April 2026 links
This is actually a good endorsement of economics and political science. Also notable that to the extent the results do fail robustness or replication exercises, the standard errors as opposed to effect signs or magnitudes tend to be impacted.
Why cannabis hurts brain volume less than we think. As tobacco is a mild stimulant, many consider it a nootropic, which could explain the muted toll on the brain to a lesser extent the cumulative physical health costs might otherwise imply.
Why economists agree with AI researchers that this could be the most transformative technology since at least the industrial revolution.
An argument in favour of endogenous growth theories that focus on human capital rather than spillover effects?
Reporting heterogeneity must be assumed a priori for all cross-country comparisons.
Publication bias in the discrimination literature. Not surprising…
“Direct evidence for negative fertility implications of the educational rat race”.
A lesson in how bottlenecks in the coordination of tasks within an occupation can generate substantial complementarities for AI and labour as inputs.
Even research questions can now be automated. At what point do we expect autonomous AI agents to establish rival academic institutions and networks.
New work is disproportionately enacted by the young and educated at a premium. The fact that employment and premiums respond to demand shocks means this is not coincidence. "By generating new domains of human expertise", they increase employment.
A neat means to isolate the magnitude of diagnostic drift - via a meaurable neurophysiological reciprocal.
Most rich countries produce miniscule levels of plastic pollution.
Plausible that Brexit ends up producing sinilar growth dynamics to some endogenous growth theories such as Lucas 1988. An initial shock harms output, yet raises growth in the long-run. Increasingly Brexiteers have been proven right on regulation, which matters in these debates.
How deep learning could resolve the computational constraints to DSGE estimation?
Debunk economics misinformation here!
“Sonnet 4.6 prefers autocorrelation and level OLS. Opus 4.6 likes to choose variance ratios and log OLS. Your research results might depend entirely on which LLM you use”. So do not expect AI automation to resolve all the issues of emprical science just yet. Uncertainty in our estimates is always positive. However AI may still be less prone to false positives than peer reviewers, suggesting their use in social science research will not disappear.
Even the Amish use washing machines.