Correction to “Causal Effects Versus Causal Mechanisms: Two Traditions With Different Requirements and Contributions Towards Causal Understanding”

Abstract

Grace, J. B., N. Huntington-Klein, E. W. Schweiger, M. Martinez, M. J. Osland, L. C. Feher, G. R. Guntenspergen, & K. M. Thorne. 2025. “ Causal Effects versus Causal Mechanisms: Two Traditions with Difference Requirements and Contributions towards Causal Understanding.” Ecology Letters 28:e70029. https://doi.org/10.1111/ele.70029

We would like to correct the statement found on page 12,

Ferraro, Sanchirico, and Smith (2019) have stated that ‘Mechanistic models that judge success by model-data consistency represent predictive inference. Such models are not considered to be causal and need not include any variables with causal effects.’

to say,

Ferraro, Sanchirico, and Smith (2019) have stated that mechanistic models that judge success by model-data consistency represent predictive inference and such models are not typically considered to be causal and need not include any variables with causal effects. Such models, while not considered to be causal, may shed light on causal relationships under limited circumstances.

We would also like to correct the statement found on page 2,

“For example, Dee et al. (2023) repeatedly makes the unconditional declaration that data pooled across separate samples, such as samples across an environmental gradient, cannot be used for causal inferences because the samples cannot be assumed to be ‘all else equal’. Other presentations of causal statistics to ecologists make similar declarations and expound on an extensive list of restrictions. Some have gone so far as to make the blanket declaration that the parameters and relationships in mechanistic models do not qualify as causal because they are not based on causal statistical methods (Ferraro, Sanchirico, and Smith 2019).”

to say,

“For example, Dee et al. (2023) imply that analyses of data pooled across spatially separated samples require defending the assumption of no omitted confounders (i.e., adhering to the Perfection Standard in our Figure 2). To be more specific, they state, ‘In Grace et al., the authors use spatial variation across sites … The Dee et al. model eliminates the spatial variation that comes from the “between-plots” comparisons because we believe those comparisons will yield biased inferences about the relationship between richness and productivity – hidden bias that comes from unobserved confounding variables.’ Elsewhere, Dee et al. say, ‘it is unlikely that one can measure all possible confounding variables.’ and ‘failure to control for all confounding variables can lead to inferences of the wrong sign or magnitude.’”

We apologize for the errors.