Friday, February 16, 2024

How to Make Nice Bar Graphs in Stata

Sure, you can just show a table, but nobody will remember that or pay much attention to start. Much better to tell your story with a picture. 

Here are some helpful tips on how: Part 1 and Part 2. Thank you, John Kane for putting this together and to David Mackenzie for finding it. 



Friday, February 9, 2024

To Post-Doc or not to Post-Doc. Is That the Question?

David McKenzie provides useful advice not only on whether or not to do a post-doc but also best practices if you do decide to do one.  

My thoughts: Post-docs vary a lot! Some are really just high-skilled RA positions--which is not necessarily bad if you're learning a lot. Others just give you a lot of free time to pursue your own work--which is not necessarily great if you don't have a clear direction for future projects. My point: There is no clear answer to the question of whether it is good or bad to do a post-doc. It really depends on the particular post-doc position you are considering and the non-post-doc position you are giving up or postponing. 

Good luck, JMCs! 

Tuesday, January 16, 2024

Potentially My Most Useful Post of the Year? Decade?: AI for Economists, Prompts and Resources

We hear a lot about the dangers of AI (and these are real!), but we do not hear enough about how useful LLMs can be. Check out these prompts and other resources for use with LLMs made available by Jesse Lastunen. And do share in the comments if you discover any additional useful prompts. 


H/T: Marginal Revolution 

Friday, January 5, 2024

Cool Data Alert: Child Penalties Across Countries

The website describes it like this: "The Child Penalty Atlas maps how the arrival of a child affects the careers of men and women across the globe." I recommend playing with the map a bit. You can see for most countries in the world, the child penalty in employment, the gender gap in employment, and the fraction of the gender employment gap explained by the child penalty.

How many new papers idea using these data can you come up with? 1-2-3, go! (UConn students, feel free to come discuss your ideas with me). 

Here is the website: https://childpenaltyatlas.org/. 

Wednesday, August 2, 2023

Cool Data Alert: Aging

At this summer's NBER SI Aging, Kosali Simon introduced new data from all over the world that can be used to study the economics of aging. She just posted her slides online

Thank you, Kosali. UConn PhD students, start exploring and let me know if anything inspires you. 

Sunday, July 16, 2023

How to be a Good Discussant

Your third year paper was accepted for presentation at a conference. Congratulations! But now you've been asked to discuss someone else's paper at that conference. What should you say during those 5-10 minutes? Every single thought you have on the paper? NO!!! 

See Chris Blattman's The Discussant's Art, an oldie but goodie. 

The most important thing to keep in mind: Your discussion is mostly for the conference participants, not the authors. You want to help the audience better understand and appreciate the presentation they heard. Speak to them! 

Of course, you've read the paper and have lots of ideas. Absolutely share these ideas with the author! They can be very valuable and appreciated, even the "you have a typo on p. 18"-type comments. But do this one-on-one with the author. Ideally, meet for coffee during the conference. Otherwise, send the detailed comments over email after the conference. 

But the discussion is mostly for the audience. Keep it short but profound. :) 

---

Chris Blattman's Tips Copy-Pasted here for convenience: 

1. Start by telling people why they should care. It is seldom obvious. What’s the big question, and what’s at stake if the paper gets it right or wrong?

2. Then summarize the paper. Break it down differently than the presenter. Pretend you are explaining it to your grandmother. Or, rather, your adult-attention-deficit-disorder grandmother. Keep it short.

3. Say more with less. Mathematically, everything you say after your best point lowers the average quality of your comment. Pick your three best points, say them briefly, then stop talking.

4. Now, say even less. Those three comments? Write out, in bullets, exactly what you plan to say. Now cross out half. What you think will take eight minutes will take fifteen. Bring it back to eight.

5. Be constructive. A colleague once said to me: “I like it when people find problems with my paper, but I like solutions more.” Finding solutions makes you think (and displays it too).

6. Don’t discuss the small stuff. Write your little comments down, and later give them to the author. Don’t bore the audience with footnotes and trivia.

7. Feel free to entertain. A discussant need not merely list ideas. You can weave in an anecdote, or frame a point with a story. At least speak from a personal point of view, not a monotone benevolent overlord.

8. Have fun, don’t make fun. If you use humor, let it not be at the expense of anyone but yourself.

9. Spell it out for us. Tell us why your comments matter. Say precisely what we learn.

10. Aim for profound. The best discussants rotate my brain 90 degrees. They reframe the problem, or propose a novel idea. I can’t tell you how to be deep. I seldom succeed myself. For me, a few things usually help. I read the paper, walk away for a day or two, then return. I ask myself questions: Do I think about a big question differently now? What convinces me, and what would convince me more? Where should the field be going?

More on Shift-Share Instruments via Borusyak, Hull and Jaravel

It is tricky to keep up with the econometrics literature on shift-share instruments. But it's important. Here is a review paper written by Borusyak, Hull and Jaravel. 

David McKenzie highlights the following: “Identification in the general case, for arbitrary formulas and designs, follows from simple adjustments based on the expected instrument: the average value of the formula across counterfactual sets of shocks, drawn from the specified assignment process. Specifically, OVB is avoided by either adding the expected instrument as a control or by using a recentered instrument which subtracts the expected instrument from the original formula. Controlling for or recentering by the expected instrument is generally necessary for identification with formula instruments, absent auxiliary assumptions on the exogeneity of shock exposure.”