David Slichter

Conceptually, what goes in your school VA? Sulagna and I found that variation in school quality within county/district was ~80% teacher quality, but there are substantial county/district effects which are not explained at all by teacher quality.

Looking forward to reading your paper!

Reasonable people can absolutely update their views about causal questions based on evidence which does not involve obviously as good as random assignment of treatment! Famous examples: effect of smoking on lung cancer, effect of CO2 on global temperature.

This is a factual question and the answer is that there exist sets of assumptions which do not include exogeneity but which identify causal parameters.

But also, if we average across many research questions, some methods will usually perform better than others. This is how I usually interpret the idea of the hierarchy of evidence. This doesn't mean that, say, RCT is universally better than DiD regardless of sample size, context, etc.

Sure, not a strict hierarchy. But the typical distance between the parameter you want and the estimate you get is not the same for all methods.

(2) If so, is the ratio of long-run to test score effects for these policies similar to the implied ratio e.g. in that JPAM paper, in which variation is not necessarily even coming from any state-level policy, let alone the same one? (3) Would the same thing happen if other states tried to copy?

Absolutely, things which increase test scores usually help long-run outcomes. My main uncertainties: (1) Were these trends actually caused by the state ed policies people have in mind?

Test scores, schools, and the geography of economic opportunity Do standardized test scores in a community indicate whether schools there are effective at producing human capital? Counties with high average test sc…

This is a nice article and the story makes sense. But worth remembering that the logical step "this place is doing well in standardized testing, therefore everyone should copy what their school system is doing" involves assumptions which don't always hold:
www.sciencedirect.com/science/arti...

AnotherSpaceSavingTipIsThatIfYouCapitalizeTheFirstLetterOfEveryWordYouDon'tNeedSpaces

Darkly funny how a strong emphasis on unbiasedness via the credibility revolution plausibly ended up making the published literature more biased.

Both policies produce winners and losers, so I wouldn't call either one costless. If you mean housing reforms increase the size of the pie, well, cutting taxes reduces DWL. But utility gains from housing reforms are more widely shared across income brackets, because indirect effects are stronger.

I'd define trickle-down as directly helping rich people with the expectation/excuse that this will indirectly help poor people. Helping the poor via vacancy chains is exactly that! But in housing, unlike taxes, the indirect effects are actually large enough for the argument to make sense.

(Though I can't vouch for the accuracy of claims about the psychology literature contained within...)

I always really liked this piece:
www.alfiekohn.org/article/five...

For OLS, this is simpler than Aronow and Samii (2016 AJPS) (or pretending you had an IV equal to treatment residualized on controls) because you don't have to construct residuals and inference is simpler.

How does the estimated effect in the first year compare with the OLS correlation between income and subjective well-being?

In that case, prices of imported goods increase, then stay (let's say) 15% elevated forever. That makes all future prices high relative to today, but doesn't make prices 30 months from now high relative to 29 months from now; so, it doesn't increase inflation measured 30 months from now.

Both apply. To an approximation, tariffs cause a one-time price increase. I would also bet on the courts ruling against the claimed tariff authority.

Good luck and don't forget to consolidate control of the security forces *before* implementing your vision for the department!

Was this the paper you're responding to? (See Figure 17 on page 179)
mdoepke.github.io/research/Doe...

The bastards!

Increase the tax base with bass taxes

Fine, I will accept a temperature scale equal to 2*C. This will save a digit from every thermostat outside the US. We can name it after somewhere in Poland. Deal?

Witchcraft!

Fahrenheit is the unit of measurement where 1 degree corresponds to the minimum temperature change I can detect and have an opinion over.

Celsius is the unit of measurement which makes it easiest to compare the temperature of your house with the boiling point of water.

Counterpoint: All other countries should switch to Fahrenheit.

I would love to see more papers do some basic Fermi problem style analysis to justify what might constitute a small or large value of the parameter they're estimating, then explicitly comment on how their empirical analysis should lead you to update between large vs. small values.

Not thinking about the magnitude of a priori uncertainty also leads to selective reporting of results based on significance instead of total imprecision (model + sampling errors).

At the same time, not thinking about magnitudes means no framework to accept a design as having acceptably small but nonzero bias. Hence the pressure to produce zero-bias designs even at the expense of severe variance and external validity problems, or of estimating uninterpretable parameters.

Yes! In econ, the fundamental problem is giving no thought to what effect sizes are a priori plausible. One result is that people have no framework to think about how much variance is too much, so papers don't get evaluated based on magnitude of SEs (except when results are non-significant).