There’s several elements to this argument and I will just go through them in steps. First, there is my marginal product argument. It is my experience that causal inference raises the marginal product of the econometrics professors work, and not the other way around. 2/n
-
-
Prikaži ovu nit
-
Routinely, students who take my class at the end tell me “ah hah! I finally get what we are doing in econometrics!” Now, here’s the thing; it’s debatable that I have helped them better understand econometrics. 2/n
Prikaži ovu nit -
But what I have done is giving them a *reason* to interact with that difficult econometrics material as a beginner. And that’s what is currently missing in our training of economics students. Oftentimes, students are exposed to econometrics without useful priors. 3/n
Prikaži ovu nit -
We think primarily in terms of technical prequels like calculus or a probability class. But then students can not articulate to themselves or each other just what these parameters are they’re estimating or why they should care truly about an estimators properties 4/n
Prikaži ovu nit -
What I believe they are missing is something to hang their hat on to justify why they should exert so much effort in learning econometrics, and causal inference does that. And here’s my second element to this argument. Causal inference crowds out nothing 5/n
Prikaži ovu nit -
It’d be different if we were saying “well if they learn to think about econometrics purely in terms of estimating causal effects, then that crowds out this other reason I believe they should be thinking about”, but we aren’t. Students come in blank slates and are often lost 6/n
Prikaži ovu nit -
But if they had causal inference first, it becomes a useful crutch to help them through the material. It fills a void rather than replacing some other crutch. Many students just do not understand at a deep level why they are doing any of that; they learn why late or never 7/n
Prikaži ovu nit -
But another reason I think to do causal inference first is I think insofar as it raises the marginal product of econometricians, then it actually could raise demand for econometrics courses. That’s speculative, I get that, but I believe it’s credible. 8/n
Prikaži ovu nit -
What I have seen is exposure to potential outcomes, DAGs, and these designs empowers students - honestly, much like elementary game theory empowers students too. They begin to think questions are within their reach. They develop optimism inside themselves, hope, confidence. 9/n
Prikaži ovu nit -
But they know they need more. I suspect we pull the marginal economist back on the track if we could have causal inference come first, and if we could, then econometricians gain bc their classes fill just slightly more. May sound optimistic but I believe it. 10/n
Prikaži ovu nit -
And what that means is a few things. One, can help with retention. Two, can help with excitement. Three, can help more generally with all our course offerings to handle the massive shift towards empirical work within the profession. 11/n
Prikaži ovu nit -
If causal inference is brought early, even mandatory, then it provides a very basic worldview that enhances every other faculty’s micro offerings so that they can take certain concepts for granted. And that’s huge bc you want them to know DD before they take labor. 12/n
Prikaži ovu nit -
I am not in the camp that causal inference is a substitute for econometrics. I believe it is a compliment. But more to my argument, I believe pedagogically it will be its best complement to that if you teach it first. You make causal inference the *prerequesite*. 13/n
Prikaži ovu nit -
I think we have a perfectly good useable model of that course in my causal inference Mixtape and my workshops and class, which is really just a distillation of how labor economists have been thinking for decades, with some Cunningham quirks. 14/n
Prikaži ovu nit -
First, you go all in on potential outcomes. ALL IN. I mean, deep. You go through the history of thought on causal inference and how counterfactuals seems to be what humans have converged on to think about causality. You tell stories and a lot of them. 15/n
Prikaži ovu nit -
Then you dig into the notation, but unlike
@autoregress , you do it correctly with Y^0 and Y^1. Convince them it’s simple and in the right hands has become able to move mountains. Then you do DAGs. I will die on that hill pedagogically bc I’ve seen it: kids love DAGs. 16/nPrikaži ovu nit -
Watch their face when you explain IV inside DAGs and contrast that with some counterfactual pedagogy. The alternative is less effective in the mode. They like learning these words, and working through the do calculus. Can’t we just throw kids a bone sometimes? 17/n
Prikaži ovu nit -
And then you do all the canonical designs, with lots of papers, and lots of coding intensive replications. And that constitutes the first course in a micro sequence. And the reason is, all contemporary micro either is all about that, or it’s heavily in conversation with it 18/n
Prikaži ovu nit -
So in conclusion, I think causal inference as I teach it (which sounds more arrogant than I intend it to sound) should be the paradigm for our curriculum at the undergraduate level, and absolutely at grad level. Even if just strategically, you should consider it. 19/n
Prikaži ovu nit -
In the right hands, used properly, a causal inference class brought in early can energize a students mind and heart. It saves her a lot of wasted time - even avoiding major errors that can be nearly impossible to push them off of later. 20/n
Prikaži ovu nit -
So I am firmly in the camp that we should build statistical inference on top of causal inference and not the other way around, even if in the students longrun, she completely ignores causal inference and goes her own separate way. 21/n
Prikaži ovu nit -
Causal inference as the base is like a surge that raises all ships. Again, either a controversial take that will get me ratioed or so obvious that it doesn’t even merit feedback, but I wanted to share my beliefs about curriculum. 22/n
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.