User talk:FishBaseProject

Comments on “Errors in Length-weight Parameters at FishBase.org“ by Cole-Fletcher et al., self-published. Rainer Froese, FishBase Coordinator, 12 July 2011

The manuscript (ms) by Cole-Fletcher et al. takes 6 species of fishes for which length-weight relationship (LWR) parameters were available in FishBase. For LWR studies taken from one author, Carlander (1969, 1977), it compared the encoded information with the source and found that all parameters were correctly encoded, but the type of length measurement was wrongly recorded in 5 cases. It also found that subspecies or diverging phenotypes had not been marked as such by the encoder. The ms then compares multiple LWRs published for the same species and finds that extreme curves produce very different, unrealistic predictions for a length of 30 cm. The ms concludes that “Length-weight tables at FishBase.org are not generally reliable and the on-line database contains dubious parameters. Assurance of quality probably will require a systematic review with more careful and comprehensive methods than those currently employed.”

The ms suggests that a new parameter L1, showing the typical length of a fish at 1 kg weight, would improve LWR and assist in detecting errors.

The ms was useful for the FishBase team, as it reported errors and made suggestions for improvements. The errors have meanwhile been fixed and a second visual tool for detection of dubious LWRs is under development. However, as a scientific paper, the ms has to pass the usual tests: 1) Does the ms present new, relevant knowledge that is of interest to the scientific community? An estimation of error rates in FishBase and description of methods to detect errors would satisfy this criterion. 2) Was the subject well-understood?

There are a variety of reasons for the variation in LWR parameters, reviewed in Froese (2006), which they cite but do not adopt or discuss. These include biases often caused by narrow length ranges or inclusion of early juveniles that have not yet completed metamorphosis, as well as seasonally different condition (weight per length to the power of three) of juveniles and adults. This makes it difficult to draw a line between reasonable and unreasonable LWRs. The ms is apparently unaware of these intrinsic LWR problems.

The ms assumes that the existence of doubtful LWR parameters not marked as such in FB stems from flaws in the methods applied to detect such cases. The ms cites the FishBase error detection tool, namely the log a vs b plot, and concludes that it is insufficient for the job. This conclusion could have been easily verified by the authors, by checking where extreme LWR values show up in the log a vs b plot. If done, all extreme cases show up as such, being far away from the center of an ellipsoid containing the LWR coordinates. Thus, the method is not to blame, only that it has not been applied to all LWR studies. FishBase keeps track of this: of over 9,500 studies, 335 cases have been checked, 323 have been marked as doubtful (YES), 12 have been marked as not doubtful (NO), the rest has not been checked, showing a blank entry as indication that no test has been performed. Thus, it seems that the ms does not understand its subject well, and that its main conclusion, that the current FishBase tool is inadequate, is not backed by data.

3) Were the methods applied appropriate for the conclusions?

FishBase contains over 9,500 LWR studies for over 3,300 species extracted from over 1,200 sources. Clearly, an analysis of about 50 LWR for 6 species from one author is insufficient to make a general statement about error rates. More importantly, there is a general rule in science, repeated explicitly in many LWR papers but ignored by the ms, that regressions such as LWRs shall not be used outside the data range used to derive them. For the six species analyzed in the ms, several of the extreme LWRs did not include length-ranges and should have been excluded from the analysis. Also, some of the stated length ranges did not include the 30 cm mark used for the proposed new method. These cases should have also been excluded from the analysis.

Consequently, the new L1 parameter proposed in the ms is only applicable to fish that reach 1 kg body weight. Thus, it is not applicable to the majority of fishes, which have a median length of about 15 cm and 30 g. Also, the authors have not realized that length at 1 kg will be very different for an eel or a box fish. Thus, the methods applied were not appropriate to support the conclusion. Summary: While the research question posed in the ms is of principle interest, the understanding of the subject and the applied methods were insufficient to justify the conclusions. The authors are, however, encouraged to work with the FishBase team in order to derive sound results for the research question. As for the Wikipedia page on FishBase, the prominent display of the flawed, self-published ms seemed inappropriate, and it was therefore removed by us, with a new sentence about error rates in FishBase, with reference to the ms. We thought that was a fair reaction.