Figure 1.

OPEN IN VIEWER

Figure 2.

Proportion of papers belonging to the 1% most frequently cited papers for German cities. Only cites with at least 50 papers are considered (see the list of cities in Table 3 in Appendix 1). White areas indicate cities without any paper. Some cities are so small that they do not appear as coloured (white) areas, but are masked by black borders.

Figure 2 shows the proportion of top-1% papers for cities with at least 50 papers. Here, those cities are in the top-group of German cities with a proportion of papers between about 10% and 25% belonging to the top-1% papers. The results reveal that Berlin, Hamburg and Munich belong to different performance classes, whereas Munich and Hamburg are ‘above average’ and Berlin is ‘below average’.

Although the size-dependent perspective in Figure 1 favours larger cities, such as Berlin or Munich, the focus on the size-independent perspective in Figure 2 favours small cities, such as Worms or Neu-Isenburg.

The spatial maps produced for this Brief Communication demonstrate that the analysis of bibliometric data on the city level produces overviews of the performance of countries with high information content. The maps avoid the overlay of performance information from institutional maps, but are more detailed, for example, than the density maps proposed by Bornmann and Waltman [10]. These authors used a raster with 500 rows and 1000 columns and a kernel width parameter of 100 km to visualise the affiliations of authors.

An important disadvantage of our city approach and the regions approach by Bornmann and Waltman [10] is the information loss. If a city has many institutions, the performance differences between these institutions are ignored. A good example is Harvard University, Massachusetts Institute of Technology and Boston University in the Boston (area).