, 15 tweets, 5 min read Read on Twitter
1/ Let's go for another round of #EconBookClub! By now, we've reached the most important chapter 4 of Guido Imbens' new paper in which he contrasts DAGs with the potential outcomes (PO) framework (link: arxiv.org/pdf/1907.07271…) #BookofWhy #Causality #Econometrics
2/ This chapter has six subsections and because my previous threads were already quite long, I decided to go thorugh them one-by-one in the following days.
3/ Section 4.1 is about the role of manipulability in causal inference. Something that has been discussed here a lot recently by the epidemiologists and also @yudapearl commented on it a while ago.
4/ To be honest, this debate always struck me as quite "philosophical", with very little practical relevance. Because, in my mind, non-manipulable (by humans) variables like gender, race or obeseity can of course be causes.
5/ We can also not manipulate the occurence of solar flares (and I don't know how helpful the thought experiment analogy is here either). Still, we don't doubt that they exert a causal effect on short-wave radio communication on earth. Don't we?
6/ However, I can very well see why the manipulability discussion is so central to PO, which defines causal effects solely in relation to an either hypothetical or actual experiment.
7/ If such an experiment weren't even theoretically conceivable, then PO practitioners are of course FUQ'd (i.e., they deal with a "fundamentally unidentified question") – as Angrist and Pischke would put it.
8/ Here I strongly disagree with Imbens when he stipulates that PO connects well with economic thinking in this regard. Because I don't think we have a problem with including a gender or race dummy in our regressions
9/ or structural / theoretical models to interpret their "ceteris paribus" effect, even though it's debatable whether these variables are directly manipulable.
10/ As long as the causal mechanisms they exert can be thought to be reasonably autonomous from other influence factors in our models, we don't have any troubles including and analyzing such variables. economics.soton.ac.uk/staff/aldrich/…
11/ But here lies the big difference. In structural econometrics as well as DAGs, we have a model to guide our thinking, in which interventions such as the ones expressed by the do-operator are all well defined.
12/ It is the model that defines the "treatment" and that tells us which other variables are confounders or mediators. PO doesn't have that. Instead of a formal model, it relies exclusively on an "RCT as gold standard" metaphor for defining causal effects.
13/ In this case, if the gold standard you're trying to adhere to isn't even conceivable – because the corresponding RCT is unimaginable – then of course you run into conceptual troubles.
14/ To summarize: The PO framework forces a debate about manipulability that feels rather artificial in economics. Once more, PO's origin in biometrics shows here, I'd say.
15/ Not manipulability but autonomy is the decisive feature for something to qualify as a cause in econometrics (see the Aldrich paper I linked to in tweet 10). And this is fully in line with the DAG framework as well as the more philosophical accounts of causal models.
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!