On the basis of clarification with the authors, this question has been updated. My original concern related to spatial patterns in butterfly abundance being related to levels neonicotinoid application, whilst other spatial variables (such as area of intensive arable agriculture) were ignored. This misunderstanding was due to some ambiguity in the data description section “This dataset contains counts of butterflies from sites in a wide range of different habitats from across the UK. We included the full dataset for each species”. However, the authors have since clarified that their response variable was a single annual national collated population index, so spatial variation is effectively averaged out each year.

However, I still have major concerns that other environmental drivers may be leading to the observed associations with annual levels of neonicotinoid applications. The authors do not account for long term trends in the butterfly populations. So if any butterflies show long term population trends, beyond those driven by weather variables included in the model, (e.g. due to changes in host plant availability or quality), then these will be attributed to the neonicotinoid variable. This is apparent in Table 2, where the species that have declined the most over time apparently have the strongest neonicotinoid coefficient. I conducted a regression of species population trend versus the neonicotinoid coefficient and find a strong relationship whereby the species that show the steepest population declines over time also have the largest negative neonicotinoid coefficient (F = 83.14, df = 1 & 15, p < 0.001). Although they did not analyse by individual species, the authors do in fact notice this pattern. In the discussion, they state “All 14 species which the model identified as being most negatively affected by neonicotinoid usage exhibited 10-year declines during the first decade of the 21st century”. This is written to suggest that neonicotinoids might explain the long term declines, but I suggest the reverse causality is true and long term trends in the species lead to the (potentially spurious) neonicotinoid coefficient. To back this assertion, consider the comma and ringlet butterflies in Figure 1. They appear to show a very strong (and probably statistically significant) increase in population size in years when more neonicotinoids were applied. Does this mean those species actually benefit from neonicotinoid pesticides? Probably not. It is probably a consequence of the fact that these two species have long term increasing population trends for other reasons, so their population sizes are higher later in the time series when there also, by chance, happens to be more neonicotinoids applied in the UK. Conversely so for the other species. Rather than use annual level of neonicotinoid application as the temporal covariate of interest here, one could probably replace it with any term that shows a long term trend, and get similar significant results. E.g. The number of houses in the UK (possible headline “Development linked to decline in UK butterflies”) or the number of X-factor christmas number one singles (possible headline “Butterflies would rather die than listen to another Simon Cowell-produced song”). And so on… More convincing evidence would be that there was an effect of annual neonicotinoid applications independent of long term butterfly trends (e.g. by including Year as a continuous variable in the model). i.e. do years with more or less pesticide applications lead to inter-annual increases or decreases in the number of butterflies? I would be very interested to see this analysis. The authors may be concerned that this approach has a danger of ‘throwing the baby out with the bathwater’, but I believe that linking inter-annual changes in butterflies to pesticide applications, or adopting a more fine-grained spatial approach (and controlling for other confounding factors), is the only way to improve confidence that there is a causal relationship using such a correlative approach.