Structural equation modeling (SEM) is a set of approaches that have seen exponential usage in the soil sciences as well as the related fields of agriculture and biogeochemistry. When correctly used and interpreted, SEM can be a powerful and flexible tool to test complex hypotheses on causality. However, the recent explosion of SEM usage in the soil sciences facilitated by user-friendly statistical programs has not been fully met by statistical expertise of users, reviewers and editors, ultimately leading to widespread contamination of the literature with inappropriate modeling and inflated or unfounded causal claims. The rise of such “SEM slop” poses a serious risk of an unreliable knowledge base and also undermines efforts and standards on what constitutes causality in the soil sciences. To address this, we diagnose major pitfalls in SEM, with an eye towards considerations specific to soil sciences, categorizable as three types: (1) Causal claims, including not satisfying causal criteria, lack of justified a priori models, not considering counterfactuals, and unqualified causal language; (2) Experimental design, including use in randomized complete block designs without complete pooling or multi-level models, inappropriate data type (e.g., ontological misalignment), and insufficient sample size; and, (3) Assessing the model, including incomplete or inappropriate model evaluation, non-qualified use of modification indices, and lack of robustness tests. There is a dual imperative for users as well as reviewers and editors to better implement and evaluate SEMs and claims of causality made with SEMs. To support this, we offer best practices and practical considerations on these three major pitfalls. These best practices will help SEM be appropriately employed as a powerful, nuanced statistical tool that benefits the soil science community.
扫码关注我们
求助内容:
应助结果提醒方式:
