Abstract

Accurate fisheries data are fundamental to fisheries management. This ensures accurate stock assessments and equitable allocation of fishing quotas, eventually leading to sustainable fisheries management. In many low- and middle-income countries, such as Malawi, fisheries statistics of commercial or large-scale fisheries rely heavily on secondary reporting systems, such as logbooks maintained by the Government through Department of Fisheries. There have been reports that large-scale fishers underreport catch and effort data in their logbooks submitted to the Department of Fisheries (Government of Malawi). This study was conducted among large-scale fishers to assess the level of underreporting of catch and effort data in logbooks. Data on catch and effort were collected from large-scale fishers in the 2023 fishing season (primary data) and compared with historical data (secondary data) from the same period on catch and effort obtained from logbooks at the Department of Fisheries.

The density plots and Q-Q plot of the secondary and primary catch data were normally distributed. The Shapiro-Wilk p-values of the secondary and primary catch data showed that the datasets were normally distributed, with p-values of 0.2871 and 0.8355, respectively. The boxplot showed that the primary catch data had a higher mean, wider standard error, and range than the secondary catch data. The two-sample t-test showed a p-value of 0.2359, implying that there was no significant difference between the two groups, although physical observation of the two dataset means in the boxplot showed that the primary data was higher than the secondary data. This difference could be a result of variability in the data collection methods.

However, to make the comparison more robust and comprehensive with the Monte Carlo machine-learning approach, nonparametric methods (permutations and bootstraps) that work well even with normally distributed data were employed. The p-value obtained from the permutation test (0.202) indicated that the two datasets were not significantly different, implying that the observed physical differences between the groups could be random.

The original bootstrap statistics were approximately -959.8, with a standard error of approximately 728.28, indicating that, on average, the secondary catch data were lower than the primary catch data by approximately 960 units. The large standard error indicates that there could be substantial sampling variability, which should be studied and improved, especially in the secondary catch data. The CI included zero, suggesting that the observed difference in means was not statistically significant at the 5% level. While the point estimate was negative, the broad CI shows that the effect might be negligible or even in the opposite direction, depending on how each of the datasets was collected. This implies that the claims that commercial fishers in Malawi underreport the logbooks submitted to the Department of Fisheries may not be true or significant. Primary data verification was performed by comparing the data with what the fish traders reported having bought from individual fishers. It would be ideal to license fish traders to collect data from them, which could be used to validate the logbooks submitted by fishers to the Department of Fisheries.