Comment by dahart | Hacker News Reader

Comment by dahart | Hacker News Reader

Parent Full thread

What exactly is the problem with that sentence? Are you sure the problem isn’t that you don’t understand what it means?

Here are a few links explaining the terms:

https://pmc.ncbi.nlm.nih.gov/articles/PMC7384548/

https://en.wikipedia.org/wiki/Poisson_regression

https://lost-stats.github.io/Model_Estimation/Research_Desig...

I don’t know why people distrust science. Sure, it’s not perfect, and scientists, like all people are subject to human problems. But there’s nothing else in the history of the world with a better track record than science. I feel like the problem is that some politicians spread FUD and prey on people’s insecurities, and unfortunately it tends to work, disproportionately on people with less resources. The problem isn’t science at all; the problem is people and politics.

I know what a linear regression is and how to examine event studies, lol. What you don't understand is that the author is leaning on linguistics to insinuate strong evidence of causation where it doesn't exist. If this was a quant in finance, they'd be out the door in days.

"The problem isn’t science at all; the problem is people and politics."

Agree completely.

> the author is leaning on linguistics

Please elaborate. I haven’t used entropy balancing or difference in differences, but those articles explain that their purpose is to try to tease out causation. What - exactly - is the linguistic trick, if they actually did use an entropy balanced Poisson regression and difference of differences?

Entropy balancing cannot fix unobserved confounders.

"Teasing out causation" is exactly why this methodology fails. You are confusing the intended purpose of a statistical tool with its real-world validity. No one is questioning what an Entropy-Balanced Poisson Regression or a Synthetic Difference-in-Differences model is designed to do.

The issue is that the authors have profoundly violated the mathematical assumptions required for these tools to actually function. Throwing high-level econometric terms into an abstract does not make the underlying logic scientific, but rather acts as a linguistic tuxedo on a fundamentally broken causal claim.

If you cannot see through economic (and other) confounders that invalidate their approach and their biased statements, I cannot help you. This isn't science. Getting an LLM to run an SDID model and spit out a result doesn't = science.

Reply on news.ycombinator.com