In a recent contribution, Greenville et al. (2021) elegantly use network analyses to investigate the relationship between threat diversity and extinction risk in vertebrates. We applaud the use of this new method, but question the authors’ conclusions.

Contrary to results from previous studies (e.g., Ducatez & Shine, 2017; González-Suárez & Revilla, 2014; Jono & Pavoine, 2012), Greenville et al. (2021) conclude that “extinction risk is not higher for species exposed to a greater number of threats” (except for cartilaginous fishes). We suggest that methodological problems weaken this conclusion.

First, research effort is one of the main predictors of threat diversity; the more research that an imperiled species attracts, the greater the number of likely threats identified (Allek et al., 2018; Ducatez & Shine, 2017). Species that are endangered also tend to be investigated less often (Ducatez & Shine, 2017) and hence will (paradoxically) be allocated fewer threats. To take this bias into account, previous analyses have included proxies of research effort. Similar corrections could be implemented in the network analysis framework by correcting the number of species per threat category by proxies of research effort, or by excluding poorly known species.

Second, most previous studies have grouped IUCN threats into a small number of categories, thereby limiting biases in threat identification due to the data available (Hayward, 2009). By considering 39 different categories (vs. less than 12 in most previous studies), the analyses of Greenville et al. (2021) are exposed to biases in threat assessment.

Third, the results of Greenville et al. (2021) do not account for interspecific variation in geographic range size. Species with broader ranges tend to be exposed to more threats, while also being at lower risk (range size is a major criterion for assessing extinction risk). Thus, we need to remove the potential effect of range size by incorporating it as a covariate, and excluding species listed as at risk because of their small range (e.g., Ducatez & Shine, 2017; González-Suárez & Revilla, 2014).

In conclusion, network analyses can be valuable but at this stage, most studies show an increase in global extinction risk as the number of threats increases. We believe that more robust network analyses, correcting for biases in threat assessment and considering range size, would bring additional insights on this fundamental question.

在最近的贡献中，格林维尔等人。( 2021 ) 优雅地使用网络分析来调查脊椎动物的威胁多样性与灭绝风险之间的关系。我们赞赏使用这种新方法，但质疑作者的结论。

与之前的研究结果相反（例如，Ducatez & Shine，2017 年；González-Suárez & Revilla，2014 年；Jono & Pavoine，2012 年），Greenville 等人。( 2021 ) 得出的结论是“对于暴露于更多威胁的物种而言，灭绝风险并不更高”（软骨鱼类除外）。我们认为方法论问题削弱了这一结论。

首先，研究工作是威胁多样性的主要预测因素之一；濒危物种吸引的研究越多，确定的可能威胁的数量就越大（Allek 等，2018 年；Ducatez & Shine，2017 年）。濒危物种的调查频率也较低（Ducatez & Shine，2017 年），因此（自相矛盾地）被分配的威胁较少。为了考虑到这种偏见，之前的分析包括了研究工作的代理。类似的修正可以在网络分析框架中实施，方法是通过研究工作的代理修正每个威胁类别的物种数量，或者通过排除鲜为人知的物种。

其次，大多数先前的研究将 IUCN 威胁分为少数类别，从而限制了由于可用数据而导致的威胁识别偏差（Hayward，2009 年）。通过考虑 39 个不同的类别（相比之前的大多数研究中少于 12 个），Greenville 等人的分析。（2021 年）在威胁评估中面临偏见。

第三，格林维尔等人的结果。( 2021 ) 没有考虑地理范围大小的种间差异。范围更广的物种往往面临更多威胁，同时风险也较低（范围大小是评估灭绝风险的主要标准）。因此，我们需要通过将其作为协变量合并来消除范围大小的潜在影响，并排除由于范围小而被列为处于危险之中的物种（例如，Ducatez & Shine，2017 年；González-Suárez & Revilla，2014 年）。

总之，网络分析可能很有价值，但在此阶段，大多数研究表明，随着威胁数量的增加，全球灭绝风险也会增加。我们相信，更强大的网络分析，纠正威胁评估中的偏差并考虑范围大小，将为这个基本问题带来更多见解。