Who Got Through the Door?
Asystole

I adjusted for every variable.
Even shoe size.

Did you adjust for knowing why?

Confounding: a shared cause can bend the view,
Push the treatment choice and the outcome too.
Sicker people get the stronger treatment plan;
Severity also raises risk. That's where we began.
Compare the groups raw, then call the treatment bad?
You left the common cause off the page you had.
Draw the fork: severity points two ways;
Your coefficient needs more than a confident phrase.
Choose adjustment with a reason, grounded in the question asked;
You clicked 'select all columns' and declared the problem passed.
I ask what caused the variables you found;
You keep adding columns till the laptop makes a sound.
A mediator sits along the causal chain;
Adjusting there can change the effect you explain.
For a total effect, blocking that route can hide
Part of the very change you're trying to describe.

Who got through the door? Who got left outside?
What did you condition on? Show me what you tried.
Common cause, common effect: trace the arrows through;
An extra column can invent a story too.
Who got through the door? Who got left outside?
What did you condition on? Show me what you tried.
Common cause, common effect: trace the arrows through;
An extra column can invent a story too.

Collider: two arrows meet at one shared effect;
Condition on that meeting point, and watch the links connect.
Imagine flu and fractures independent in a town;
Either can get you admitted. Now narrow the sample down.
Among admissions, those causes can appear related;
A link inside the ward that your selection created.
The hospital door became a statistical audition;
Your sample got a storyline from its admission condition.
Selection can mean who joins or who stays;
Loss to follow-up can filter different ways.
Bias needs a mechanism; explain who disappears;
A giant sample won't repay those missing-data years.
Restricting, stratifying, regression can all do
The conditioning that opens a path you never knew.
You said, "Control for everything; that's how legends work."
You controlled the common effect and gave the data a smirk.

Who got through the door? Who got left outside?
What did you condition on? Show me what you tried.
Common cause, common effect: trace the arrows through;
An extra column can invent a story too.
Who got through the door? Who got left outside?
What did you condition on? Show me what you tried.
Common cause, common effect: trace the arrows through;
An extra column can invent a story too.

Confounder: a common cause in this simple scene.
Mediator: carries part of the effect between.
Collider: common effect; two arrows meet.
Name the path and question before you take a seat.
A variable's role depends on the diagram and aim;
A label on a column doesn't settle every claim.
I choose a defensible set, then state what I assume;
You brought every variable and ran out of room.

Who got through the door? Who got left outside?
What did you condition on? Show me what you tried.
Common cause, common effect: trace the arrows through;
An extra column can invent a story too.
Who got through the door? Who got left outside?
What did you condition on? Show me what you tried.
Common cause, common effect: trace the arrows through;
An extra column can invent a story too.

I filtered this crowd to fans who love my show.
"Selection."
Fair. Open the door.
Let the critics get a row.