econtwitter.net is one of the many independent Mastodon servers you can use to participate in the fediverse.
A Mastodon instance for Economists.

Administered by:

Server stats:

163
active users

Peter Hull

@dlmillimet @nic @seema_econ @ben_golub @paulgp I disagree. There's a meaningful econometric and practical difference between design based identification strategies that can specify shock counterfactuals and model based ID strategies that can't

@instrumenthull @dlmillimet @nic @seema_econ @ben_golub Might be a bad terminology choice though (although understandable given path dependence)

@instrumenthull @dlmillimet @nic @seema_econ @ben_golub @paulgp Right - one can say RDD is a quasi-experiment in a sense distinct from observational data writ large.

@arindube @dlmillimet @nic @seema_econ @ben_golub @paulgp RDD is an interesting one. The usual formalization with continuous potential outcome CEFs is not a "natural experiment" as DiNardo or we would say -- it's a "quasi experiment," leveraging a model on unobservables. But then there's a design-based "local randomization" view where we think of an RCT in the bandwidth and can e.g. permute shocks to the left and right. I think this distinction + extensions are currently very underexplored!

@instrumenthull @arindube @dlmillimet @nic @seema_econ @ben_golub same with diff in diff (staggered random timing rollouts)!

@paulgp @instrumenthull @arindube @dlmillimet @nic @seema_econ @ben_golub This is interesting. Would one of you please suggest a recent article that discusses design- versus model-based identification strategies? Probably something of yours that I should have read but have not. Thanks.

@BlaneDavidLewis @paulgp @arindube @dlmillimet @nic @seema_econ @ben_golub I not so humbly recommend

dropbox.com/s/eg91bo1ghik31gi/
Which gives a very general approach to design-based identification, and

dropbox.com/s/pktwrdvppcddqtw/
Which discussed how design- and model-based strategies lead to different negative weighting problems in OLS specifications