Syndicate content

Recent comments

  • Reply to: Worm Wars: A Review of the Reanalysis of Miguel and Kremer’s Deworming Study   2 hours 28 min ago

    Hi Heather,

    Nice question. I'd argue that it's still needed for two reasons:

    1. Without satisfying SUTVA, your impact estimates are biased. So, drawing a graph of estimates in your "control" group by (geographic or social) distance, you're getting a better sense of what the true treatment effect is.

    2. To the extent that spillovers across clusters are significant and the unit of intervention/delivery is a cluster, then you can get away with treating less than 100% of the country and save money. This is an extension of the within cluster, herd immunity argument, but it might still matter. We talk about this in our spillovers paper that I cited in the post, at which you may take a closer look...

    Thanks for commenting.

    Berk.

  • Reply to: Worm Wars: A Review of the Reanalysis of Miguel and Kremer’s Deworming Study   7 hours 31 min ago

    Hi Berk: I have a question that has always nagged me about the HKN paper, possibly related to the idea of "too much policy change on the basis of a single paper." Now seems like as good a time as any to ask, though it is a departure from the replication/reanalysis discussion!

    As you note, externalities are a key part of the original analysis and, indeed, the impact claims and cost-effectiveness of the intervention is bolstered by the spillover effects into neighboring clusters. I leave aside any issues of within-cluster spillovers / forms of herd immunity and rather focus on arguments made about the spillovers from treated to comparison clusters.

    When using a study to argue for the scale-up of an intervention, to what extent is it appropriate to rely on estimates that incorporate spillovers to make the case? Presumably scale-up is meant to eliminate untreated clusters, so that there are no more clusters (within, say, national borders) to which to spillover. It would be kind and possibly good public health policy to want border areas to be treated in/directly but I imagine most decision-makers focus most heavily on their own country.

    It is always important to account for externalities, both in estimating impact and understanding the mechanisms by which an intervention has its effects. But, if we take 'scale-up' to essentially mean 'let's eliminate untreated clusters,' I wonder where an analysis of cross-cluster externalities comes in forming evidence about the decision to scale.

    Thanks! And, of course, thanks for your important comments and insights on internal replication / reanalysis. Hopefully other commenters will manage to stay on topic.

  • Reply to: Paying Taxes Because You Want To [Trust Me, You Really Want To]   3 days 2 hours ago

    This is a good point. When I was Secretary of Mayor Offices in Spain in the late 80’s, the Mayors sent letters to local tax payers that did not pay on time before imposing penalties to them, including information about the services the town delivered, the percentage of citizens that had already paid, and the penalty surcharge in case of delay. This system worked quite well and more than 80% of the pending payments were done after the reception of the letter. They combined very efficiently enforcement, reciprocity, and services delivery.

  • Reply to: Blog links Jan 9 2015: Angrist and Niederle on pre-analysis, problems of phase-ins, French-speaking field coordinators needed, and more…   6 days 11 hours ago

    The biggest problem with headlines interpreting the the PLoS ONE study, of course, is that paper *replication* is equivalent to paper *production*. It's hard to argue that, when replicating a given output, it is much easier to replicate using a WYSIWYG system, for which LaTeX is not (unless one uses a front-end such as LyX). But production is a different exercise altogether: it involves thinking of what one wishes to convey. This naturally reduces error production and, I would argue, would not give rise to the sorts of "errors" identified in the study. One may even find the clutter-free form more conducive for focusing on writing!

    Just as important, the interaction of this production process with standard LaTeX applications would often result in automated output that is far more error-free than the replication exercise would entail: tables would be automatically output into TeX-friendly form, with just minor editing for style; text editors would pick out typos in a way that is more easily observed by a writer than by a copyist; and references are automatically cross-checked using BibTeX so any non-matching references are immediately caught.

    One aspect, less highlighted in the study, is that TeX capabilities for collaborative writing is definitely more primitive. Indeed, as a user of both, my decision in favor of TeX for research articles is premised on the greater efficiency with which one can crank out scientific output, whereas when I am crafting a CV or a newsletter, I revert to Word, given the focus on formatting.

  • Reply to: Paying Taxes Because You Want To [Trust Me, You Really Want To]   1 week 3 hours ago
    Another reader -- offline -- highlighted to me the point that improving government services should come first. The L&S article highlights the positive correlation across countries between impressions of government services and tax compliance. So yes, first order are having a functioning government and a reasonable enforcement mechanism. Then, the nudges above.