Measuring epistemic success of a biodiversity citizen science program: A citation study

selecting the publication year, the document type, and the journal title). Let us note that some journals are not referenced in WoS, and were not taken into account in the statistical analysis (namely: Diversity; Current Zoology; Nature Climate Change; Peer-Review Journal; Computa- tional management science; Methods in ecology and evolution; Urban ecosystems; Regional Envi- ronment change; British birds; Ecology and evolution; Conservation Letters; Insect Conservation and Diversity). 17 Vigie-Nature articles are then excluded. Besides, we also excluded from the analysis articles published in generalist journals, namely: Science, Nature, Proceeding of the Royal Society: Biology, Biology Letters, Plos One. 13 VN articles were then removed from the analysis. Finally, this second data set includes 78988 articles. Third, we also used the articles from the ‘‘Environment and Ecology” category from Clarivate Analytics. Between 2009 and 2019, this category gathers 841 673 articles. We directly extracted some statistical figures from the data base (citation thresholds for the 10%, 20% and 50% most cited papers. These data are available here: https://esi.clarivate.com/BaselineAction.action). Statistical analysis. All the calculations, and in particular statistical tests were carried out by using Matlab 2019b. The statistics were developed through a longitudinal analysis using the mvregress function in Matlab (specifically conceived to lead this kind of longitudinal analysis aiming to compare temporal series by modeling them under the form of polynomials). With

this function, we operated multivariate regrcessions to fit, with different models, the time- series of yearly citations for the different groups of papers: VN (Vigie-Nature articles), BC (Biology and conservation articles), and VN journals articles. In its more complex form, the regression model takes into account the time T (0 to 12 years), the origin O of the articles (BC, VN or VN journals), and the interaction between time and origin TxO. For a two-degree poly- nomial, the global equation model is the following: yij ¼ bo þ b1Gi þ b2tij þ b3tij 2 þ b4Gitij þ b5Gitij 2 þ εij with yij the value (number of citations) for article i at time tij (in years), Gi the origin of the paper (equal to 1 for VN and 0 for BC or VN journals), εij the error in fitting yij, and (βo.. . .β5) the coefficients to be found. Apart from this global model, two reduced models can be designed: the first one ignores the O variable (MR1), and the other one ignores only the inter- action terms β4 and β5 (MR2). The mvregress function then finds the β terms and the likelihood objective function for these models (the global one and the reduced ones). We then led two kinds of analysis. First, for the general comparison of citation history, we computed the MR1 model and we compared it to the global model which includes O as a variable. We then com- pared these two models in their ability to fit the data, by using a likelihood ratio test (using the chi2cdf function in Matlab). The null hypothesis is that the two models are equivalent, which

would mean that the reduced model without O is sufficient, and thus that the origin of the paper does not determine the citation history. Second, for the analysis of the evolution of the relative citation rate of Vigie-Nature articles, we compared MR2 and the global model (which includes interaction terms). We applied the same likelihood ratio test to determine if the interaction of time and origine had a significant effect on the citation history profile. In other terms, this analysis allows to determine if the dif- ferences measured in the yearly citation rate change in time. The results we obtained are pre- sented in the next section. Results Comparison of citation history We first computed the citation history for the 123 papers extracted from the Vigie-Nature (VN) bibliographic database. We calculated the average number of annual citations as a PLOS ONE A citation study of a biodiversity citizen science program PLOS ONE | https://doi.org/10.1371/journal.pone.0258350 October 11, 2021 5 / 13