Ongoing article board #10

markjrieke · 2024-09-11T20:32:18Z

Schedule

Here's the abstract & why I'm doing this (see previous tweet examples)
Confirmation of bias/variance reduction in continuous outcomes
Confirmation (mostly) of bias/variance reduction in binary outcomes
Simulation --- how much correlation is needed to see variance reduction
Simulation --- can improve parameter estimates by incorporating reported uncertainty directly.
Notes:
- see the linked paper in this tweet
- See this thread in which I replicate the results for a unit-level continuous outcome, but not for a unit-level binary outcome
- See this tweet in which I admit the folly of my ways

election model in detail (thicc math)
a non-centered parameterization of the multivariate student-t distribution (see here for how to add $\nu$ as a parameter, rather than data)
Math and randomness of DnD
Error handling / tryCatch wrappers
Shipping a quarto document as a part of a package
Quarto code factories: write a variable number of quarto sections in a report
How you can extract a distribution for the probability of an event when probability isn't a model parameter

Information removal in Bayesian models (i.e., can we simply subtract from the log-prob "undo" prior information/data we want to forget?)
Half-normal parameter recovery (if I figure it out lol)
modeling from first principles: basketball
NCAA march madness forecast (difficult difficult lemon difficult)
Measuring pollster herding
- feels like you should be able to get a state level measurement of how much a pollster herds by comparing the probability of their result (given a trendline) to a uniform distribution over (0:1)
- degree of herding depends on the model/qty of polls
- not sure how this should affect an average

See here for conf materials from 2024