>>16891239
I think the fundamental issue with "social sciences" is they're essentially philosophy masquerading as science. And there's nothing wrong with philosophical endeavors, but I'm sick of psychologists and economists and sociologists peddling their shaky theories and suppositions as hard science.

>>16891276
ironically most sociologists and psychologist professors are retarded as fuck they would simply tell their students lack of innate ability within cognitive ability (IQ) is the reason why blacks cant excel at subjects and drop out of highschool or fail even. thats why ben shapiro has to destroy them about standardized testing, being born a genius or making one which is impossible anyway, or if we should be a socialist state which is also another fucking retarded idea seems like a retard fest

>>16891239
its useful for composite calculations like FSIQ though thats the only context in which I know factor analysis

File: 1741961007938637.png (626 KB, 1000x750)

626 KB PNG

>>16891239
blame philosophy

>>16891309
you seem upset that the psychometrics field isnt real science

>>16891310
it's useful if you dont care about valid results I guess

>>16891239
Its a bit better than PCA alone right becuase you account for errors in variables.

also you'd be surprised what you can do with linear methods.

linear methods are good becuase they are fast and you can reason about them.
the moment you pull out non linear stuff it gets very hard, unless you can do some dimensional analysis like is common in chemical engineering (look up buckingham pi).


linear ain't perfect but at least you can reason about it.

trust me this stuff is better than throwing AI at everything, and thats what people do these days.

the amount of junk papers that just use an ANN or what ever to model shit is insane, literally useless publications

>>16891381
cont.

ideally bayes is better and simulating stuff and setting priors but that shit is hard and infeasible a lot of the time

>>16891381
it's plain unreliable and inaccurate for social sciences. Human behavior is non-ergodic. They're agents that interact and mutate eachothers state. You cannot model that with linear equations, you get inaccurate results. Worse than no results, FA will give you some kind of result, that is just wrong.

If you want to at least try to do statistically model human behavior you can try something like CHMMs

>>16891334
its the realist science nigga it actually has reproducibility since so many g loaded batteries have been made since the inception of ravens

you just fucking hate how you are a midwit

>>16891381
indeed anon also thanks for telling about pi it said a bunch of big words about being dimensionless idk what exactly the math equates to tho

>>16891408
I saw a bunch of matrix multiplication thingies and multi dimensional matrices but nothing that reduces the feature set of any problem can you explain it more

>>16891407
>its the realist science nigga
wtf does that even mean? Accurate? It's not even close to accurate

>>16891239
>>16891276
https://www.youtube.com/watch?v=wX5II-BJ8hI

>>16891449
>midwit sci enjoyer not going to backup claims with evidence ok nigga

apparently a correlation of .76 of APM is irrelevant as a test of g the only variable of main pattern recognition in the phenomenon of psychometrics doesnt mean jack delly shit ok so what now we dont believe regression? you know your shitty chatgpt account uses regression nigga KERNELS

>>16891485
the fuck are you talking about, I backed it up in the post you were responding to >>16891392

are you senile?

>>16891392
I do a bit of work using this stuff for simple data and its unreliable then too.
it is sort of convenient for vaguely describing things, but it really misses the point.
you can use PCA and find the most variable features, or some other method like partial least squares PLS to find covariance, but it tells you nothing about causality.

CHMMs seem interesting ill have to look into them.

annoyingly for some tasks Bayesian tools are unusable. they are either computationally intractable or they are too pessimistic i.e. actually evidence based.

if the status quo is a shitty model that looks good, its hard to compete with it with your fancy and slow Bayesian model that gives worse predictions, even if they are actually of a better quality.
science is dead. people just wanna justify what they already believe most of the time

>>16891449
if you dont know what that means then you shouldnt be here

>>16891531
you are lost lmao

>>16891422
>>16891408
yeah the wiki on buckingham pi is unhelpful

the idea is basically say you have a problem of 5 parameters, but some of them are awkward to measure. you can reexpress the problem as a dimensionless value versus something that you can easily measure.

heres probably a good intro:

https://engineering.purdue.edu/~wassgren/teaching/ME30800/NotesAndReading/DimensionalAnalysis_BuckinghamPi_MethodOfRepeatingVariables_Reading.pdf

or a video describing the idea:

https://youtu.be/TaTb6RsSEiQ


this sort of stuff is useful in generalised linear modelling where you can, based on prior knowledge express the process or system as some functional relationship and then model it from there and solve your inverse problem (i.e. regression) or what ever you actually want to do.

tl;dr its infinitely more useful to be able to describe the behaviour of a system with respect to some dimensional qualities rather than just blindly doing PCA / factor analysis and looking at the structure of data points.
hope that helps :)

>>16891567
fuck that video long as shit lmao

thanks anon

so this is basically for modeling 3rd dimensional fluid dynamic shit that is mainly FE, but hand selected? not creating dimensionality in the sense of spreadsheets or ML so hand selecting non repeating ones? thats so much theory just to say select non repeating variables to include in the dimensional array or whatever after transforming them with the equations thats pretty cool

can you use this for orderbook data thats multidimensional and fill in the gaps to engineer some trading algorithm? i might have to explore that reducing dimensionality in any way possible is fun in finance

makes no fucking sense chatgpt lied to me fucking bullshit

yeah nigga buckingham pi is a pile of shit none of this makes any sense my chatgpt was hallucinating the equations and dimensionality of selecting input variables is manual theres no automation so there is nothing like PCA going on with a custom calculation all shieet

>>16891562
I think the problem is that there just isnt good enough tools for some inquiries. Network Psychometrics seems it could be the right approach but I dont know how mature it is. EGA maybe

>>16891627
probably for trading, but time series data is weird because well everything is some random function of time.

at the moment i dont think theres any reliable way of doing trading, other than just gambling on polymarket lmao

the economy makes no sense anymore.

>>16891648
yeah llms are awful at thinking. they can do something someone else has done but yeah they wont be able to do this.

the whole point of the approach really is to reparameterise things in a way you can actually measure and explain things with. so really you have to think about the question you want to answer.

>>16891648
oh and consider trying PLS instead of PCA.

theres also PaCMAP, which is similar to pca but more advanced and can give better seperation.
easy enough to vibe code it with llms to get it working.

cheers

>>16892140
PacMAP

https://github.com/YingfanWang/PaCMAP

but at the end of the day if your data is crap and has like scaling problems you need to do a lot of preprocessing to fix it

>>16892140
bro no

>>16891239
Linearity is proportionality.
You would be suprised at how many things in real life are proportional with one another (with diminishing returns at the ends).
In any case, they are not studying "what function best interpolates this", they are studying proportionality between X and Y.

>>16891381
You have gaussian processes regression, splines, conformal prediction, bayesian networks, probabilistic graphs and so on.
These niggas haven't caught up and are stuck doing what fisher and pearson were doing 100 years ago

File: oh-no-george-carlin.gif (78 KB, 220x220)

78 KB GIF

>>16891239
>Who ever came up with the retarded idea to try to statistically model dynamic processes as linear equations?
Low IQ thread.

>>16892141
this doesnt make any sense

Boudieu was a big offender. He really liked MCA and claimed it was much more that what it is.

I think the dimensionality reducer pi theorem is gay as fuck

>>16891239
i am a brainlet
but wouldn't this only work if you made factors for every possible subset of the given variables, thus exploding your problem complexity to an exponential degree and wiping out any gains in calculation efficiency?
seems like if this was reliable at all even with proven correlations and a huge amount of empirical data... using it to predict the weather with a small sub-samble of weather data would be trivial.

i think "can it be used to accurately model and thus predict some part of the weather over even medium timescales" should be the real litmus test for how useful stuff like this is in understanding complex systems. that dataset is continuously generated at high reliability and has a ton of advantages over most sociological data, so if your "i don't want to admit this data is basically worthless for empirical mathematical study" modeling trick can't do the weather it probably can't be more accurate about sociological systems with less and worse data.

>>16891567
good video

>>16891313
:^)

>>16891276
The origins of science are from philosophy.
Philosophy didnt used to be a dirty word before people ruined it.