Boris Sobolev
@sobor.bsky.social
Causal questions, counterfactual answers
Pinned
Boris Sobolev
@sobor.bsky.social
· Jan 18
Probability of benefit in clinical practice
YouTube video by SobolevSpaces
youtu.be
ATE? nah…
We ask how many patients would get better with treatment but not without it.
youtu.be/ms4z2i7JzVY
We ask how many patients would get better with treatment but not without it.
youtu.be/ms4z2i7JzVY
@pwgtennant.bsky.social Peter, your repost should bring us who study SES in health to re-evaluation of how we do research. We should focus on causal questions, causal methodology, DAGs, probability of benefit, recovery from selection bias, transportability of effects, causal fairness.
January 25, 2025 at 9:53 PM
@pwgtennant.bsky.social Peter, your repost should bring us who study SES in health to re-evaluation of how we do research. We should focus on causal questions, causal methodology, DAGs, probability of benefit, recovery from selection bias, transportability of effects, causal fairness.
ATE? nah…
We ask how many patients would get better with treatment but not without it.
youtu.be/ms4z2i7JzVY
We ask how many patients would get better with treatment but not without it.
youtu.be/ms4z2i7JzVY
Probability of benefit in clinical practice
YouTube video by SobolevSpaces
youtu.be
January 18, 2025 at 3:07 AM
ATE? nah…
We ask how many patients would get better with treatment but not without it.
youtu.be/ms4z2i7JzVY
We ask how many patients would get better with treatment but not without it.
youtu.be/ms4z2i7JzVY
sometimes, a flower is just a flower
December 24, 2024 at 4:33 PM
sometimes, a flower is just a flower
Un laberinto de tentaciones
December 23, 2024 at 3:34 AM
Un laberinto de tentaciones
If you've never tasted the dishes prepared by la chefa Anayeli Suazo of Tentaciones, you don't know what a gourmet meal is and what it looks like.
December 21, 2024 at 3:49 AM
If you've never tasted the dishes prepared by la chefa Anayeli Suazo of Tentaciones, you don't know what a gourmet meal is and what it looks like.
Nothing can beat sleeping until noon on vacation
December 20, 2024 at 11:34 PM
Nothing can beat sleeping until noon on vacation
I had a quick chat with Elias Bareinboim on the way from the a/p #neurISP24…
He reminds me of Mozart, who barely has time to write down all the revelations of causality pouring on him from above. 😊
He reminds me of Mozart, who barely has time to write down all the revelations of causality pouring on him from above. 😊
December 14, 2024 at 3:13 PM
I had a quick chat with Elias Bareinboim on the way from the a/p #neurISP24…
He reminds me of Mozart, who barely has time to write down all the revelations of causality pouring on him from above. 😊
He reminds me of Mozart, who barely has time to write down all the revelations of causality pouring on him from above. 😊
Wondering about #causal mediation analysis in comparative effectiveness research? Check out our framework for the mediating effect of treatment time of 2 treatment methods for complex coronary artery disease.
doi.org/10.9778/cmaj...
doi.org/10.9778/cmaj...
November 23, 2024 at 7:00 PM
Wondering about #causal mediation analysis in comparative effectiveness research? Check out our framework for the mediating effect of treatment time of 2 treatment methods for complex coronary artery disease.
doi.org/10.9778/cmaj...
doi.org/10.9778/cmaj...
There’s this misconception that 'causality is in the design.' Apart from being uninformed, this assertion is fed by a bigger fallacy: that causality is a product of our mind rather than a feature of the material world.
Pearl’s major contribution is grounding causality in materialism.
Pearl’s major contribution is grounding causality in materialism.
November 20, 2024 at 8:39 PM
There’s this misconception that 'causality is in the design.' Apart from being uninformed, this assertion is fed by a bigger fallacy: that causality is a product of our mind rather than a feature of the material world.
Pearl’s major contribution is grounding causality in materialism.
Pearl’s major contribution is grounding causality in materialism.
Oh, I agree that causal claims come from reasoning, and that design is a part of it.
I just don't see how reasoning using words can be superior to reasoning using mathematical notation.
Pearl’s formulas easily describe RCT. But trialsts have a hard time to describe the estimand of their experiment.
I just don't see how reasoning using words can be superior to reasoning using mathematical notation.
Pearl’s formulas easily describe RCT. But trialsts have a hard time to describe the estimand of their experiment.
Again, notation is not important for causality is the design. For example, blind evaluation is important for causality. Whats the notation for that?
November 20, 2024 at 5:31 PM
Oh, I agree that causal claims come from reasoning, and that design is a part of it.
I just don't see how reasoning using words can be superior to reasoning using mathematical notation.
Pearl’s formulas easily describe RCT. But trialsts have a hard time to describe the estimand of their experiment.
I just don't see how reasoning using words can be superior to reasoning using mathematical notation.
Pearl’s formulas easily describe RCT. But trialsts have a hard time to describe the estimand of their experiment.
So, you admit that your notation doesn’t capture the difference btw association and causation?
And! and your reasoning ‘it’s in the design’ is not reflected in your notation, right?
And! and your reasoning ‘it’s in the design’ is not reflected in your notation, right?
Precisely. Causality is in the design of the study, not in the notation.
November 20, 2024 at 5:15 PM
So, you admit that your notation doesn’t capture the difference btw association and causation?
And! and your reasoning ‘it’s in the design’ is not reflected in your notation, right?
And! and your reasoning ‘it’s in the design’ is not reflected in your notation, right?
I’m glad that you are using a rigorous notation for statistical hypothesis testing. Because it offers me an opportunity to ask, where is causality here?
H0: E(Outcome|Treatment) = E(Outcome|Placebo) is from 100 years ago.
November 20, 2024 at 5:01 PM
I’m glad that you are using a rigorous notation for statistical hypothesis testing. Because it offers me an opportunity to ask, where is causality here?
Notation gives us a means to express our ideas in rigorous and transparent fashion. Whether it is for radius and
circumference, or for a value that
outcome variable would assume in unit u, if it were assigned to a treatment group regardless of any factors that might influence treatment assignment.
circumference, or for a value that
outcome variable would assume in unit u, if it were assigned to a treatment group regardless of any factors that might influence treatment assignment.
What is absurd is to think that the notation used for the parameter has any importance for causal evidence. 🤣
November 20, 2024 at 4:30 PM
Notation gives us a means to express our ideas in rigorous and transparent fashion. Whether it is for radius and
circumference, or for a value that
outcome variable would assume in unit u, if it were assigned to a treatment group regardless of any factors that might influence treatment assignment.
circumference, or for a value that
outcome variable would assume in unit u, if it were assigned to a treatment group regardless of any factors that might influence treatment assignment.
You mean, this Fisher? 👇🏻
A denier of causal relationship btw tobacco smoking and lung cancer? yet ‘seeing’ causality btw race and the quality of ‘human stock’? Btw ‘high intellectual calibre’ of parents and their offsprings? Inverse link btw ‘fertility’ and ‘achievement’ of most valuable classes?
A denier of causal relationship btw tobacco smoking and lung cancer? yet ‘seeing’ causality btw race and the quality of ‘human stock’? Btw ‘high intellectual calibre’ of parents and their offsprings? Inverse link btw ‘fertility’ and ‘achievement’ of most valuable classes?
November 20, 2024 at 4:17 PM
You mean, this Fisher? 👇🏻
A denier of causal relationship btw tobacco smoking and lung cancer? yet ‘seeing’ causality btw race and the quality of ‘human stock’? Btw ‘high intellectual calibre’ of parents and their offsprings? Inverse link btw ‘fertility’ and ‘achievement’ of most valuable classes?
A denier of causal relationship btw tobacco smoking and lung cancer? yet ‘seeing’ causality btw race and the quality of ‘human stock’? Btw ‘high intellectual calibre’ of parents and their offsprings? Inverse link btw ‘fertility’ and ‘achievement’ of most valuable classes?
The most amazing part of the EBM proselytism is that it cannot specify the estimand of an RCT. The Pyramid evangelists keep repeating with religious fervor that the causal effect is in the design, but they refuse to reveal what the causal effect is.
Credo quia absurdum
Credo quia absurdum
November 19, 2024 at 11:12 PM
The most amazing part of the EBM proselytism is that it cannot specify the estimand of an RCT. The Pyramid evangelists keep repeating with religious fervor that the causal effect is in the design, but they refuse to reveal what the causal effect is.
Credo quia absurdum
Credo quia absurdum
There’s this misconception that causal evidence comes from data.
It comes from reasoning.
Unless, of course, the High Priests of EBM Pyramid deny us mere mortals the ability to think.
It comes from reasoning.
Unless, of course, the High Priests of EBM Pyramid deny us mere mortals the ability to think.
November 19, 2024 at 8:02 PM
There’s this misconception that causal evidence comes from data.
It comes from reasoning.
Unless, of course, the High Priests of EBM Pyramid deny us mere mortals the ability to think.
It comes from reasoning.
Unless, of course, the High Priests of EBM Pyramid deny us mere mortals the ability to think.
I’ve always been puzzled by my aversion to interpreting RCT outputs as evidence. Finally, I figured it out.
An RCT’s output, vaunted as assumption-free, are in fact not falsifiable. There is nothing to falsify.
That what makes EBM Priests so pompous: they present infallible evidence.😂
An RCT’s output, vaunted as assumption-free, are in fact not falsifiable. There is nothing to falsify.
That what makes EBM Priests so pompous: they present infallible evidence.😂
November 19, 2024 at 7:40 PM
I’ve always been puzzled by my aversion to interpreting RCT outputs as evidence. Finally, I figured it out.
An RCT’s output, vaunted as assumption-free, are in fact not falsifiable. There is nothing to falsify.
That what makes EBM Priests so pompous: they present infallible evidence.😂
An RCT’s output, vaunted as assumption-free, are in fact not falsifiable. There is nothing to falsify.
That what makes EBM Priests so pompous: they present infallible evidence.😂
I agree.
The PO notation is not needed to understand the counterfactual nature of the question that a randomized experiment purports to address.
The notation of Pearl’s First Principle would suffice.
The PO notation is not needed to understand the counterfactual nature of the question that a randomized experiment purports to address.
The notation of Pearl’s First Principle would suffice.
The evangelists of observational studies fail to realize that for the ATE in randomized trials the potential outcome notation is completely unnecessary for the evidence. In observational data it is directly nonsense almost always.
November 19, 2024 at 6:04 PM
I agree.
The PO notation is not needed to understand the counterfactual nature of the question that a randomized experiment purports to address.
The notation of Pearl’s First Principle would suffice.
The PO notation is not needed to understand the counterfactual nature of the question that a randomized experiment purports to address.
The notation of Pearl’s First Principle would suffice.
When I hear 'I don't like Pearl's causal hierarchy', it reminds me of an old story.
A tourist at the Louvre looks at La Gioconda and declares: 'I don't like it!' The curator nearby responds: 'Oh, she's seen so many people over the centuries that nowadays she decides who gets to like her.'
A tourist at the Louvre looks at La Gioconda and declares: 'I don't like it!' The curator nearby responds: 'Oh, she's seen so many people over the centuries that nowadays she decides who gets to like her.'
November 19, 2024 at 5:59 PM
When I hear 'I don't like Pearl's causal hierarchy', it reminds me of an old story.
A tourist at the Louvre looks at La Gioconda and declares: 'I don't like it!' The curator nearby responds: 'Oh, she's seen so many people over the centuries that nowadays she decides who gets to like her.'
A tourist at the Louvre looks at La Gioconda and declares: 'I don't like it!' The curator nearby responds: 'Oh, she's seen so many people over the centuries that nowadays she decides who gets to like her.'
The evangelists of EBM gospel fail to realize that the ATE in randomized trials not only involves counterfactuals, but actually answers a counterfactual question.👇🏻
November 19, 2024 at 4:51 PM
The evangelists of EBM gospel fail to realize that the ATE in randomized trials not only involves counterfactuals, but actually answers a counterfactual question.👇🏻
There’s this misconception that probability of diverging outcomes of alternative treatments is just a theoretical consideration
Here’s the probability of benefit (survival if treated early, mortality if not) estimated empirically.
Each bar shows the range of its possible values.
Here’s the probability of benefit (survival if treated early, mortality if not) estimated empirically.
Each bar shows the range of its possible values.
November 18, 2024 at 8:44 PM
There’s this misconception that probability of diverging outcomes of alternative treatments is just a theoretical consideration
Here’s the probability of benefit (survival if treated early, mortality if not) estimated empirically.
Each bar shows the range of its possible values.
Here’s the probability of benefit (survival if treated early, mortality if not) estimated empirically.
Each bar shows the range of its possible values.
The types of questions we ask determine the types of answers we get. 👇🏻
November 18, 2024 at 8:27 PM
The types of questions we ask determine the types of answers we get. 👇🏻
Amazing! Pearl’s probability of causation coincides with the causal effect P(Y_x) - P(Y_x′) under some conditions.
ie, the joint probability of opposing counterfactuals
P(Y_1=1 & Y_0=0)
equals the difference of marginal probability P(Y_1=1) and, wait for it, P(Y_0=1)
ie, the joint probability of opposing counterfactuals
P(Y_1=1 & Y_0=0)
equals the difference of marginal probability P(Y_1=1) and, wait for it, P(Y_0=1)
November 18, 2024 at 8:20 PM
Amazing! Pearl’s probability of causation coincides with the causal effect P(Y_x) - P(Y_x′) under some conditions.
ie, the joint probability of opposing counterfactuals
P(Y_1=1 & Y_0=0)
equals the difference of marginal probability P(Y_1=1) and, wait for it, P(Y_0=1)
ie, the joint probability of opposing counterfactuals
P(Y_1=1 & Y_0=0)
equals the difference of marginal probability P(Y_1=1) and, wait for it, P(Y_0=1)