Presentation is loading. Please wait.

Presentation is loading. Please wait.

Nick Bloom, Labor Topics, Spring 2010 LABOR TOPICS Nick Bloom Peers at Work.

Similar presentations


Presentation on theme: "Nick Bloom, Labor Topics, Spring 2010 LABOR TOPICS Nick Bloom Peers at Work."— Presentation transcript:

1 Nick Bloom, Labor Topics, Spring 2010 LABOR TOPICS Nick Bloom Peers at Work

2 Nick Bloom, Labor Topics, Spring 2010 Comments on Mas and Moretti (2008) Great paper: innovative way to address an interesting topic – peer effects within workplaces Huge dataset High frequency productivity measurement Interesting setting Points to think about and learn from this I want to discuss: (A) Reflection problem (B) Bootstrap (C) Presentation (graphics and robustness test)

3 Nick Bloom, Labor Topics, Spring 2010 Reflection problems (1/2) One of the central issues in addressing spillovers is distinguishing these from: (A) Unobserved shocks (B) Sorting (selection effects) These issues are often called the “Reflection Problem” after Manski (1993). Ways to deal with this are: (A) Controlling for (instrumenting) unobserved shocks - Do this here using very long-run data, so no SR shocks (B) Having random matching by pairs - Do this here by claiming the shift matching in random (and then test this)

4 Nick Bloom, Labor Topics, Spring 2010 Reflection problems (2/2) (C) Using a distance metric to put more structure on the estimation - They do this with the facing/behind till distinction Overall I think they do a convincing job of addressing the key issue in the “peer effects” and “spillovers” literature

5 Nick Bloom, Labor Topics, Spring 2010 Bootstrap and robustness tests One issue they faced was in generating appropriate standard- errors in the regressions. They had a generated regressor (predicted Θ i ) in the second step – this has error around it so needs its SE adjusted Easiest way to do this is Bootstrap – keep re-drawing (with replacement) from the original data to look at the distribution of the coefficients. - Idea is treats the sample as the population Very computationally intensive (need to re-estimate everything 1000 times over) so they did a more complex Bayesian alternative

6 Nick Bloom, Labor Topics, Spring 2010 Presentation – graphics and robustness They used fantastic graphics to prove their results – if you can always have some graphs of results, particularly with this kind of sharp discountinuity effect They also always tested their key claims – for example that assignment of people to shifts is “random” - If you ever make a claim in a paper always try to test this as much more convincing


Download ppt "Nick Bloom, Labor Topics, Spring 2010 LABOR TOPICS Nick Bloom Peers at Work."

Similar presentations


Ads by Google