RSSAmplifier

Blog

Probably Definitely

apithymaxim.wordpress.comRSS feed ↗10 posts

Latest posts

What are we doing here? Brain supplements edition

Posting my notes from a little research on Alpha BRAIN, which is a supplement for your brain. Joe Rogan recently mentioned it on his podcast as the only such treatment that had undergone a randomized trial with the Boston Center Continue reading

Can you use current AI models for observational causal inference?

My tentative answer is: not yet! It s really common that a randomized controlled trial gives different estimates compared to observational studies. For example, physicians used to think that vitamin E reduced heart disease based on observational studies where they merely Continue reading

What in causal inference is true and non-obvious?

The great mathematician Stanislaw Ulam challenged the great economist Paul Samuelson to name a principle in the social sciences that was both true and nonobvious. Samuelson thought for a bit, then replied, Ricardo s theory of comparative advantage. (source) I ve always Continue reading

Skepticism of the Oakland guaranteed income study

A research center recently released an appraisal of Oakland s guaranteed income program, where low-income families got $500 per month. Oaklandside says that Researchers found a range of benefits in no-strings-attached money, and former mayor Libby Schaaf says This is really Continue reading

Are there more realtors than homes for sale? Stocks vs. flows

This NY Times Daily episode includes the statistic that there are more realtors than homes for sale. It s an eye-catching claim and has been repeated in other places. What s interesting is that it s not wrong, it s just imprecise. It s true Continue reading

Use meaningful names in your code examples

Which of these code chunks is easier to grok? (They illustrate how to nest Pydantic models.) Option 1: Option 2: I think Option 2 is clearly better. It maps to structure that most readers will have intuition for, which makes Continue reading

Measurement error in control variables is bad

I think measurement error as a topic is somewhat old-fashioned, but it s important to keep in mind how much it can bias an allegedly causal estimate. Here s a super simple simulation. We want to study the impact of pollution on Continue reading

regplot: Python’s response to binscatter

Stata s binscatter package has some awesome functionality: it shows a binned scatterplot with a fitted regression line. You can also adjust for controls, alter the functional form, and split it up by group. Python s seaborn package has an awesome near-alternative Continue reading

Minimum working example of clustering sentences

Here s the fastest way I could see for clustering sentence meanings using Python. The KMeans clustering succeeds at identifying the semantically similar sentences in the list. Here s the output: This is the same framework used by Anthropic in their Clio Continue reading

The implausible(?) effects of smoking on birth weight

Hull Angrist make a great point about randomized encouragement studies with imperfect compliance, using as their example the effect of colonoscopies on colon cancer mortality. Medical papers often report only the first stage , the effect of the randomization on Continue reading