, 19 tweets, 8 min read Read on Twitter
1/ Alright folks, this will be the last round of #EconBookClub, discussing Guido Imbens' new paper on PO vs. DAGs (link: arxiv.org/abs/1907.07271). #EconTwitter #BookofWhy #Econometrics #Causality #CausalInference #MachineLearning #AI
2/ Today I will go through the final chapter 4.6 and the paper's conclusion. Maybe you will notice that I skipped over section 4.5. I had some thoughts on this chapter, but I feel that I didn't understand Guido's point well enough. So I prefer to leave it aside for the moment.
3/ In chapter 4.6, Guido discusses the returns to education example, which is a classic question in economics and which is also closely connected with the development of PO techniques in econometrics.
4/ He discusses a couple of seminal indentification strategies – among them covariate balancing, instrumental variables, and fixed effects methods – that have been used by economists to tackle this question, and presents them in DAG form.
5/ Finally, the chapter ends with this conclusion:
6/ So the message that you can read between the lines is similar to what we've seen in previous chapters. DAGs might be a nice visualization tool here, but PO was already successful enough so that we actually don't really them anymore.
7/ Unsurprisingly, my answer will be similar to before too. Sure, there's not much need for DAGs when using only the four identification templates we're already familiar with. But it's exactly about moving beyond these simple templates.
8/ Finding the right hammer for the nail, not a nail for one of our four hammers, that's what DAGs are about. There is, for example, a lot to be said about Guido's unconfoundedness graph.
9/ Why are there only two confounders (I know it's a stylized example)? How do they relate to each other? Are there other variables that we could possibly use to block backdoor paths? Are there colliders we need to watch out for? Are some variables maybe
10/ only affecting education and not earnings, so that we can use them as instruments? 45 years since Mincer and we came no bit closer to answering these questions, because we always restrict attention to four simple templates. Isn't that sad?
11/ I think we should start tackling these big questions, even if they might be tough to answer. And in my view, DAGs could be a great tool to store the knowledge we obtain – step by step – about a field over time.
12/ DAG methodology is so rich by now. It offers a unified framework for dealing with so many of the daily issues we are faced with in applied empirical work. It's really worth having a look at it! dropbox.com/s/pcfxmcncofnq…
13/ Also – and this is something where I firmly disagree with Guido – the idea of structural causal models and interventions therein is so natural to economists, if we could just get used to the boxes and arrow, we'll find them to be very compatible with our methodology.
14/ The main message of the #BookofWhy (which is also touched upon in the recent @SamHarrisOrg podacast with @yudapearl, btw) is that science has more or less ignored the question of causality up until the end of the 1980s.
15/ Econ is actually a notable exception here – one we're quite proud of. But also our credibility revolution is only a child of the 90s. Before, the C-word was likewise a no-no for us. Since then, two important streams of literature on causality have developed independently.
16/ Computer scientists picked up early ideas by Sewall Wright and developed powerful tools for causal inference based on graphical models. Economists were drawn to Don Rubin's potential outcomes framework and developed a huge literature rich with clever empirical applications.
17/ If you ask me, now it's about time to overcome these institutional silos and start to explore what the other has to offer. Guido's paper is a great starting point for this endeavor and I'm very happy that the paper exists.
18/ What is still miss though is a tone that would truly invite collaborations between both disciplines. An honest discussion about the pros and cons of one approach compared to the other is great. And it's of course totally fine if you have your preferred side.
19/ But we should get beyond the methodological battles that we're having for so long already and start talking about where we can join forces. Disciplinary silos are so 1990s, let's finally break them down!
Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Paul Hünermund
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!