ABC | Volume 111, Nº3, September 2018

Viewpoint Santos et al For a model of self-citation governance in ABC Arq Bras Cardiol. 2018; 111(3):423-426 Table 1 – Total of references per year, number of references of articles published in Arquivos Brasileiros de Cardiologia (ABC) and number of valid references for estimation of the impact factor from 2000 to 2016 Year References per article SD References or articles published in ABC SD References of impact SD 2000 28.11 2.01 1.29 0.28 0.00 0.00 2001 28.06 1.85 1.21 0.21 0.06 0.02 2002 29.44 1.52 1.18 0.22 0.18 0.07 2003 29.29 1.52 1.18 0.22 0.18 0.07 2004 29.14 1.25 1.14 0.18 0.19 0.05 2005 31.32 1.52 1.36 0.20 0.06 0.02 2006 30.25 1.01 1.38 0.16 0.20 0.42 2007 27.88 0.70 1.54 0.17 0.30 0.04 2008 25.85 0.81 1.30 0.17 0.35 0.07 2009 27.27 0.72 1.92 0.20 0.39 0.07 2010 29.07 0.55 2.00 0.16 0.31 0.04 2011 29.24 0.89 1.47 0.18 0.46 0.07 2012 29.19 0.75 2.00 0.20 0.50 0.06 2013 28.41 0.75 1.92 0.19 0.30 0.05 2014 29.70 0.84 2.11 0.24 0.31 0.07 2015 29.23 0.84 1.59 0.20 0.38 0.08 2016 29.17 0.84 1.65 0.22 0.46 0.09 Mean 28.87 0.24 1.61 0.04 0.30 0.01 Data expressed as mean and standard deviation (SD) Stability of all parameters selected in ARIMA model was assessed by estimation of eigenvalues and their graphical display inside the unit circle of the inverse root of the ARIMA polynomials. Significance level was set at two-tailed p < 0.05. Statistical analysis was performed using Stata software (version 14.2). Results of the analyses are described in four stages, as follows: Stage 1. A total of 1,875 articles were analyzed, corresponding to all “original articles” published between 2000 and 2016 in ABC. Table 1 describes the number of references per article, the number of references of articles previously published in ABC and the number of “valid” references for the impact factor calculation. Data were described per year, as mean and standard deviation. Stage 2. Numerical and graphical analysis of autocorrelations did not reveal temporal, periodic or seasonal trends. Similarly, results of the Q-test were compatible with white noise in models in which lags were not included (p = 0.49) and in models with inclusion of up to 20 lags (p = 0.27), i.e., there was a random change of signal, with no temporal trends or autoregressive phenomena associated with such variation. Stage 3. ARIMA regression models were tested. The model that met the adequacy criteria, yielding the lowest AIC values, used, as parameters, a “p” (lags) of 6, a first‑order “d” (difference) of 1, and “m” (or cut offs after lags, in “leads”) of 3. All coefficients had a p-value greater than 0.05, including the first-order difference and sigma, which test the hypothesis of variance in time series different from zero. The model stability was considered satisfactory in numeric terms, for showing eingenvalues lower than 1 in absolute number, as well as in graphical terms for showing the inverse root of polynomials inside the pre-established circle. Stage 4. A sequential, six-month window was adopted as parameter in the rolling window regression. A Poisson regression model was used for analysis of countable data, with a robust estimate of both variance and covariance. Distribution of IRR of “valid” references is depicted in Figure 1. An IRR near 1 indicate absence of volatility, i.e., considering the total number of references, the rate of “valid” references did not significantly change throughout the analysis period. Therefore, the pattern of bibliographic referencing remained unchanged in the last 17 years. As shown in graph 1, IRR was extended by nearly 5%, ranging from approximately 0.98 to 1.04 from 2000 to 2016. Analysis of time series during a 17-year period enabled a detailed description of parameters of bibliographic referencing distribution and self-citation pattern. These data may serve as a basis for future comparisons between different journals and within the same journal. Our main finding was the stationary pattern of self-citations in the bibliographic referencing of original articles, published in ABC between 2000 and 2016, considered “valid” for the impact factor calculation. This suggests that ABC resisted the temptation to encourage self‑citation of their reports to increase its impact factor. In Brazil, the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) a Brazilian government agency run 424