At the City University of New York Graduate Center, 365 Fifth Avenue (between 34th and 35th street), room 6002. The topic: causality and statistical learning. Announcement is here (scroll down). It says that if you would like to attend any event, please respond by emailing datamining@gc.cuny.edu

I’m also giving a shorter talk on the same topic in the Sustainable Development Seminar Series 4pm Monday 29 Apr in room 407 International Affairs Bldg (at 118th St. and Amsterdam Ave.).

It’s just a coincidence that I’m giving the same talk twice. I was asked at different times to speak for these groups. When someone asks me to speak, I let them pick from recent talks on the list.

Every time I see these slides, I get confused by the second slide on age and happiness (slide 12). The graph on slide 11 (the Economist graph) measures happiness; slide 12 measures probability of depression. Doesn’t it make complete sense that the happiness should show the inverse relationship as depression does? Those two things look consistent to me! It’s only the GSS slide (13) that differs. What have I missed?

@john see link for some of the story http://andrewgelman.com/2010/12/26/age_and_happine/

Any chance either of these will be videotaped?

On similar subjects, and not directly conflicting in time (although Friday will hurt), Bernhard Schölkopf (MPI-IS) is talking at the Courant Institute at NYU on Thursday (tomorrow, April 25) and Friday (April 26):

THURSDAY, APRIL 25, 2013

THE XXVIIIth COURANT LECTURE:

5:00 P.M., WWH 109

Title: Statistical and Casual Learning

Bernhard Scholkopf, Max Planck Institute

FRIDAY, APRIL 26, 2013

THE XXVIIIth COURANT LECTURE:

10:00 A.M., WWH 1302

Title: Inference of cause and effect

Bernhard Scholkopf, Max Planck Institute

Abstracts? Papers?

I’m interested in the statement that “each causal question requires its own data analysis” which is on the slides from this talk.

I’m tasked with analyzing a rather big observational data set on sleep quality, sleepiness and work environment. Coming from a machine learning and computer scientist background I have some reading up to do on how to analyze observational data (I’ve solely worked with experimental data before). But I can’t get my head around what point prof. Gelman is trying to make regarding causal questions and how each requires its own analysis (other than that it’s important). I tried to Google it (ofc, haha) but to no avail.

Anyone?

/David

David:

The advice, “each causal question requires its own data analysis,” comes from me and is not published anywhere as such. But it is implicit in standard presentations of causal inference in statistics and econometrics, in that methods for causal analysis for observational data are typically presented in the context of there being some particular causal quantity being identify. The idea is that if you have, say, three different causal effects you’re trying to estimate, you identify them one at a time, for each constructing an observational study using what data you have available. Really, my advice here is just an elaboration of what is implied in standard textbook presentations. The reason I give the advice is that, in practice, people often attempt to run one single regression from observational data and then just read off a bunch of coefficients and give them direct causal interpretations. That typically won’t make much sense.