Sci-Hub has gained fame and notoriety for enabling free access to over 45 million paywalled articles and book chapters, purportedly collected through use of institutional log-in credentials. The site has been operational since 2011 and has received widespread attention recently, partly due to a lawsuit threatened by Elsevier, that considers it a major breach of copyright.
In April of this year, John Bohannon , a freelance journalist working for Science magazine, in cooperation with Alexandra Elbakyan, Sci-Hub’s founder, released and analysed 6 months of Sci-Hub download data for the period September 2015-February 2016 (see: Who’s downloading pirated papers? Everyone). The data consist of DOIs (digital object identifiers) of the material downloaded, date and time of download, and attributed location of download (city, country and GPS coordinates). Location of download was determined by IP geolocation; no individual IP addresses were released. The full dataset is available from Dryad under a CC-0 license, enabling full re-use.
This data enables an enquiry into what it is that people are turning to Sci-Hub for. Do people use Sci-Hub to get papers they do not otherwise have access to, or do they (also) go to Sci-Hub for convenience: a one-stop shop to get access, without having to navigate library and publisher websites?
To investigate this from the perspective of my own university in Utrecht, I first tried to estimate the proportion of Sci-Hub downloads from a city like Utrecht that is likely to come from academics with access to university journal subscriptions. This is reported in this blogpost. Second, I looked at the percentage of Sci-Hub downloads from Utrecht that would have been available through our university journal subscriptons. This will be reported in part 2 of this blog series.
Sci-Hub downloads from the Netherlands- university vs. non-university cities
I first made a subset of all Sci-Hub download data attributed to The Netherlands. This dataset is available on GitHub for others to analyze as well. Due to its manageable size it can be opened in a regular spreadsheet program.
My first question, as described above, was whether downloads from university towns can be assumed to at least partly represent academic use, i.e. use by people that have institutional access to a university’s journal subscriptions. While there are no direct data on this, there are some indications that it is indeed the case.
Looking at the cities with the most downloads (> 1000) (excluding Amsterdam, see below) all university cities in the Netherlands make an appearance (Figure 1) . Only a few other cities have such a high download count. The fact that also smaller university cities, like Delft and Wageningen, have high downloads could be seen as further indication of a role for the academic community in requesting these downloads (although these university cities also have a lot of science tech companies , with no access to university suscriptions but potentially a sizeable demand).
Figure 1 Sci-Hub downloads from Dutch university and non-university cities
A note on IP-geolocation and Amsterdam
IP-geolocation is imprecise at best, and downright unreliable at worst (see also Interpreting Sci-Hub data
). A prime example of this is Amsterdam, which on its own is responsible for over half of all downloads from the Netherlands and which has 20-30 times as many downloads as other major Dutch cities. A plausible explanation is that for many ISPs (internet service providers) IP-geolocation likely defaults to Amsterdam rather than to the actual location of the end user. This would mean that downloads attributed to Amsterdam could come from anywhere in the Netherlands.
To further investigate the relationship between population size and number of downloads, I also plotted the number of downloads per city to the number of inhabitants (Figure 2, inhabitants per municipality, source: https://nl.m.wikipedia.org/wiki/Tabel_van_Nederlandse_gemeenten). Here again, university cities generally come out on top, while many cities that are comparable in size have much fewer downloads. For the largest cities in the Netherlands, this analysis is obviously hampered by the fact that they almost all have universities, the one exception being Den Haag.
Is this one exception enough to disprove the hypothesis of a relation between number of Sci-Hub downloads and the presence of an academic community, at least in larger cities, or are there confounding factors that contribute to the download count in Den Haag, specifically? Among these could be the fact that Den Haag is the seat of the government as well as home to many international organizations. A closer look at the type of papers downloaded might give some indication of the communities these downloads are coming from, but this was beyond the scope of my investigation.
Among smaller cities, there are also some notable exceptions, like Zwijndrecht, Heemstede and Bergschenhoek. Although there might be a correlation with the presence of research companies in these cities, without doing a comparative analysis of cities with and without such companies, this is no more than a conjecture.
In general, despite obvious caveats as the unreliability of IP-geolocation and the inability to distinguish between academic and non-academic use on the basis of these download data, these data lead to a cautious assumption that downloads from university cities could to some degree be attributed to the academic community in these cities.
Identifying academic use of Sci-Hub – other data
Bastian Greshake, a German PhD student, has done some more work on identifying academic vs. non-academic use, with Sci-Hub data linked to aggregated institutional IP-addresses at country level (included in his article ‘Correlating the Sci-Hub data with World Bank Indicators and Identifying Academic Use‘ on the Winnower). He was able to do this because he requested and received additional data from Sci-Hub. These data show that, averaged over 18 ten-day periods, 4% of downloads from the Netherlands can be attributed to university IP-addresses. This is probably a lower bound due to the age and possible lack of completeness of the list of IP-addresses used*.
We can use these data to estimate the percentage of downloads in a Dutch university city, like Utrecht, that come from within the university’s IP range. For this, we need to make some assumptions on the large batch of Sci-Hub downloads originally attributed to Amsterdam, namely:
- the majority of downloads attributed to Amsterdam are attributed only to the ISP-level, and could be coming from anywhere within the Netherlands;
- the number of downloads directly attributable to Amsterdam is to be estimated at 5000 (slightly larger than that of Rotterdam, the second largest city in the Netherlands).
This leaves us with approximately 65000 downloads that can be directly attributed to specific cities/towns (168,877 total minus 109,131 ‘Amsterdam’ downloads plus 5,000 downloads estimated to be directly attributable to Amsterdam). Of these 65000, approximately 30,000 (25,762 from university cities except Amsterdam plus 5,000 added for Amsterdam) are from university cities. If (and this is a big ‘if’) we furthermore assume the set of 104,131 downloads attributed to ISPs is in reality similarly distributed across university- and non-university cities and towns, we can estimate that 45% of Sci-Hub downloads in the Netherlands come from university cities.
Together with the data on aggregated university IP-addresses (4% of downloads in the Netherlands coming from university IP addresses), this would suggest that, as a lower bound, approximately 9% of downloads in university cities like Utrecht are from university IP addresses (Figure 3).
Figure 3 – Sci-Hub downloads from Dutch university IP-addresses and university cities (numbers are rounded estimates)
Actual use of Sci-Hub by people that have institutional access might well be higher – think of academics and students working from home and not using their institutional access (e.g. via proxy or VPN), but going to Sci-Hub instead. Even with this in mind, contribution of downloads from university IP-addresses in university cities seems small in relation to the observed difference in downloads between university cities and non-university cities. An explanation unifying these observations can be that being a university town means much more than just having a university IP-range within your municipial borders. It also means having a highly educated population and a concentration of scientific information-intensive companies and institutions (incl. universities), which together could explain the results of the scatterplot (Figure 2).
* I have not checked this list for presence and accuracy of all Dutch university IP addresses – if someone would be inclined to do so I would be interested in the results. The calculations provided in this paragraph are really meant to be very rough estimates.
This leads us to…
With all necessary caveats, the available data obtained from Sci-Hub by John Bohannon and Bastian Greshake support the idea that Sci-Hub downloads from a university city like Utrecht can to a certain degree be attributed to download requests from the academic community. To further investigate whether use of Sci-Hub is mainly a case of access or (also) of convenience, I examined whether the individual Sci-Hub downloads attributed to Utrecht were also available through other means, e.g. through institutional journal subscriptions. Read more about that in part 2 of this series!