Model clades are vital for comparative biology, and ascertainment bias is not a problem in practice: a response to Beaulieu and O'Meara (2018)

In several previous publications we have questioned the significance of largescale phylogenetic studies, highlighting instead the value of analyzing smaller, comprehensively sampled clades (Donoghue and Edwards, 2014; Edwards et al., 2015, 2017). In the recent issue of the American Journal of Botany on the Plant Tree of Life (2018, vol. 105, issue 3), Beaulieu and O’Meara (2018) explored an opposing concern, namely, that the study of selected small clades could lead to incorrect conclusions about evolutionary processes. Specifically, they performed a series of simulations designed to test whether studies of smaller clades present an “ascertainment bias”. The core idea is that comparative biologists tend to study clades that show variation in a trait of interest and, correspondingly, fail to focus sufficient attention on clades that are not so variable in that trait. Consequently, this form of taxonomic sampling could result in a biased inference about the rate of change in the trait of interest across the larger clade that includes both variable and nonvariable subclades. Surprisingly, their simulations showed that this was not generally the case. In fact, averaging across many smaller, variable subclades, there turned out not to be a significant deviation from the rates that were estimated for a more inclusive clade based on a random sample of a similar number of species. However, they did show that sampled subclades exhibited much higher variance around the estimated transition rates for the more inclusive clade. Beaulieu and O’Meara (2018) concluded that ascertainment bias is a major problem in comparative biology and argued that large, global phylogenetic analyses are needed to overcome this bias. In the same special issue, Folk et al. (2018) concurred with their assessment of the “risks” of relying on smaller studies. Many details of the Beaulieu and O’Meara study are worthy of careful examination, but here we reflect solely on the central conceptual issue that they raised for comparative biologists. Our main concern is that they implicitly misrepresent the motivation behind the study of smaller variable clades. As reflected in their simulations, their premise appears to be that we study small tractable subclades (e.g., Viburnum) to estimate transition probabilities for a more inclusive clade (e.g., Campanulidae). But this is typically not the goal at all. That is, most comparative biologists are not studying a particular subclade to generate such an estimate for a more inclusive clade. And, indeed, it would be a big risk to extend the inference for a subclade to more inclusive clades, not only because the sample of species is typically much smaller, but also because, as Beaulieu and O’Meara correctly pointed out, our estimate might then be highly biased by the particular subclade in question. If we were, in fact, attempting to estimate rates for a more inclusive clade, there would be an ascertainment bias problem. However, because those who study smaller clades typically are not intending to extend such estimates beyond the clade of interest, ascertainment bias, as described by Beaulieu and O’Meara, is simply not an issue in practice. One might then ask: Why bother studying a smaller variable clade if one is not interested in estimating more global transition rates (or speciation rates, or extinction rates, etc.)? Asked another way: What good are such targeted studies if not to render broader INVITED SPECIAL ARTICLE: COMMENTARY