Brouhaha: There’s a bit of a brouhaha underway in Canada following the publication of a paper by members of Fisheries & Oceans Canada titled ‘Spatiotemporally dependent relationship between salmon lice from salmon farms and infestations on juvenile Pacific salmon in BC Canada’ (Jeong et.al. 2025) The paper says that a mixed effects regression model revealed a significant positive relationship between farm derived infestation pressure and sea lice prevalence on wild salmon’.
According to activist and ‘sea lice expert’ Alexandra Morton, an earlier version of the paper had concluded that salmon farms do not elevate sea lice infestations on wild salmon. In a subsequent complaint to the Canadain government, sixteen scientists, who Ms Morton says are highly qualified, deemed the results to be unsubstantiated because of the omission of key data, which Ms Morton suggests was a red flag of potential scientific fraud. When the internal documents were finally released, it appears that a senior scientist had changed the original wording from ‘do influence infection’ to ‘sea lice infection of wild salmon could not be explained by lice in salmon farms’. Those opposing salmon farming were outraged by this revelation.
Ms Morton has demanded that amongst other actions, a review of the scientists involved should take place with consideration of potential scientific fraud, a trend she says is recognised since the tobacco industry tried to reduce the claimed impact of smoking.
I had previously read about the activist complaints about DFO science but when I received this paper, I didn’t realise that this was the research that was the centre of these accusations. I just looked at it for what it was.
The paper measured numbers of sea lice on salmon farms by estimation and standardisation to account for differences in location and fish numbers as well as for missing data. For wild salmon, only certain species were included in the measurement as were certain stages of sea lice. In addition, samples with few fish were ignored. This meant that 15% of the samples were discarded. In total 21,180 fish were included in the study. I will return to this number later.
Finally, the numbers were treated with a mixed effects logistic regression model. The paper includes the complex formula used from which (some) researchers concluded that there is a significant positive relationship between farm -derived infestations pressure and sea lice infestation on wild fish.
My own conclusion is that whether the researchers decide that there is a positive relationship or not, the way that the decision was reached was far too complex. This is not surprising given the increasing focus on modelling in sea lice research. The analysis has become far too remote from the biology, the ecology and the understandings of parasite interactions.
I mentioned that the paper had identified 21,180 wild salmon to be used in the study. These had all been caught and inspected for sea lice infestation over a period of eight years from 2016 to 2023. Fo me the most important statistic is that just 12.7% of these fish were infested with sea lice meaning that 87.3% of the sampled fish were lice free.
It is a real puzzle as to how any conclusion that there is a positive relationship between farms and wild fish be drawn when only 12.7% of the fish were infested. Interestingly, the paper includes a graph showing the infestation levels and it is clear that the distribution is a classic aggregated distribution of parasite infestation, but it does not seem that this is considered in this paper. Instead, they seem to treat it as a statistical expression referring to it as an over-dispersed negative binomial. They both might look the same graphically, but they are in fact worlds apart.
My own view is that no conclusion can be effectively drawn from this study. It relies too heavily on a certain interpretation of the narrative and not enough on the fact that sea lice are a living entity rather than a modelled concept.
Although I believe that this paper does not contribute to the discussion about sea lice and wild salmon, there is one interesting aspect to this paper which is of interest. This is that there is recognition that sample size has an impact on the outcome of any analysis. In this case, the researchers highlighted that the number of fish sampled ranged from 1 to 60 with an average of 15 fish and the most common number (mode) being 30 fish. When they applied filters to the samples, such as which species of fish, which life stage of louse, they also discarded all samples of less than 10 fish (9.3% of samples) as these were not considered representative.
I have not really seen consideration of sample size in sea lice research papers even though in Taranger’s original 2012 report he recommends a minimum samples size of 100 fish. However, he does not clarify whether this sample should be collected in one go or in a number of attempts. After all, a sample made up from 100 different samples is unlikely to be the same as one sample of 100 fish caught at the same time. As it can prove difficult to catch 100 fish for sea lice analysis, it seems most researchers have ignored Taranger’s recommendation. Certainly, the majority of POs were assessed, during the last analysis for the Traffic Light System, on samples of less than 100 fish.
In Scotland, there is a protocol in place which says that there should be a minimum sample size of 30 fish, which appears to be based on a statistical estimate, rather than with reference to parasite distribution. Again, it is unclear, whether the recommended sample size should be caught in one go or whether those who wrote the protocol consider it acceptable to be made up of several smaller samples.
Renewed interest in the question of the size of samples arose this week because I discovered that Scottish government scientists had on 9th June quietly posted an updated version of the spreadsheet detailing lice counts on wild sea trout in Scotland. The previous version had run from 1997 to 2019, and this has been amended to 1997 to 2024 adding another 4,440 records from 114 locations. This is up from 100 locations listed in the previous spreadsheet. Between 1997 and 2024, there were a total of 1,781 separate samplings. Of these 18 samplings are recorded as catching no fish although I suspect that there are actually many more cases of zero catches that have not been recorded.
The new spreadsheet is still riddled with errors and inconsistencies which seemingly Scottish government scientists appear unable to resolve. For example, the samples labelled 2025 which I have previously highlighted remain in this spreadsheet. There are also issues with locations such as with Flowerdale in Wester Ross. There are actually three separate entries for Flowerdale. These are:
Flowerdale – sampled 2010 to 2017 and then again in 2022
Flowerdale Est – sampled 2018 to 2019, 2021, then in 2023 and 2024
Flowerdale Estuary – sampled 2019
I would suggest that these are all the same location. I wonder that if such simple errors are ignored, then how can any other data or data handling be trusted.
However, ignoring the errors, then what is of most interest from the dataset is that only 18.5% of the samplings met the 30 fish minimum catch demanded by the protocol. This means that 81.5% of the samplings should be considered invalid.
Whilst the BC sample found that 87.3% of fish were lice free, the comparable Scottish data is that 47.5% of the fish were free of lice. The difference might be explained by the fact that those sampling in Canada work for an independent research company, whilst in Scotland, those sampling work for one of six Fisheries Trusts aided by volunteers. Whilst they all work to a protocol, it will be inevitable that there will be differences between the way they work. For example, it is clear from the data that some of the trusts are much better at catching larger samples of fish than others. In addition, the many more small samples will also influence the spread of the distribution. Sixty percent of the samplings caught ten fish or less. Nearly twenty percent of samplings caught just one fish. Finally, the Fisheries Trusts work on behalf of the angling sector therefore may not be considered independent. It will be interesting to see whether SEPA have opted to recruit an independent company to continue these sea lice counts or whether they stick with the Fisheries Trusts.
The controversy about the Jeong et.al. (2025) paper and their claim that there was a positive relationship between salmon farms and sea lice infestation on wild fish reminded me that last year, Scottish government scientists also published a paper that claimed a significant positive association between lice on salmon farms and juvenile lice on wild trout (Ives et. al. 2024).
When the paper was published, I did highlight some simple issues such as the paper claimed a total of 5042 wild trout caught from 31 sites. However, the raw data indicates that the fish caught from 31 sites actually totalled 5043 fish. It is a minor point, but it is inaccurate. The paper also highlighted that 67% of the fish sampled carried no juvenile lice at all which begs the question how using just 33% of the fish sampled the authors can claim a significant association between these fish and salmon farms.
Although the raw data remains the same in the old and updated spreadsheets, I have revisited all the data and what has promoted me to do so is that whilst the paper states that the sampling was undertaken according to a standardised protocol (SFCC 2009) the detail was omitted. However, the Scottish Government web page providing the link to the new dataset includes the following:
(Note: The link doesn’t actually work).
It is difficult to know whether the protocol is strictly adhered to or whether some use it as a guide, Certainly, the full dataset includes dates from January to October and not just April to September as stated.
Although the dataset runs from 1997 and the paper was submitted in January 2023, the data used runs from 2013 to 2017 because they say that changes in the way the data was presented occurred before and after those two dates. However, it doesn’t say why the way the data reported was changed. The paper also stated that the data was collected between late April and July with the remainder sampled through September. However, the focus of most previous work was the three months of May to July and as just 55 fish were sampled outside this period, my own review of the data excludes these few fish to limit samples to the three key months.
As I have already highlighted each sampling event aims to catch at least 30 fish, which I presume means each netting, but this Is not made apparent in the dataset. Finally, the government statement says that the fish caught should be either post smolt trout or finnock. Finnock are first year sea trout and larger than post smolts. Unfortunately, it’s not made clear what size any finnock might be. However, finnock are now recorded in the annual catch statistics and the average weight of all finnock caught in Scotland in recent years is 330g. This would appear to be much larger than a typical post smolt trout. The Wild Trout Trust describe a typical finnock as weighing 225g which is still quite large. Regardless, the sea lice spreadsheet does not describe the samples taken by weight but rather by length.
The previous paper that explored the Scottish sea lice data contained in the wider spreadsheet was Middlemas et. al. (2013), a paper that shares an author with the more recent Ives et. al. (2024). This paper considered sampled fish that might carry detrimental numbers of lice to individual fish, and they used a threshold determined in the paper by Wells et. al. (2006). (Alan Wells is now head of Fisheries Management Scotland representing wild fish interests). This paper determined that 13 mobile lice might negatively affect the fish if the fish were up to 70g in weight (although the paper is subject to some unanswered questions. The Middlemas paper then argued that only fish of less than 198mm should be used in their analysis as these equate to fish of 70g in weight. Thus, following on from their determination, if the data used by Ives is adjusted to fish of this size, the number of fish assessed would fall from 5042 to 3351.
Returning to the issue of sample size of 30 fish, then using the paper’s original data, three of the 31 sites should be classed as invalid since the total number of fish caught at these sites is less than 30 fish each. However, when the sample data is analysed for the 3351 samples, then 12 of the 31 sites should also be excluded as none of the separate dated samples reach 30 fish. In total, just 36 out of the 238 samplings met this target which is equivalent to 15.1%. This is below the 18.5% success rate of the total spreadsheet.
If only fish caught in samples of over 30 fish are used in the study, then the total number of fish that should be analysed would be 1534, not the 5042 used in the paper. A surprising 3530 fish should be deemed to be invalid for this research.
Table 1 of the Ives paper highlights the number of wild trout that are either with or without lice infestation for different life stages of lice. Thus, 67.6% of fish have no juvenile lice, 63% have no mobile lice stages and 53.6% have no lice of any stage.
By comparison, the percentages for the reassessed data are 84.2% of fish are lice free for juveniles and 81.1% lice free for mobiles. In addition, 74.1% of all fish are free of lice of any life stage.
When the spread of lice throughout the sampled population is plotted, the graph is a classic aggregated distribution.
I have referred earlier in this commentary to the paper by Middlemas et. al. (2013). That paper discussed overdispersion and describes it as the presence of greater than expected variability in a data set. By comparison, the book ‘Evolutionary Ecology of Parasites’ by Robert Poulin states that whilst parasitologists often use the word overdispersion to describe the uneven distribution, the word is confusing to many biologists suggesting a wider dispersion rather than clumping. This is why the term aggregation is preferred. The Ives paper makes no mention of overdispersion and refers only to aggregation in terms of placing data together. This lack of reference to parasite distribution is concerning for a paper considering the impact of parasites on their hosts.
The authors of the Middlemas paper suggest using the Wells 13 lice threshold on wild fish of around 70g to determine any impact. Fish infested with 13 and more mobile lice are subject to physiological stress and potential death from sea lice infestation. The number of fish exposed to 13 mobile lice or more using this data is just 1.2% of the sampled fish. I would therefore suggest that it is not necessary to rely on the complex statistics used by Ives et. al. (2024) to determine whether there is a significant impact or not on wild fish from sea lice associated with salmon farming. Surely, a potential impact on 1.2% of the sampled fish sends out a clear message and thus is hardly worth any consideration. I ought not be surprised that the Scottish Government scientists are so reluctant to discuss such findings as these, but they are. After all, a potential impact of 1.2% does not fit in with their narrative.
Never mind the Brouhaha in Canada, there should be more than a Brouhaha taking place in Scotland.