class: center middle main-title section-title-3 # Causation and RCTs --- layout: true class: title title-3 --- # Randomized controlled trials .box-inv-3.medium[RCTs are the stereotypical experimental trial] -- .box-3[Treatment group gets treatment] -- .box-3[Control group does not] -- .box-3.medium[Average outcome in treatment group −<br>average outcome in control group =<br>**causal effect of program**] --- # The magic of randomization .box-inv-3.medium[Randomization makes it so that people don't self-select into the treatment] -- .box-inv-3.medium[Randomization also makes it so that differences in pre-treatment characteristics don't really matter] --- layout: false class: bg-full background-image: url("img/02/gold-standard.png") --- layout: true class: title title-3 --- # The "gold" standard .box-inv-3.medium[Some believe RCTs<br>are the *only* way to find causation] -- .box-3.small[You'll see this in the comment sections for articles about COVID-19!] --- # The "gold" standard .box-inv-3.medium[Saying RCTs are the "gold standard" implies that all causal inferences will be valid if you do the experiment right] -- .box-3[That's not the case! We don't care if studies are experimental!] -- .box-3[We care if our causal inferences are valid] --- # When to not randomly assign .box-inv-3.medium.sp-after[You can't RCT everything!] -- .box-3[Don't use RCTs…] -- .box-3[…when you need immediate results] -- .box-3[…when it's unethical or illegal] --- # RCTs and DAGs .pull-left-narrow[ .box-inv-3[RCTs intervene<br>in a DAG] .box-inv-3[Delete all arrows pointing into the treatment node] .box-inv-3[No more confounders and backdoors!] ] .pull-right-wide[ <img src="02_rcts_files/figure-html/rct-dag-big-1.png" width="100%" style="display: block; margin: auto;" /> ] --- # How to analyze RCTs .box-inv-3.medium.sp-after[1\. Check the balance between<br>the treatment and control groups] -- .box-inv-3.medium[2\. Estimate the difference in<br>outcome between the groups] --- layout: false class: title title-3 section-title-inv-3 # Our turn &nbsp; .box-3.large[Let's analyze an RCT with R!] --- class: title title-3 # Next up &nbsp; .box-inv-3.medium[Closing backdoors with regression]