Syndicate content

Recent comments

  • Reply to: Can Wealth Taxation Work in Developing Countries? Guest post by Juliana Londoño-Vélez   13 hours 54 min ago

    Dear Julians Congratualutaions on your research work. It is relevant that you share your findings with Colombian Government, particularly with Alberto Carrasquilla, Minister of Finance so that he reflects upon the fact that greater enforcement will improve wealth tax collection as you claim: "With better enforcement, wealth taxes can complement progressive income taxes to reinforce progressivity and address inequality in contexts where elites are difficult to tax."

  • Reply to: 8 lessons on how to influence policy with evidence – from Oxfam’s experience   6 days 5 hours ago

    Wonderful especially point 8

  • Reply to: 8 lessons on how to influence policy with evidence – from Oxfam’s experience   1 week 11 hours ago

    Présentation of the évidence must align with the political interest of the policy makers to have a positive social impact.

  • Reply to: Tools of the Trade: a joint test of orthogonality when testing for balance   1 week 12 hours ago

    Hi David -

    This is interesting. Thanks for sharing. It occurs to me that the approach you describe has some serious limitations. For one, a statistically significant overall F test from the type of model you describe would provide evidence of covariate imbalance, but would not directly tell you whether or not that covariate imbalance would generate bias in the association you ultimately measure between T and Y. If some Xs are associated with T but have no association with Y, they would generate no bias. Moreover, it is possible that the bias generated by one set of Xs could be offset by bias in the other direction generated by another set of Xs. This concern, though, has a good remedy: after running a simple model of T predicting Y, add subsets of Xs to that model based upon post-hoc analyses following a statistically significant overall F test.

    Perhaps a bigger concern involves ways in which this approach could be abused, all of which I think relate to Type 2 error. If you were an naughty investigator who wanted to avoid finding statistically significant evidence of covariate imbalance, there are various strategies you might employ. One would be to have a small sample, and thus little statistical power to detect covariate imbalance. That strategy, though, would be self-defeating when it comes to what is presumably the main goal of the study: to measure the effect of T on Y. Another strategy that would not be self-defeating, however, would be to pack the regression model with a whole bunch of garbage Xs -- covariates that are poorly measured, or for which there is no good reason to think they have anything whatsoever to do with either T, or Y, or both. Given a set of covariates for which there truly is imbalance, the overall F test will detect that if those are the only covariates in the model; will have a good chance of detecting it if there are some but not a ton of other covariates in the model; and will have little chance of detecting it if the model is packed with a bunch of other covariates.

    All of which is to say, I guess, that I don't think the overall F test or any other statistical approach can fully adjudicate these issues in the absence of honesty, integrity, and good judgment on the part of the investigator.

    - Bart

  • Reply to: What’s the latest in development economics research? Microsummaries of 150+ papers from NEUDC 2018   1 week 15 hours ago
    Thank you for catching this! I'd propose that the abstract is ambiguous: It first refers to "the dismantling of a...policy reform" and then to the consumer welfare impact of the "policy reform," so it's not obvious whether the latter policy reform refers to the dismantling or to the original policy reform. 

    But the paper makes clear that you are correct! "I find that dismantling the policy increased consumer welfare by 1.62 percent." Thank you for catching that. I'm updating the post.