Imagine making a diagram that tells how to turn empathy, loneliness and low self-esteem into profit, and not thinking you are the scum of the earth.
That makes it worse because that suggests this shit actually works.
How the fuck did they get that significance level with that numerical precision, what the fuck?
Also, listening to advertisement is bad :cat-confused:
In Thanksgiving my little cousins wanted to play games on my phone, I hadn't played anything other than the odd chess match and it's insane how many ads they cram into every minute of gameplay. I taught them how to use airplane mode to avoid ads, I don't want them getting brainwashed.
Its sad because i think it has alot of potential if done right, especially if you do it longitudinally with mixed effects and what not
this diagram is bullshit because it had barely any observed variables feeding into the latent variable to pick apart.
edit: jk I just read the subscript saying they hid them for brevity
The problem with latent variables in psychology is that you can't empirically measure the construct the explicit variables are ostensibly tapping into. It's a huge issue in social psychology (my field of study), and I'm not sure how we can best deal with it.
agreed thats what higher levels of stats are for like SEM.
Sure you can’t directly measure but you can use variables that are measurable that make up the construct. think of it like putting a blanket over something like say a ball you can still gain of information about the ball and its relationship to other things, its properties etc. sure its indirect but it can absolutely be more rigorous than things are now. each latent variable in this is likely a sort of amalgamation of observed variables that make it up. thats the best way I can explain it tbh. the statistical method used here absolutely has merit IF DONE RESPONSIBLE AND CORRECTLY. the current incentive model biases these though to push for “positive” results being more publishable . despite insight still being available from less flashy results
I know what you mean by this, but even if you measure something very accurately, there is philosophically no way to be sure that it is mapping on to the latent construct you claim it is. The usual methods that are used in psychology (reliability analysis using Cronbach's alpha, principal components analysis, and factor analysis) don't actually tell you if your explicit variables actually relate to your latent variable; they just tell you how well your measures correlate with one another. It's not like physics where we can make mathematical predictions about how an electron will behave when manipulated in a very specific and controlled way. Human psychology and behavior can be influenced by so many factors that it's effectively impossible to conduct any experiment and be assured that the results are due to your experimental manipulation and not some other factor you were unable to control for.
This short write-up is a better explication of the problems with SEM and causal claims that are often made with it than I am able to make, hope you find it interesting: https://www.rasch.org/rmt/rmt221d.htm
I think that until we (speaking only for my field of social psychology) have developed a more sound methodological foundation that we're better off doing qualitative, bottom-up research instead of conducting experiments and making causal claims.
I really appreciate this wrote up and that link Ill look into it. Of course its not an exact science but I absolutely believe in the potential SEM related methods like Latent growth models etc has especially done longitudinally with mixed effects which improves a case for causality. the thing that sucks is most of social psychology (hell majority of psych) are doing anovas and regressions instead of taking it up a notch if you feel me? regardless i appreciate your devotion to trying to keep social psych grounded as a field! psychology as a whole has a lot of stinkers in it haha lots of p hacking, file drawyer, and false positives etc lol
I agree that longitudinal mixed models are better for causal inference. Wish I had the same optimism as you about SEM methods!
Yeah, there's a lot of bad work done in social psych, but the field is waking up to that which makes it an exciting time - lots of work is being overturned and we have an opportunity to fix things from the ground up. I hope we can do work that is actually useful to more people than just marketing agencies lol
The way I see it those investing in methods enough to be using SEM acknowledge the limitations and with enough education (which is really pumping even 3 years ago when i was in a lab) ways of applying it can definitely be shifted imo. When I was taught it they definitely said use a metric fuck ton of salt lmao.
I think shifting to a Hierarchical linear approach will will also be necessary as many of these things are nested. tbh i haven’t had enough advanced knowledge to know how the methods have been applied to together other than intra individual approaches in latent growth curves (i worked with them independently and probably a rather surface level tbh) you are much more knowledgeable though than me so Im sure your skepticism and lack of optimism is very warranted
its sad because SEMs (the name of the stats they are using) has so much potential 😭
