Reporting the accuracy of small-scale fishing data by simply applying Benford’s law

Abstract

Small-scale fisheries (SSFs) play essential economic, social, and cultural roles for many fleets in the Atlantic region. The basis of fisheries assessment and management is the landings reported by observers or fishers. Even though the information from the landings is essential, it can be subject to a fisher’s bias, such as the tendency to inflate catches and the refusal to fill the logs. The feasibility for managers using field surveys to validate these datasets is held up by the high costs, shortage in monitoring, and the lack of prior information. Alternatively, Benford analysis overcomes those limitations since it can be applied directly on the logbook data. This method is growing in popularity, underlining its suitability to many economic and biological fields. Hence, in this study, we inspected small-scale fisheries data by the Benford’s distribution, aiming to validate fish landing data from 27 points in Brazil’s northeastern region over 3 years. Our results suggest that 20% of landings data are considered highly imprecise (misinformation), especially from non-motorized canoes. Also, harbors in remote locations provide poorer quality data, specifically monthly catch values reported by several boats. The way we mine our data affects the sensitivity of the analysis, with monthly data being less prone to be accessed by this method than daily information. As the results match our prior knowledge on the location, we endorse the suitability of the method and reliability for assessing accuracy in fishing data. Hence, we recommend that it ought to be used as an audit tool for SSF landing data aiming to enlighten data reliance and support managers for planning management actions.

A Benford’s law-based framework to determine the threshold of occurrence sites for species distribution modelling from ecological monitoring databases

“…For instance, Polidori and Hage 38 applied Benford’s law to evaluate the accuracy of elevation, slope, and stream order from a digital elevation model and found large errors in elevation. Moreover, Noleto–Filho et al 39 demonstrated that the compliance assessment of a Brazilian fishing dataset with Benford’s law could identify the cause of unreliability. These findings suggest that Benford’s law can be used as a solid criterion for evaluating the reliability of monitoring datasets 40 .…”

Aplicación de la ley de Benford a la pesquería argentina de corvina rubia (Micropogonias furnieri)

“…La ley de Benford es una herramienta utilizada actualmente en la detección de fraudes y manipulaciones en grandes bases de datos (Nigrini 2012). Esta ley se basa en que la ocurrencia de los dígitos en una cifra positiva posee una probabilidad logarítmica.…”

“…A pesar de que la distribución de frecuencias de Benford no es intuitiva debido a su naturaleza no uniforme, ha sido observada en una amplia variedad de conjuntos de datos, incluyendo áreas como finanzas, auditorías, dimensiones geográficas, fraudes electorales, reportes oficiales de casos de COVID-19 y declaraciones de impuestos, entre otros (Diekmann 2007;Mebane 2008;Silva y Filho 2021). En el ámbito de la biología pesquera, la implementación de técnicas estadísticas basadas en la ley de Benford han sido utilizadas para la verificación de los reportes de capturas de la langosta del Atlántico canadiense (Homarus americanus, Graham et al 2010), de ballenas en el océano Pacífico (Cerri 2018), atunes en el Golfo de Guinea (Tsagbey et al 2017) y, recientemente, para validar la información de captura de la pesca artesanal en el noreste de Brasil (Noleto-Filho et al 2022). Sin embargo, la utilización de este tipo de herramientas no es ajena a controversias concernientes con la magnitud de las desviaciones observadas a la ley y el tamaño de la muestra (eventos cry wolf o exceso de potencia).…”

Table_1.XLSX

“…It is important to note that in this instance, the p-value is not a measure of significance but an absolute value to use for comparison to accept or reject the null hypothesis. Since our z-statistics are >2.77 and <0.0056 (Table 1), we will not reject the null hypothesis because both proportionate values equate to each other (41).…”

“…The Z-statistic looks at a two-sided p-value for each observed proportion and the BLs proportion (41). It is important to note that in this instance, the p-value is not a measure of significance but an absolute value to use for comparison to accept or reject the null hypothesis.…”