-
Impact of coal power on excess deaths.
There is a nice figure from Mortality risk from United States coal electricity generation regarding the impact of coal generated pm2.5. Initially, I was a little bit skeptical as coal pollution is presumably associated with poverty which has myriad health implications. On the other hand, this figure is very compelling which shows that in a number of cases, total excess deaths decreased after a scrubber was installed. Hard to argue with that data, which is helpfully presented as a rich time series – allowing one to draw that conclusion without being told!
-
Cartel Size Paper in Science
I just finished reading through the main paper on [Mexican Drug cartel sizes in Science]() (I still really need to read the supplement, which is where all the actual details are…). One of the things I love about Science and Nature is that they will publish “preambles” or “contexualizers” for articles. [In this case, there is a lengthy preamble by Caulkins, Kilmer, and Reuter]() that – I suspect – was the result of a review of this paper that pointed out a number of issues that would be impossible to resolve. In addition to pointing out what seems to be a fairly large hole in the specific justification (e.g. that the result is indeterminate because it’s a ratio), it attacks the entire mathematical modeling edifice itself as too crude an approximation for the complex dynamics involved in cartels “Cartel members are not billard balls or atoms locked into mechanistic reactions to external shocks”. Yet, while this is done with concise and technical precision – the rebuttal is ultimately laudatory and accurate captures the ambition and scope of the effort. It’s nice to see people agree on the utility of big ideas and insights even if they may have huge disagreements on the details – and it’s also nice that Science tries to support this style of debate.
Now, for my own take, I think the objections to the mathematical modeling have the potential to be overblown. The goal with this type of mathematical modeling is to capture just enough of the truth to be useful. In this case, it seems any more detailed modeling would be impossible to do. e.g. the problem with more complicated agent based models is that it’s hard to know if you have baked in any effect due to the extremely large number of choices they require and establishing the this is not so is often problematic and time consuming.
-
Prediction-powered inference
New paper in Science on using ML-based methods to improve classical confidence intervals or p-values. (also on arXiv) The idea is to use a large collection of unlabeled data along with an ML method to estimate labels on the unlabeled data. Then we can use that data to improve the confidence interval given all the data. This reduces the amount of data you need to make a key “discovery” as shown in Table 2.
-
Light reading...
I reviewed a book proposal for Cambridge University Press. They often gift books to people who do this, which I love. While I was looking at their catalog, “Quantum Mechanics: A mathematical introduction” by Larkoski (of SLAC), caught my eye. It’s supposedly a quantum mechanics course based on linear algebra.
It’s been a while since I’ve read a preface that makes me fairly convinced I’ll enjoy a technical book. The perspective articulated is exactly the kind of derivation and analysis I enjoy and learn best from. It’s technical and explanation towards a goal. Supported by simple methods.
The first few chapters have been fun to read. Although, this is material I know fairly well, so we’ll see how things go once the depth picks up.
subscribe via RSS