Walmsley Sam F, Morrissey Michael B
Scottish Oceans Institute, School of Biology, University of St. Andrews East Sands St. Andrews United Kingdom.
Dyers Brae House, School of Biology, University of St. Andrews Greenside Pl St. Andrews United Kingdom.
Evol Lett. 2021 Nov 9;6(3):234-244. doi: 10.1002/evl3.258. eCollection 2022 Jun.
Many biological traits covary with body size, resulting in an allometric relationship. Identifying the evolutionary drivers of these traits is complicated by possible relationships between a candidate selective agent and body size itself, motivating the widespread use of multiple regression analysis. However, the possibility that multiple regression may generate misleading estimates when predictor variables are correlated has recently received much attention. Here, we argue that a primary source of such bias is the failure to account for the complex causal structures underlying brains, bodies, and agents. When brains and bodies are expected to evolve in a correlated manner over and above the effects of specific agents of selection, neither simple nor multiple regression will identify the true causal effect of an agent on brain size. This problem results from the inclusion of a predictor variable in a regression analysis that is (in part) a consequence of the response variable. We demonstrate these biases with examples and derive estimators to identify causal relationships when traits evolve as a function of an existing allometry. Model mis-specification relative to plausible causal structures, not collinearity, requires further consideration as an important source of bias in comparative analyses.
许多生物学特征与体型共同变化，从而形成一种异速生长关系。由于候选选择因子与体型本身之间可能存在关联，确定这些特征的进化驱动因素变得复杂，这促使多元回归分析得到广泛应用。然而，当预测变量相关时，多元回归可能产生误导性估计的可能性最近受到了广泛关注。在这里，我们认为这种偏差的一个主要来源是未能考虑大脑、身体和选择因子背后复杂的因果结构。当大脑和身体预期会在特定选择因子的影响之外以相关方式进化时，简单回归和多元回归都无法确定选择因子对脑容量的真正因果效应。这个问题源于在回归分析中纳入了一个（部分）是响应变量结果的预测变量。我们通过实例展示了这些偏差，并推导了估计量，以在特征作为现有异速生长函数进化时识别因果关系。相对于合理因果结构的模型误设，而非共线性，需要作为比较分析中偏差的一个重要来源加以进一步考虑。
