@DrJBhattacharya
Closing parks was one of the worst things. I’have never understood how this was rationalized (getting fresh air is supposed to be anti-covid measure). Or maybe they thought Covid was super airborne.
A week ago I finished teaching and grading my new course on (applied) causal inference with ML and AI tools. As prizes to best performing grad and undergrad students, I have given out a dozen of Pearl's Causality book.
On March 5th, it became illegal for Russians to protest war, and such protests could be equated to “radicalism/terrorism” (with a jail sentence). Using this opportunity, I would like to register my protest against this war. Please stop the war. It is never too late to stop a war
Recommendations for future students in econometrics and statistics. For programming, invest in both Python and R. For math, probability or stochastic processes, or both (I'd recommend engineering dept over math dept courses; with the exception: .)
recent trends observed: people apply for assistant professor jobs with 60 publications, after post-doc, for grad school with 4-10 publications, after pre-doc... very different from my time (well, I am getting old, I guess), when you applied to either position with 0 publications
“All happy statisticians are happy in their own way; but all the unhappy ones are all alike — they all do causal inference with observational data”. L. Tolstoy in “Anna Karenina” (found this quote on the internet, so it must be authentic)
We have released a revised version of our paper titled "Long Story Short: Omitted Variable Bias in Causal Learning." You can find it at . In the paper, we (
@analisereal
;
@syrgkanis
) derive the OVB formula for nonlinear models and develop sensitivity
Causal DAGs are an incredibly useful tool for empirical work in economics. They provide a clear way to explain your assumptions for causal effect identification as well as isolate threats to identification. An example of using Dagitty for 401(K) analysis.
Evelyn Kitagawa of the University of Chicago developed the Oaxaca-Blinder decomposition analysis in 1956, roughly 20 years before Oaxaca and Blinder did. I didn't know this until a student with a background in sociology pointed this out to me. Wikipedia does not elaborate much on
Over the Christmas break, I have gone and “re-rediscovered” with paper and pencil the backdoor, frontdoor, d separations etc. DAGs are thrilling and it is one huge development that econometrics definitely missed out on. Never too late to catch up though.
I was somewhat skeptical of causal diagrams/ causal DAGs (even though I started to use them in econometrics teaching in 2015), but now.... I even think of RCTs in terms of DAGs.
Just came across this beautiful paper on NNs for causal inference under un-confoundedness:
Just writing this or myself to put it on the reading list for a new course in the Spring of 2021 (if we make it :-) )
Happy to have our paper with Paul and Hiro appear in JoE. Back in May we found that masks, lockdown policies contributed to mitigating Covid-19 (contrary to popular views that mitigation happened all due to private response) Policies also matter, and DAGs were super-helpful.😀
After 6 or 7 years of work, and a couple of failed attempts at “general interest” journals, I am happy to see our paper on shape-enforcing operators finally appear in Journal of Machine Learning Research: 😀
Who invented OLS: Legendre or Gauss? Legendre published it first in 1805. Gauss published it later, though kinda used earlier. Pythagoras of Samos invented ANOVA and OLS for the case of 2 observations (n=2), about 2300 years before that.
Causality and causal DAGs... I need to explain this to freshmen undergraduates in 35-40 minutes (hopefully, convincing them to do our B.S. in CS, Economics, and Data Science). Could anyone recommend high-quality, self-contained material for such a lecture?
@yudapearl
@PHuenermund
Came across this paper that does panel regressions of Y (Covid growth rates) ~ P (adopted polices) + fixed effects. The elementary modeling mistake is not to control for past growth rates and new recent cases. Call these variables I (information variables). 1/n
(Russian-speaking) economists wrote the letter to stop the senseless, unjust war. (This is of little help, of course, but we try.) The only thing that this war does is to multiply the horror and sorrow for the people of both Ukraine and Russia.
I've just finished watching the new documentary "Don't look up" on Netflix... It is amazing and psychedelic. Easily the film of the year. Plus that Russian (Leonardo DiCaprio aka Lev Smirnoff) really did an amazing job. May many more Oscars pour onto him in the New Year.
Are there good econometrics packages in Python for doing basic stuff (like panel data, clustered standard errors etc)? I am thinking of switching part of my teaching to Python.
Could investing in early childhood development be one of keys to eliminating systemic racism? The data show striking differences in early childhood development. Figures below on test scores are from our JASA paper, reanalyzing Fryer and Levitt (AER). Left: 1y; Right: 7y.
Simple, sharp, general, and flexible omitted variable bias analysis in causal machine learning, from Victor Chernozhukov, Carlos Cinelli, Whitney Newey, Amit Sharma, and Vasilis Syrgkanis
Why do so many people seem to think that panel data models are difference-in-difference models? (The research design in (possibly dynamic) panel data model is conditional random assignement of policy variable given state variables and latent confounders ("fixed effects").
Putin claims to "protect" Russian speakers, but huge proportion of refugees are Russian speakers; Russian speaking children get killed in "precision strikes"; Russian speaking people fight Russian troops near Kharkiv and Kyiv; and anti-war protesting Russian speakers get jailed.
A small team of econometricians initiated the launch of Economics section of ArXiv in 2017 (which now has 3 subbranches, Theory, Econometrics, General). The growth in the number of submissions has been fantastic over the 3 years.
@JoergStoye
I'm teaching a "data science for economists" course this semester.
If you're interested in learning more about
#rstats
, Git(Hub), programming, databases, cloud computation, ML, etc., I'll be making all of my course material publicly available here:
Since the Maidan of 2014, I have been using Ukraine-themed DAGs in my teaching, as a tribute to my friends, family in Ukraine, the town where I grew up (Mukachevo), and the hope to live in a civilized country. Plus, the DAG below is really awesome science by Philip Wright 1928.
I just came across this nice quote by Democritus c400 BC: “I prefer to discover a causality rather than become a machine learner”... Evidently, some people were far ahead of their time. What do you say, a random Twitted wanderer, so I can quote you in my lecture note?
I came across this very nice blogpost by
@PHuenermund
on value of C-DAGs in econometric research (One correction: causal DAGs were native to econometrics, Wright, 1928, Wold 1954, Tinbergen, 1940). Time to go back to the roots?)
I hope Putin has courage to step down as President. The war is a tragic mistake that took many innocent lives. It puts Russia in a very bad position for many generations. It destroys relations with Ukraine, connected to Russia with millions of family ties. Unthinkable tragedy.
For teaching purposes, I am looking for empirical applications of neural embeddings in economics, primarily aiming at structural causal inference. We've recently produced one here, but I am curious to learn of more examples.
I am beginning to read JMLR paper by Halbert White and Karim Chalak on limitations of DAGs/SCMs in dealing optimization, equilibrium, learning, which are of "central interest to economists and econometricians". Comments/suggestions/joint reading is welcome.
@yudapearl
This is how the communist system started in the former Russian empire — by making one party dominate the politics and eradicating basic freedom of speech. The communist system resulted in millions of deaths across the globe and immense suffering of many people. “Quality of
I’ve been reading some neo-classical econometrics papers from 80-90s. Hansen-Singleton (Econometrica)
And Berry-Levinnshon-Pakes (Econometrica). They start out with stochastic economic models (to define structures/ Causal models). 1/n
Super-nice paper by M. Igami "AI as Structural Estimation.." appeared in Econometrics Journal recently. Econometrics Journal now enjoys a great stream of papers, thank to revamping editorial policy (shorter papers, "minor revision" only policy, very fast turn around times).
Students often "accept" AI/ML (with the zeal of the convert, forgetting everything from econometrics), while some professors just "reject" it as irrelevant. Here's the third way:
I was prepping lecture notes and tried to look up a paper of mine; found some really ugly mathematics, as usual, but (hopefully?) useful. In the end, I summarized this in the lecture note as "under regularity conditions".
@akbarpour_
Another thing I do at MIT: grades in Econometrics I are based 100% on HW (as an option); if you submit all the HW, you are guaranteed B+ (and with a little extra effort more). This takes off stress and allows students to focus on learning (based on students' responses).
Masks were stupidly politicized during Covid era. Japan did not have any business lockdowns, and mostly relied only on using masks and public health recommendations, and did spectacularly better than U.S. or Sweden, for example. The graph shows deaths per 100K.
Russian economists published a letter in support of fair treatment of Alexey Navalny and in support of free elections in Russia. The rationale given is "sustanined economic growth", but really I think it is mainly about basic decency.
Question: how does one create a DAG to study equilibria? An answer was given by economist Philip Wright, 1928, by simply introducing the supply and demand functions as *latent* nodes, which together determine observed equilibrium price and output. No wonder QJE rejected this. 😀
Ok. Im doing C-DAG lectures notes and R cloud notebooks for teaching (one with “dagitty” and one with “dosearch” packages). Are there other cool packages (R or Python) I should look at? Also, anything that auto ports a graph G in dagitty format to TikZ/latex format?
The tyranny of the null hypothesis. Imagine a decision maker, where false acceptance of null is very costly, yet he blindly uses decision based on P< a=.01 (or .0...1). This implies that the null has low acceptance cost. None of this impasse arises in Bayesian decision framework.
Hello friends, love you all. But why so many ppl on Twitter appear so intense, uneasy? Is it b/c "hello", "friends", "respectfully", "thank you", "you are welcome", are counted towards the tweet character count? Maybe they shouln't, and this platform would be better off.
With the amazing Denis Chetverikov and Yuta Koike, we
just finished up a new paper that derives a high-dimensional CLT with optimal dependence on log d (log of dimension), envelope B, and (up to log n) optimal dependence on n (sample size).
Replying to a probable bot: I was protesting the US wars as well! I am a principled pacifist. I believe that all people are brothers and sisters and the life of each person is sacred. Never too late to stop any war!
I’ve been feeling lazy and unproductive as far as I remember myself. Now that I bought this book, all problems solved. In the nutshell: just work 8 hours a day. (Don’t know if this includes weekends).😀
Wayne Shafer, an amazing educator and superb researcher passed away… At UIUC he taught a really amazing mathematical economics class, covering general equilibrium, cooperative game theory and other topics, that convinced me to pursue graduate work in economics. He also wrote a
The unconfoundedness assumption: (informally) "all the variables affecting both the treatment T and the outcome Y are observed"
Magic: "the use of means (such as charms or spells) believed to have supernatural power over natural forces." Meriam-Webster
In an extraordinary move, the three new arrivals emerged from their Soyuz capsule after docking with the space station wearing bright yellow jumpsuits with blue stripes, instead of the standard-issue blue uniform
@daniela_witten
When I was taking a walk around streets of Cambridge university, UK, a fancy convertible car stopped by, and students in the car asked if I could sell them some pot. I infer that the local supply comes from Poles, Russians etc, and I must look like one of them😀
Our review for "The Matrix Resurrections":
"The actors’ sincerity and effortlessly synced performances have always been this series’ greatest special effects, and watching them slip back into their old roles is a pleasure," writes our critic.
The "good" news outweigh the bad: >2000 students/faculty from Saint-Petersburg University sign the anti-war petition (despite this act now bing punishable). This outweighs the letter of 600 produced in support of the war (with all the admin encouragement).
Hey, friends, how do you make sure there is no unobserved confounder/omitted variable in any of your causal analyses? The same questions applies to experiments with some Y's missing? 😀
I miss having World Congress of Econometrics Society in Milan 2020: hunting for the next good session, meeting people you've never met before, chatting, and having drinks with friends. It seems impossible to replicate the live experience via Zoom, esp. big conferences. Any +s?
@phl43
Philippe, I wanted to thank you very much for your critique. Hiro, Paul, and I are taking a deep dive and will provide a detailed response in a format of a paper. The estimated amount of work is at least one month.
Recent Impact factors for some journals that publish work by econometricians:
JRSS(b) 5; Review of Econ Studies 4.890
Econometrica 4.2; JASA 3.9; AoS 2.99; Econometrics Journal, 2.19; Biometrika 1.67; Journal of Econometrics, 1.57.
@JaapAbbring
I teach 14.382 (Econometrics I) using chalk and blackboard. I have to teach again in the spring via Zoom (most likely), and would like to do avoid using slides. Any recommendations would be helpful.
I didn't have a chance to contribute to wording (no option was given), but was happy to sign it. I would like to see the return of election at town and city levels, competitive parliamentary and presidential elections. Election are our true Russian tradition (Novgorod & South).
@k_sonin
Kremlin should set an example and start with self. Siberian Shaman Gabyshev said a few years ago… “God told me that Putin is not human but a demon and has ordered me to drive him out.”
A Russian shaman has been just sentenced to a psych ward, with the diagnosis “overestimation of importance of his own persona”. He walked across Siberia 20km a day to talk senses to Kremlin folks.
Sad statistics from yesterday: nearly 3K daily deaths from Covid-19, with 1.3K coming from Florida alone. This exceeds the peak for daily deaths in the first wave of pandemic in 2020.
Finally: Large Scale RCT on Masks in Bangladesh. A modest "nudge" increased community use of masks in treated villages: 13% to 42%. For surgical masks: symptomatic seroprevalence dropped 35% in 60+ group and 13.5% overall (for cloth masks 8.5%)
@abuluck
However, the most prolific data collector is the Count of Monte Carlo. He is so famous that no-one bothers to cite particular papers. Everyone just cites her/his name only.