This paper examines MIKE and PIKE data to determine the primary causes underlying African elephant mortality patterns. Their approach expands upon past studies addressing this issue by adding new covariates, more poaching data, and new modelling methods.

This is an important paper. However, I have some concerns about the “one-shoe-fits all” approach they have taken, treating all elephant populations the same. That approach seems logical if asking what factors apply to all sites as a determinant of poaching. However, that approach does not address what predicts the largest amount (versus proportion/site) of total elephant killings across the continent as a whole. I am concerned that ignoring major differences between populations may be skewing their results.

The most dramatic population differences concerns forest versus savannah elephants, which are now recognized as separate species by IUCN, potentially warranting treating them separately. For example, large numbers of forest elephants may be very different from large numbers of savannah elephants. Combining both forest and savannah elephants into the same analysis may obscure the effect of large size because what is considered large depends on the species. Most forest elephant sites have very few elephants compared to most savannah elephant sites. More broadly, PIKE values can easily be skewed in sites with very few elephants, potentially causing disproportionate error if each site is considered equal.

Cover may be a poor discriminator of variation in poaching among forest elephants because cover is relatively high everywhere. By contrast, cover is far more variable in savannah elephant habitat and may thus be an important discriminator of poaching for savannah but not forest elephants.

The authors may argue that they are unable to divide their data by species because the composition of forest versus savannah elephants in many countries is still unknown, especially in West Africa. However, most protected areas have a predominance of one species, which is fairly well-known (especially to several of the well-informed co-authors on this paper). Alternatively, dividing their data by region (as opposed to forest/savannah) may provide at least a partial solution that is better than none.

Conducting their analyses site by site may also be a problem if major ivory trafficking syndicates operating in a specific areas are driving the greatest amount of the illegal trade overall. It is now well known that there are a small number of major poaching hotspots in Africa where the greatest number of elephants are being killed. Hotspots matter most in terms of total number of animals killed but are given equal weight to non-hotspots in their model. Yet, PIKE looks at proportion of illegal killing relative to all other mortalities at a given site, providing a relative number within each site. That raises the question: Is the aim of the paper to determine what factors common to all sites best predict the proportion of illegally killed elephants, or, what factors predict the greatest number of illegal killings continent-wide. Both questions are important. However, it is equally important to articulate what question is being asked. Covariates applied to places where major criminal syndicates are acquiring their ivory may respond very differently in those places compared to sites that do not have enough elephants to become a hotspot. The same may be true for newly emerging, or yet to be discovered, hotspots.

Consider, for example, Figure 4 in their paper. Household wealth (left side of Fig 4) appears to have a strong negative relation to PIKE continent-wide. However, the impact of household wealth on PIKE is far weaker in the two biggest poaching hotspots in Africa (Tanzania and the Tridom). NE Gabon is a major forest elephant poaching hotspot and has some of the highest income and highest PIKE. Non-hotspots like CAR, DRC, and Mali show the opposite trend with strong negative correlations between wealth and PIKE. Some countries may have very few elephants to kill, making “random” variables appear more extreme. For example, North-Central Namibia has good enforcement and low PIKE, but hardly any elephants to kill. How does that affect the overall model?

The Kavango-Okavango Transfrontier Conservation Area (KAZA) has low killing overall and good wealth, but recently emerged as a major poaching hotspot, driven by major trafficking syndicates, even though household wealth in those areas has not changed. (What changed was the president of Botswana. The former president had strong policies aimed at combatting IWT, including shoot to kill policies, and a total hunting ban. The new president was focused on culling as a means to control human wildlife conflict, suggesting at one point to sell the resultant surplus of elephant meat for dog food.) It is noteworthy that in lines 325-327 the coauthors state that their findings differ from Schlossberg et al (2020), who (unlike the current paper) did not find correlations between elephant mortality and human development or governance. The co-authors explain this difference by arguing that Schlossberg focused only on savannah elephant whereas the current paper focuses on savannah and forest elephants. Which approach is right? The Schlossberg paper, by the way, also showed a significant increase in poaching in southern Africa largely attributed to poaching in the KAZA.

In summary, this paper appears to be asking what factors best explain poaching across all sites (one shoe fits all) versus what factors explain the largest absolute number of poaching incidents overall (i.e., places where most killing occurs across the entire continent). It is unclear which approach may be best. Regardless, it is very important that the paper clearly articulates what their model may NOT explain (i.e., what factors are causing the largest number of elephants to be killed in Africa). I am concerned that failure to clearly articulate their study question may give higher weight to variables like household wealth than is warranted as a factor causing the greatest amount versus proportion of illegally killed elephants. That could, in turn, mislead decisions about how to mitigate the greatest amount of elephant poaching across the African continent as a whole.

This paper did a good job addressing my previous concerns. The analyses and interpretation of results are clear, well described, with careful consideration of alternative explanations. They also do a good job pointing out limitations of the data. I have just two points that I would like to see clarified/addressed.

First, at times, the paper equates ivory syndicates with poachers (e.g., lines 108-112, 365-368)). This needs to be justified. Syndicates tend to rely on poachers to provide the product they consolidate for export. However, these syndicates tend to operate far away from where the poaching occurs. Moreover, the strength of connections between poacher and syndicate is a loose one, with syndicates relying on middlemen to acquire ivory from a wide array of poachers. I recommend the authors make clear that they are asking what drives poachers, not syndicates.

Second, the authors note higher PIKE rates forest compared to savanna elephants and note a lack of explanatory variables to address this difference (lines 421-425). However, they do not discuss the possibility that this species difference is driven by price, despite price being an important explanatory variable in their model. Forest elephant ivory is more preferred than savannah elephant ivory, both by buyers and carvers, and consequently garners a higher price. This is because forest elephant ivory is far denser than savannah elephant ivory making it better suited for making intricate carvings. This possibility that price is driving this species difference should be discussed, especially since no other variable seemed to explain these species differences in PIKE rates.