ABC | Volume 111, Nº3, September 2018

Viewpoint For a Model of Self-Citation Governance in Arquivos Brasileiros de Cardiologia Marcos Antonio Almeida-Santos, 1 Deyse Mirelle Souza Santos, 1 Beatriz Santana Prado, 1 José Augusto Barreto-Filho 2 Programa de Pós-graduação em Saúde e Ambiente da Universidade Tiradentes, 1 Aracaju, SE – Brazil Núcleo de Pós-Graduação em Ciências da Saúde da Universidade Federal de Sergipe, 2 Aracaju, SE – Brazil Mailing Address: Marcos Antonio Almeida-Santos • Avenida Gonçalo Prado Rollemberg, 211 Sala 210. CEP 49010-410, São Jose, Aracaju, SE – Brazil E-mail: maasantos@cardiol.br , marcosalmeida2010@yahoo.com.br Manuscript received October 23, 2017, revised manuscript December 04, 2017, accepted December 08, 2017 Keywords Periodical Index; Periodicals as Topic; Cardiology; Citation Databases; Journal Impact Factor. DOI: 10.5935/abc.20180160 Arquivos Brasileiros de Cardiologia (ABC) has been the official scientific publication of Brazilian Society of Cardiology ( Sociedade Brasileira de Cardiologia, SBC ) since 1948. Since its beginning, ABC has continuously published articles on a wide range of topics in cardiology, becoming the main organ of dissemination of scientific work in Brazil and Portuguese-speaking countries. Since 1950, the articles have been indexed in the main international databases (Institute of Scientific Information - ISI Web of Science; Cumulated Index Medicus - MEDLINE; Pubmed Central; EMBASE; SCOPUS; SCIELO and LILACS), 1 and are currently published in two languages (English and Portuguese). Thomson Reuters publishing company, owner of the Journal Citation Reports (JCR) statistical database, created the concept of impact factor in 1955, aiming to provide an instrument to evaluate the performance of scientific publications in a comparative and quantitative manner. Impact factor is calculated by dividing the number of citations of articles published in an academic or technical journal indexed in the ISI by the total number of articles published in that journal during the two preceding years. These estimates include approximately 3,300 journals, 200 subject areas and 100 countries. 2 Impact factor is a metric for journal assessment, widely used by researchers to choose the best journal to submit their papers. However, its validity as an indicator of scientific impact has been questioned due to, among others, self-promoting journal self-citation in attempt to increase their impact factor. 3 Overuse of self-citation practices is punishable, and caused the exclusion of 5 Brazilian journals from the JCR for one year (2013). 4 For this reason, efficient control measures of this practice should be developed and implemented by scientific journals that seek to adhere to ethical precepts recommended by the scientific community. In light of this, we developed an original model of regulation of self-citations of scientific publications, by analysis of ABC publications in terms of internal bibliographic referencing, known as self-citation, considered valid for the impact factor calculation, between 2000 and 2016. The search was performed on the ABC database available at (http://www.arquivosonline.com.br) . Inclusion criteria were the period from 2000 to 2016, and the texts classified as “original articles”. Aiming to avoid heterogeneity of data and eventual publication bias, manuscripts classified as “editorials”, “letters to the editor”, etc. were excluded from the search. Original articles were classified by month, year, total number of references, number of references of articles published in ABC, and number of “valid” references for impact factor calculation, i.e., articles published within two years prior to current publication. Temporal trend analysis was performed for the number of “valid” publications and its relation to the number of citations in the same journal and the total number of references. Assumption of temporal stability was assessed by rolling window regression, including the following parameters – monthly periodicity, non-recursive sampling (i.e., a fixed window size), overlapping window of subsamples in a six‑month range beyond sample size, fixed at 1,875, of “original articles” published between January 2000 and December 2016 in ABC. Rolling window regression was used for analysis of countable data (Poisson distribution), with the number of references considered “valid” for calculation of the impact factor, per month, used as dependent variable, and total number of references per month used as independent variable. Coefficients obtained from successive intervals of six months were exponentiated to represent the “effect size” of temporal series as incidence rate ratio (IRR). Values near 1 (margin of error of 5%) for a pre-established period indicated absence of the effect. Values lower than 1 indicated reduction and values greater than 1 suggested increment in incidence rate, with valid references. To evaluate potential influence of journal volume proximity, previous trends, future predictions, preference for certain periods of the year and stationary processes, autoregressive integrated moving average (ARIMA) model was selected according to pre-estimation and post-estimation. Pre-estimation was assessed by periodgrams, correlograms, partial and total autocorrelation plots with 95% confidence interval based on the Q test (or “portmanteau”) and the Bartlett's test. Post-estimation was assessed using the Akaike information criterion (AIC) and smoothers for detection of the Gaussian white noise in time series graphs. These are characterized by a trend for asymmetry, lack of correlation with time, and presence of stationary processes. Result of ARIMA model was described according to time operators, such as “lag”, “lead” and “difference”. used in the analysis of the best performance. 423