Hacker News new | past | comments | ask | show | jobs | submit login

If your standard for "conclusive" is "study must fork the universe to compare the same business with different timeline" then you may be waiting a while.

Regardless, in this case, where the ongoing harm could be very large, we have to at some point go with the best evidence available.




> If your standard for "conclusive" is "study must fork the universe to compare the same business with different timeline" then you may be waiting a while.

My standard is authors (and random internet commenters) to have an inkling of the progress in causal analysis and related progress in the last two decades.

Edit: Some resources:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9300826/

https://journals.sagepub.com/doi/pdf/10.1177/1536867X0800700...

https://www.hsph.harvard.edu/miguel-hernan/research/causal-i...

From 1999:

https://liberalarts.tamu.edu/sociology/wp-content/uploads/si...


I'm quite familiar with causal analysis [1]. What are your qualifications in the field?

The paper performs half a dozen standard robustness checks [2]. Which exactly do you believe are missing? Be specific, and include reasons why we should reasonably expect the opposite outcome.

[1] Senior ML Researcher @ Google, top 3 PhD, etc.

[2] Placebo analysis, instrumental variables, MTE analysis, breaking down the effect into plausible mechanisms such as reduction in front-line staffing, limiting results to nursing homes owned by the same chain, etc

edit: Instead of linking random surveys on causal inference, please make a specific claim, such as "the authors should have used propensity scoring", or "causal inference studies should all be ignored." Otherwise, there's no way to have a debate.


> I'm quite familiar with causal analysis [1]. What are your qualifications in the field?

That is nonsense.

This is not a pissing contest about qualifications, or a contest cherry picking research.

It is a rolling catastrophe - decades in the making - for USA people. In general it is that money is being transferred in industrial quantities from everybody to the very rich. Specifically money is getting sucked out of the health system for the benefit of a few at the cost of the many.

FFS socialise your medical system. Spread the cost like sane people and drive out the greed heads

If you can send a rocket to the moon.....


You should have read back up the thread, and see who turned it into a pissing contest and ended up distracting from the moral and ethical arguments by vague quibbling about the methodology.


> and see who turned it into a pissing contest

Not relevant to anything


Agreed! I only mentioned it because "curmudgeon" chose an ad-hominem argument in lieu of a concrete criticism of the methodology.


That's nice of you to list where you work. On HN there are many qualified and intelligent people, so we generally prefer to talk about the issues directly.

> half a dozen standard robustness checks

My original comment isn't arguing they used tools improperly, it's an epistemology argument. It's about the limitations of what information could even be gathered from this kind of study.


I think it's hilarious that somebody who has been here for all of 11 days wants to tell somebody who is coming up on their 10-year HNiversary what "we generally prefer".


I'll let you figure that one out on your own.


Oh, I did. I figure it means you're the kind of person who knows he's going to be a dick but wants to do it under a disposable identity to as not to harm his score. So you're not just a jerk, but a calculated jerk.


This was a reply to cscurmudgeon's objection, not yours. (You can tell because my comment is a reply to his comment.)


I mean... people who go to cancer clinic have cancer at a higher rate than those who don't. Cancer clinics cause cancer. And therapists create mentally unstable people, etc.

You wouldn't have to fork the universe. I think a randomly distributed trial would work, and could be reasonably made to work. Not sure though.

FWIW, I think we should prevent private equity in lots of places, especially hospitals.


I think doing a randomly distributed trial would be very challenging. How do you imagine funding it? How do you imagine getting PE to agree to buy random things rather than strategic things? Do you expect them to turn down 50% of almost-closed deals to keep it closer to random? How would you keep things being distorted by the money you'd be giving to PE to get them to change their behavior?

And what additional knowledge do you hope to gain versus a study design like this one that would be justified by the much larger expense?


> study must fork the universe to compare the same business with different timeline

Unfortunately, that might be what's necessary to know. I am not knocking the research quality, but there are limits on what we can conclude, and obviously other important factors have not been sufficiently isolated.


By this standard, should we not believe that smoking causes lung cancer? After all, a controlled experiment on humans has never been done.


Those studies aren't similar in methodology or subject at all. Can you acknowledge there are unmeasured social factors at play in this simple survey analysis, or not?

If so, I'm happy to speculate about the epistemology of tobacco studies.


> Those studies aren't similar in methodology or subject at all.

They both used observational health data to create a causal conclusion on mortality. How is that not a similarity in methodology?


I mean... people who go to cancer clinic have cancer at a higher rate than those who don't. Cancer clinics cause cancer. And therapists create mentally unstable people, etc.

You wouldn't have to fork the universe. I think a randomly distributed trial would work, and could be reasonably made to work (woth Gov. help). Not sure though.

FWIW, I think we should prevent private equity in lots of places, especially hospitals.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: