its over ppl. Design based inference has won, the next 5 years are just a coronation
imagine caring about MLE lmaoooo we have R packages for that sweetie
"CI is an unscientific fad" .... that happens to be incredibly in demand by Facebook, political orgs, digital marketing, data journalism, policy evaluation, NGOs.
Anyone with money who cares about getting things right (so, non-academics) does CI and will continue to do CI
Clearly, Facebook is the acme of social science! Also, the trope "X and only X is True Science and is the future" is a recurrent feature of academic rhetoric. It's word for word what the rat choice people used to say. Anyone heard from them lately? Fads gonna fad. It's not that this stuff goes away, -but it gets absorbed and there's a new craze.
its over ppl. Design based inference has won, the next 5 years are just a coronation
imagine caring about MLE lmaoooo we have R packages for that sweetie
"CI is an unscientific fad" .... that happens to be incredibly in demand by Facebook, political orgs, digital marketing, data journalism, policy evaluation, NGOs.
Anyone with money who cares about getting things right (so, non-academics) does CI and will continue to do CI
Anyone with money who cares about getting things right (so, non-academics) does CI and will continue to do CI
This is a really bizarre statement. The trends are moving towards big data analysis as the de facto in demand skills among many larger companies. Big data usually doesn’t involve causal inference.
There will always be a market for people having basic or advanced statistical skills - most governments and companies rely on this type of analysis, not causal inference or experimental methods.
The size of your data is completely orthogonal to causal inference problem.
Sometimes you want to measure causality, sometimes you don't. Only in a handful of fields causal inference is so important, which makes polisci (and econ people) to think about ML and applied stats in the terms causal vs. predictive. But the reality is that this split is very non-informative about the actual work done in more ML heavy areas.
Anyone with money who cares about getting things right (so, non-academics) does CI and will continue to do CI
This is a really bizarre statement. The trends are moving towards big data analysis as the de facto in demand skills among many larger companies. Big data usually doesn’t involve causal inference.
There will always be a market for people having basic or advanced statistical skills - most governments and companies rely on this type of analysis, not causal inference or experimental methods.
The size of your data is completely orthogonal to causal inference problem. Sometimes you want to measure causality, sometimes you don't.
I never stated it was one or the other. But if you think most big data is set up in a way to do actual causal inference stuff, you are solely mistaken. It's a bunch of disaggregated data with a bunch of noise largely collected independent of each other. Yes, you can do certain quasi-experimental approaches with it, but it's rarely structured in a manner that allows you to do this cleanly. It's largely just brute statistical work with extremely large-Ns to increase confidence.
I'm not really sure what type of problems you have in mind, but I've been working as a DS/RS in FAANG and your description doesn't seem to match any of the projects I've seen.
There is nothing magical about more data, it's just that it conflates with other type of researchers that are interested in other types of questions. If the objective is trying to classify content of pages, people don't need to a have a robust and interpretable estimate of how each parameter affects the outcomes. If the objective is to measure the sales impact of an online campaign, then having a robust and identifiable estimate of a parameter is important.
There is no amount of data that will solve your identification question. And the idea of brute force really doesn't capture well how ML problems are solved - if anything you would spend more time thinking about the model itself that in a causal-inference setting.
The size of your data is completely orthogonal to causal inference problem. Sometimes you want to measure causality, sometimes you don't.
I never stated it was one or the other. But if you think most big data is set up in a way to do actual causal inference stuff, you are solely mistaken. It's a bunch of disaggregated data with a bunch of noise largely collected independent of each other. Yes, you can do certain quasi-experimental approaches with it, but it's rarely structured in a manner that allows you to do this cleanly. It's largely just brute statistical work with extremely large-Ns to increase confidence.
its over ppl. Design based inference has won, the next 5 years are just a coronation
Yes, the increased attention to research design is a positive thing and will (hopefully) remain with us.
But that doesn’t mean we will continue the current trend of research that carefully identifies effects on a very specific sample but tells us nothing about the broader world in any meaningful way.
She should be up for tenure like last year and yet only has 5 publications. How does this make any sense?
she negotiated a 2 year maternity leave with UCLA, from my understanding. Wouldn't that stop the clock? or does the clock continue on maternity leave?
So she has a rich hubby who funded this, presumably with childcare? This sort of thing should shorten the tenure clock, not lengthen it.