Moral of the story: IV designs have much larger uncertainty + publication bias/file drawer --> Much bigger IV estimates than OLS (with potentially much greater biases).

Our recent paper on toxic political ads, if anyone is still looking for light weekend reading:

apoorva.lal No it's simpler than that. Just stacking regressions with fixed-knot 1D splines.

'Today, Florida is fighting back against the global elite’s plan to force the world to eat meat grown in a petri dish or bugs to achieve their authoritarian goals.'

Completely bonkers.

Remind me, again, who’s deciding what people can and can’t eat?

In this vein, I think often about this Matt Graham paper. When polled, zoning/land-use policies often show substantial portions of the electorate w/out an opinion on these issues. This means that the people who do have opinions don't have strong ones!

m-graham.com/papers/Graham_…

weather and crime

pubs.aeaweb.org/doi/pdfplus/10…

Periodic reminder that Heathrow airport is wider than Manhattan

Relatedly: The research people learn of early in grad school is almost exclusively from the top 1% in quality.

Partly why second-year Ph.D. students are a brutal audience—they are comparing the random paper they are seeing to the top 50 ever.

just occurred to me impostor syndrome might also be a form of network effect: on average, the peers we interact with are more successful than us, because more successful scientists have more exposure

Mick Jagger is 80!

I'm trying to learn how NaNs are encoded in floats, apparently there are many extra bits in every NaN that can be used to encode side-information (en.wikipedia.org/wiki/NaN#Signa…). Has anyone looked at encoding information for debugging NaN-outs? Seems like it could be very useful.

Important new The Lancet article examines mortality among 1.5m people released from prisons in 8 countries from 1980-2018. Death rates peak in the first week of release, driven in large part by alcohol and other drug poisoning (h/t Evan Roberts). thelancet.com/journals/lance…

The 1.6pp forecasting error two months ahead is larger than all but one forecasting error from when these forecast data start (in 2007) and COVID. And the previous similar-sized (but opposite sign) error was 2008-Q4.

Really enjoyed Lihua Lei talk today on model agnostic covariate-based tightening for partial id bounds. Very thoughtful work: as.nyu.edu/departments/ec…

gojiberries.io Dan Goldstein psycnet.apa.org/record/1979-22… This is a wonderful paper of his on how lack of ability can co-exist with confidence.

gojiberries.io This is one of my favorite papers. I thought at first it was specific but now I realize it's general and part of why we think the world is more predictable than it is

The illusion of predictability: How regression statistics mislead experts
E Soyer, RM Hogarth

!!

Omar Wasow | @[email protected] Nicholas A. Christakis Recent study of protests finds, with one exception, no effect on public opinion and elections.
nber.org/papers/w32342

Flatter rent gradients/bid rent curves — e-bike edition