wild to know that every time I'm on Bluesky, somewhere obscured and opaque to my sight and knowledge is endless roiling, cascading fields of Drama, occasionally interpenetrating my sense faculties full of sound and fury, like seeing lightning flash from a mile away, the sky lit up with death threats

Variable correlated with GDP, or log(variable) correlated with log(GDP)?

What if you don't want to predict life outcomes but instead want to understand how people think?

Agree that it can't be resolved by switching statistical models.

The main disadvantage here relative to a Q-sort is probably that standard psychometric scales are not very exact, so this won't be predictive on an individual level. Whereas maybe if you start out by ranking X and Y relatively within-person, maybe it becomes more accurate on an individual level.

I'm not sure how necessary it is to do it like a Q-sort, by the way. If you have some outcome variable Z that you are interested in, then you can put X and Y on the same scale by regressing Z onto them, and then you can compare the predicted outcomes from X vs from Y respectively.

But yes, the idea of rating variables relative to each other within a person, instead of rating people relative to each other within a variable, is very much Q-sort-inspired.

My thoughts on this were inspired partly by thinking about Q methodology, but Q methodology still AFAIK tends to focus on the overall pattern of the variables rather than the strongest variable, and it doesn't do root cause analysis and therefore also suffers from symptoms rather than causes.

One can then iterate, e.g. screen for people who have the outcome without having the main cause, and see if they have some other cause. But if you do that, the clustering will sort of emerge naturally without any algorithms.

... the subset with the outcome in question will probably already be a minority of the sample, and it's probably more productive to be looking for the main cause, rather than untangling all the causes.

... Clustering makes the most sense if you have some polycausal outcome of interest (e.g. sickness), and you want to group together people with similar causes for the outcome. But if you study people "representatively" (or, y'know, college students, point is generic sample)...

Yep. Though I also think in most psych journal articles, there would be a lack of sample size necessary for clustering to make sense. ...

Also, for it to work well, one kinda needs to precede it by a root cause analysis so one isn't just getting confused by superficial symptoms, and statistical clustering algorithms just don't do this at all.

This doesn't work well with standard statistical algorithms because those algorithms try to capture the whole patterns of issues the patient is facing, rather than the most severe one, and thus the clustering is willing to trade away accuracy in the worst issue for accuracy in the holistic view.

That is, patients might have a number of issues, and these issues can be ranked relative to each other in severity. The cluster the patient belongs to is then defined by the most severe issue they are experiencing.

I think the biggest part of the problem is that clustering algorithms are misspecified and could not be correctly specified. Useful clustering is often more like argmax than like the conditional independence that clustering algorithms usually optimize for.

Sent an access request.

Is there anywhere one can get the bot weights directly, or an API for computing correlations for pairs of items, or an API for items, or similar?

See deep double descent

I've got similar numbers in surveys on reddit.

Different tests give correlated results.

Pretending virtue seems worse, but open hypocrisy seems better.

Well, it's up to you, I can't force you to fix yourself.

Emotions are there for a reason.

Some judgements feel too shameful to broadcast, and then it's ok to seek a less judgemental community or a better thought-through judgement before broadcasting it.

Develop your own judgement in places where you're currently deferring to another person or to the consensus of a community. And then broadcast your judgement to others.

It helps others achieve higher states of consciousness.

Then you'd probably have a fun time in the psych ward. At least that's my experience.

benjaminrosshoffman.com/predation-as...

A challenge with AI adoption is that organizations are not built to a Grand Plan where AI can just be slotted in by leaders, but rather socially constructed, semi-random & in flux

Here's an anecdote from a paper on how a CEO realized he didn't understand how things really worked (& that nobody did)