Reducing bias in experimental ecology through directed acyclic graphs

Abstract

Experiments are a fundamental tool ecologists use to quantify causal relationships, in particular, by using randomized control trials (RCTs) that often regarded as the "gold standard" for causal inference (e.g., Hariton & Locascio, 2018). Under RCTs, researchers randomly assign units or individuals into treatment and control groups to eliminate potential confounding between treatment assignment and outcome, thereby increasing the internal validity of experiments (Holland, 1986;Rubin, 1974). Many of our foundational insights in

Parental early life environments drive transgenerational plasticity of offspring metabolism in a freshwater fish (Danio rerio)

“…F 1 embryos were brought to the Dalhousie University Aquatron and each family was divided into two thermal treatment groups ( figure 1 ): a Constant 27°C, which reflects a thermal optimum for reproduction and growth in zebrafish [ 34 , 35 ], and a Fluctuating temperature that varied sinusoidally on a diel basis (22–32°C; figure 1 ). This regime was intended to reflect natural thermal variability [ 34 ], while exposing fish to transient stressful temperatures, minimizing constant temperature-induced pathologies [ 21 , 34 , 38 – 40 ].

Figure 1 Split-clutch, factorial design of a long-term transgenerational plasticity experiment.

…”

“…To test whether changes in offspring RMR were mediated via changes to egg size, we also fitted Model 2, correcting for ‘Egg Size’ (diameter). Using two separate models allowed us to avoid overcontrol bias [ 40 ] in our estimation of ‘Later’ Parental Temperature in Model 1 as this treatment impacts egg size [ 41 ]. We interpreted non-negligible effects of Early or Later Parental Temperatures on offspring RMR as evidence of TGP, and interactions between Parental Temperatures and Test Temperatures as differences in the thermal sensitivity of TGP between Test Temperatures.…”