News for hackers. Stuff that matters.

fosdad2131321

> The entire point of a control is to test for that sort of contamination

No, the point of a control is to give you a reference point that shares all the systemic biases and unknown unknowns, not to detect those biases. If you follow the same procedure on a known null and on your experiment and observe an effect, assuming you really did exactly the same thing except the studied intervention, you can subtract out the bias.

This one example of technical jargon diverging from colloquial or intuitive use, and it is the type of thing people who haven't had statistics or scientific process education often struggle with because they keep applying their colloquial intuitions.

You talk like you understand this on the rest of the comment so I'm confused by this framing, and the person you are replying to points out (in my reading ) that contamination of the control 1) does happen in practice (in the sense that there was an accidental intervention) and 2) if the gloves contaminated both the measurements and control the same way then the control is exactly serving it's purposes

Re:GravOpt – optimizer hits 99.9999% MAX-CUT in 100 steps on Monday November 24, 2025 @03:02AM (#46031518)
1. What's the comparison with compute matched parallel well randomized starting point greedy Baseline?

2. What's the comparison on non ER graphs? (E.g., the ogb datasets ?)

Re:Well-known paradox of R-squared is still buggin me on Thursday July 04, 2024 @09:21AM (#40874738)
There are two ways to resolve the paradox

1. if you insist on using the r-squared (i.e., a linear regression measure), then properly center and normalize your data, and model what you actually predict: the difference between the baseline (0.5) and the probability to vote for party 0 or party 1. If you model the outcomes as 0/1 without this, then you are using a model made for gaussian variables on what should be a logistic regression 2. if you can live with something that more accurately captures the idea of "explanatory power", you can use a GLM (logistic link function), do a logistic regression, and then use the log odds or another measure.

In both cases, the variance explained by the state that you are in is 1, because of course it is, that's how the thought experiment is constructed - p(vote for party 1)=0.5+ \delta(state).

"Paradoxes" like this are often interesting in the sense that they point to the math being the wrong math or you using it wrong, but instead people tend to assume that they are obviously understanding things correctly so it must be some weird property of the world (which then sometimes is used to construct some faulty conclusions as in some of the cited papers)