Hacker News new | past | comments | ask | show | jobs | submit login
The incentives are all wrong (causal inference edition) (columbia.edu)
63 points by luu 7 days ago | hide | past | web | favorite | 3 comments





Not sure what that's about, but I just do not see the fitted line as reflecting the bubbles.

Why not draw a straight line sloping slightly upwards towards the north side?

Basically, there's a gap near 74 on the north side, and if you think that actually means something then what does it reflect in the real world? If it doesn't mean anything, then the claim of statistical significance seems fishy.


That's exactly Andrew Gelman's point. A different, more sensible analysis, could have (most likely would have) resulted in a nothing-to-see-here conclusion. That conclusion would not be a publishable result, and publishing is the main incentive in the field. Ergo "the incentives are all wrong".

Andrew Gelman by the way, is the father of the MCMC (Markov Chain Monte Carlo) engine STAN (together with Bob Carpenter [1]).

[1] https://en.wikipedia.org/wiki/Stan_(software)


The point was much clearer to me after I had read the precursor article:

https://statmodeling.stat.columbia.edu/2013/08/05/evidence-o...

"What I am suggesting is a two-step: the authors retreat from their strongly model-based claim of statistical significance, and the journal accept that non-statistically-significant findings on important topics are still worth publishing."




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

Search: