Conceptual framework for investigating causal effects from observational data in livestock1

Abstract

Understanding causal mechanisms among variables is critical to efficient management of complex biological systems such as animal agriculture production. The increasing availability of data from commercial livestock operations offers unique opportunities for attaining causal insight, despite the inherently observational nature of these data. Causal claims based on observational data are substantiated by recent theoretical and methodological developments in the rapidly evolving field of causal inference. Thus, the objectives of this review are as follows: 1) to introduce a unifying conceptual framework for investigating causal effects from observational data in livestock, 2) to illustrate its implementation in the context of the animal sciences, and 3) to discuss opportunities and challenges associated with this framework. Foundational to the proposed conceptual framework are graphical objects known as directed acyclic graphs (DAGs). As mathematical constructs and practical tools, DAGs encode putative structural mechanisms underlying causal models together with their probabilistic implications. The process of DAG elicitation and causal identification is central to any causal claims based on observational data. We further discuss necessary causal assumptions and associated limitations to causal inference. Last, we provide practical recommendations to facilitate implementation of causal inference from observational data in the context of the animal sciences.

Toceranib phosphate in the treatment of canine thyroid carcinoma: 42 cases (2009‐2018)

“…Potential prognostic explanatory variables (eg, sex, age, tumour stage, tumour location, presence of clinical signs) and all 2‐way interactions for inclusion were considered. To prevent overcontrol bias, other variables (eg, type of prior therapy, concurrent therapy, supportive medication) were not considered for model selection as the realization of these variables happened at the time of subgroup assignment 57 …”

Translating insights from the seed metabolome into improved prediction for lipid-composition traits in oat (Avena sativaL.)

“…, population structure). We sought to partition latent factors into those consistent with the former possibility (a biological process) and those consistent with the latter (a confounding effect) ( Bello et al 2018 ). Since we showed that most population structuring of metabolites was caused by drift, we expect their coordination to be largely random, and therefore unrelated to their functional class.…”

“…Although it may seem reasonable to suggest that the observed covariance among metabolites is due to a biological cause that is manifested in the metabolome, making causal inferences from observational data is nontrivial due to the presence of confounding factors ( Spirtes et al 2000 ; Rosa and Valente, 2013 ; Bello et al 2018 ). Given these data were collected on a structured population, it is expected that some of this covariance can be attributed to population structure, which can influence the construction of latent variables if not taken into account ( Phillips et al 2001 ).…”

Examining the relationships between phenotypic plasticity and local environments with genomic structural equation models

“…Secondly, in the current study, BN were constructed using a mixture of observational and experimental data. In the absence of randomization, dependencies observed in Bayesian networks constructed using observational data may be driven by unobserved confounders, thereby making causal claims based on the data problematic (Bello et al, 2018, see for review). Nevertheless, causal relationships can be learned from the data and should be used to generate hypothesis for further studies.…”

“…SEM utilize a system of linear equations to model the interrelationships between multiple dependant variables. Once introduced into the quantitative genetics frameworks pioneered by Henderson (1984), these approaches provide a means to partition multiple phenotypes into direct and indirect genetic components according to a predefined network structure (Gianola and Sorensen, 2004;Valente et al, 2013; Bello et al, 2018). In matrix form, the structural equation model is given by where all matrices are defined according to the MTM described above.…”