|A couple of honeybees, the topic of this post. Image via Pixabay.
By now, many people have seen the two new studies claiming harm from neonicotinoids to honeybees, one was conducted in Europe and one in Canada. There have also been several articles criticizing the work for overstepping the data and drawing unsupported conclusions. I have a few issues with these studies beyond what has already been stated, so I'll shares some of the issues I have here and I won't touch on the other criticisms that others have covered.
Measuring everything without correction
In the European study, the researchers collected quite a few data points and took many different measurements. In total, there were 258 data points taken. Despite all of this, the researchers did not adjust for multiple measurements. From the supplemental materials and methods: "We did not apply Bonferroni corrections as the lack of independence for the majority of the response variables (e.g. different life stages of honey bee) meant that there was no valid level for the correction." This is a serious issue that I'll explain below.
First it's important to understand why correction for taking multiple measurements is vital. The main reason why researchers do this is because when many things are measured, the likelihood of false positives being found increases as the number of measurements increases. To counteract this, researchers will adjust the P-value to limit the risk of false positives (called a Type I error in statistics). The Bonferroni correction is a common method for doing this. It adjusts the resulting P-value so that it is smaller when more measurements are taken. Some researchers do not like it because it tends to be more conservative; however, not using it, or other methods for correction of multiple tests, introduces false positives in research.
In the supplemental materials and methods, the researchers claim that they were not taking independent measurements. However, this is not entirely accurate. The best way to describe these measures is semi-independent. As an example, the number of brood and their health directly impacts the future number worker bees and the survival numbers the next winter. However, the number of workers the following winter does not impact the number of brood the previous spring. Furthermore, their reason for not using Bonferroni Correction does not make sense as the measurements are semi-independent. The Bonferroni Correction is only one method to reduce the rate of false discovery. This method may or may not be correct in this case, but there are other methods for correcting the false discovery rate of dependent data. It is very risky to not perform any correction of large data sets like this when there are so many measurements being taken. It really doesn't look like a statistician was consulted for this paper and the quality suffers as a result.
One only has to look at the data in order to see that not correcting for multiple corrections could be what has led to the confusion surrounding the paper. The figure below shows each measurement in each of the three locations for both of the neonicotinoids tested. The results are confusing and there is no consistent result by treatment or country; the positive and negative results are mixed in with no particular link to a given treatment or country. This looks like the type of result you would see if there were false positives in the study from not correcting for multiple comparisons. Because of the lack of correction for multiple comparisons, which is commonly done in cases even when the measurements are semi-independent, we can't draw conclusions from the statistical tests that were run for this data set.
|The results of the European honeybee/neonicotinoid study compiled by Dr. Peter Campbell Sr., environmental specialist and head of product safety research collaboration, Syngenta, and first published by Jon Entine and Henry Miller. Shared here from Thoughtscapism. The light green cells are neutral results from neonicotinoids on honeybee health, the dark green cells are positive impacts of neonicotinoids on honeybee health, red cells are negative impacts from neonicotinoids on honeybee health, and white cells represent measurements that were not taken due to lack of bees. CLO = clothianidin and TMX = thiamethoxam.
In order to talk about other issues that exist with the conclusions of the European study, we need to forget everything that I just said about multiple measurment corrections and assume that the results are correct. This is because even if the results are accurate, there are still issues with the interpretation of those results that make the conclusion that neonicotinoids harm honeybees problematic.
Lack of a consistent effect
Looking at the raw data presented above, one thing is clear. There is no clear impact of neonicotinoids on any of the measurements taken. For example, the number of larval cells at flowering for thiamethoxam (TMX) showed an improvement in response to seed treatment in Germany, a negative response was seen in Hungary, and a neutral response was seen in the UK. For clothianidin (CLO), it was neutral in all three locations. The other critiques focused on the number of positive and negative effects among all the measurements, but they didn't address a key issue here. If a particular neonicotinoid was having an impact on honeybee health, it should be consistent across locations (this is why field researchers conduct experiments in several locations). If neonicotinoids in general were having a negative impact, then the effect should have been consistent across locations and treatments. Lacking this, the conclusion that neonicotinoids are negatively impacting honeybee health is not supported. This conclusion was toned down in the discussion for the article, but the press release had this as a firm conclusion and led to quite a bit of confusion in the reporting. It's another example of the disturbing trend of science by press release where the conclusions are touted without showing the needed supporting data. This often leads to bad science reporting as press releases will often overstate results. I've previously spoken about this issue here.
What about the varroa mite and the viruses?
The authors briefly touch on the differences in varroa mite infestations in the different countries. On average, Germany had the lowest rate of varroa mite infestation at 1.04% (+/- SE 1.00) followed by Hungary at 2.12% (+/- SE 1.34), and the UK having the highest at 8.05% (+/- SE 1.34). The authors then mention that the UK had a different hive treatment than Hungary and Germany, adding in another layer of variation that isn't accounted for. However, no mention is made of testing for any of the viruses that the varroa mite transmits. Simply put, this is a huge oversight on the part of the authors for this reason: both the varroa mite and many of the viruses they transmit weaken the immune system of bees. In fact, the varroa mite and at least one of these viruses, Deformed wing iflavirus, can act synergistically to reduce honeybee health to the point that the colony collapses. This is an excellent review on the impacts of the varroa mite, honeybee virus infection, and nutrition and how all three impact honeybee health. This review also discusses bee viruses and provides details on some of the more common ones. With the clear data linking honeybee health decline and honeybee viruses, it is not appropriate to measure honeybee health without addressing these viruses. This is because the viruses could impact the honeybee immune system to the point that the bee is more susceptible to other factors, such as neonicotinoids, and make it seem like those factors have an impact on honeybee health. This can lead to incorrect conclusions, such as neonicotinoids negatively impact honeybee health.
The Canada study: just a single season
I'll now address some of the concerns I had with the Canadian honeybee/neonicotinoid study. The biggest problem I have is that they based the laboratory exposure levels on a single growing season rather than monitoring the neonicotinoid levels over multiple years. Field work needs to be replicated for at least two years but often three or more years are required. One of the reasons for this is because pesticide residues can vary from year to year. Let's stop and think about that for a second. If pesticide residues can vary from year to year, does it make sense to use just a single growing season to determine what a field relevant dose of neonicotinoids is? All they can state is that the residue levels seen in the fields tested that year had an impact when given to bees. They cannot use the results from a single year to make blanket conclusions for all areas and years. For all we know the seed treatment could have been excessive that year (accidents do happen) and the converse is equally as true. This is precisely why replication of years is crucial for field work. Multiple years of measurements are needed to draw conclusions of any value.
What about the viruses and varroa mite?
In the supplemental materials and methods, the authors of the Canadian study state this: "We actively managed the colonies during the season, including adding empty honey ‘supers’ (i.e., a shallow 5-11/16” D x 19-7/8” L x 16-1/4” W chamber) and removal of swarm cells, but we did not chemically treat the colonies to control hive pests or diseases." In addition to not controlling hive pests or diseases, they also did not measure them. There is a reason why this matters. Previous work has clearly demonstrated that infection with bee viruses negatively impacts honeybee foraging behavior. Nosema also negatively impacts honeybee behavior. Not accounting for these diseases, let alone controlling for them by treating them, is a huge misstep as it introduces potential variation to the study that cannot be accounted for. Because of this, we do not know if diseases altered the foraging behavior of the studied honeybees and caused them to gather more pollen with neonicotinoids where it was more plentiful (in areas that were "treated"). Previous work has demonstrated that combining diseased honeybees with pesticides further reduces honeybee health, so ignoring this important finding in previous literature does not make sense.
The bottom line It's difficult (even impossible) to account for all factors in a field study. However, if there is a known factor that has been shown to reduce the very thing you are measuring, then it must be accounted for in the experimental design. Neither paper is exceptionally well designed for an agricultural study. Each study has issues that make it hard to draw conclusions from the data due to confounding factors that were not accounted for. If these papers had been sent to either an entomology or agronomy journal, the issues in each would have held up both papers. Both studies do not adequately address previous research on the topic in the experimental design, introduction or discussion. The format could be partially responsible for this, but these issues should have been addressed in the design of the experiment. Publication of weak studies like this only serve to confuse people, especially when coupled with poor science reporting that relies heavily on information in press releases.