Illustrated by means of Simulated Experiments in R

On this article, I clarify tips on how to use backdoor criterion within the experimental setting to pick good management variables, or, keep away from choosing the dangerous ones, utilizing Directed Acyclic Graphs (DAG). I began my very own causal inference journey by means of potential outcomes mannequin, which was launched in my earlier article. I simply “found” DAG just lately from taking Prof. Jason Roos’ wonderful experimentation and causal inference class, and actually like DAG as a framework to simply theorize and visualize a causal mannequin. It facilitates the identification evaluation by making variables included within the mannequin and the assumptions made concerning the relationships between these variables salient. Because of this, it additionally helps with figuring out the confounding variables and analyzing tips on how to de-confound.

I assume that the readers already perceive the fundamentals of DAG (if not, Scott Cunningham’s Causal Inference Mixtape is a useful begin), and I consider that the quickest option to get a grasp of the backdoor criterion is by way of examples. Due to this fact, I’ll proceed as follows: first, I’ll lay out the query we need to reply and supply the DAG illustration for us to simply conceptualize it; Subsequent, I’ll clarify what backdoor criterion is and present the way it ought to be carried out in our particular instance; Final, I’ll run the instance by means of simulated experiments.

The info science drawback we need to clear up is the next: we need to know what interventions can successfully affect folks to behave extra sustainably. To take action, we design randomized experiments with remedies geared toward encouraging folks to scale back their electrical energy consumption. Let’s suppose that we now have completed the primary experiment by which we used financial reward because the remedy (e.g., present card). However we’re questioning whether or not there’s a cheaper option to obtain the same and even bigger impact through the use of behavioral interventions. Due to this fact, we design a second experiment by which we use info nudges because the remedy (e.g., handled individuals will obtain an e mail notification reminding them to scale back consumption with…