Hi all,
Our next virtual meeting will be at 12pm (EST) Wednesday, March 24
(tomorrow), where Avi Feller (UC Berkeley) will present "Varying impacts of
letters of recommendation on college admissions: Approximate balancing
weights for subgroup effects in observational studies." This is joint work
with Eli Ben-Michael and Jesse Rothstein.
*Abstract*:
In a pilot program during the 2016-17 admissions cycle, the University of
California, Berkeley invited many applicants for freshman admission to
submit letters of recommendation. We use this pilot as the basis for an
observational study of the impact of submitting letters of recommendation
on subsequent admission, with the goal of estimating how impacts vary
across pre-defined subgroups. Understanding this variation is challenging
in observational studies, however, because estimated impacts reflect both
actual treatment effect variation and differences in covariate balance
across groups. To address this, we develop balancing weights that directly
optimize for "local balance'' within subgroups while maintaining global
covariate balance between treated and control units. We then show that this
approach has a dual representation as a form of inverse propensity score
weighting with a hierarchical propensity score model. In the UC Berkeley
pilot study, our proposed approach yields excellent local and global
balance, unlike more traditional weighting methods, which fail to balance
covariates within subgroups. We find that the impact of letters of
recommendation increases with the predicted probability of admission, with
mixed evidence of differences for under-represented minority applicants.
*Zoom link*:
https://harvard.zoom.us/j/97787602526?pwd=Uzh3bVVVS0F4TEVYQTJlV3BQNjcydz09
*Schedule of the workshop*:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Looking forward to seeing you all tomorrow!
Best,
Soichiro
--
Soichiro Yamauchi
PhD candidate
Harvard University
URL: https://soichiroy.github.io/
Hi all,
Our next virtual meeting will be at 12pm (EST) Wednesday, March 17, where
Suzanna Linn (Penn State University) presents research on "Causal
Inference in Dynamic Systems." This is joint work with Clay Webb.
*Abstract*:
A causal inference revolution has been under way in political methodology
for the better part of the last decade. Time series analysts have not been
major contributors to this revolution because the tools that have been
developed thus far do not fit our data. Existing methods either require
analyst to pool observations so that analysts can differentiate treatment
and control units, require analysts to identify or develop suitable control
series, or require the analyst to exercise control over treatment in a time
series experiment. Our goal is to identify the assumptions and conditions
required for analysts to make causal inferences with observational time
series data when observations cannot be pooled, control series are
unavailable, and counterfactuals cannot be forecast. We highlight the
critical assumptions for causal identification in structural dynamic
systems: partial equilibrium recursivity and conditional exogeneity. We
discuss the conditions when these assumptions are plausible, outline tests
for conditional exogeneity and structural non-causality, and consider the
potential limitations of the proposed framework. When the proposed
assumptions are met, standard Granger non-causality tests provide a means
for analysts to recover causal estimands.
*Zoom link*:
https://harvard.zoom.us/j/97787602526?pwd=Uzh3bVVVS0F4TEVYQTJlV3BQNjcydz09
*Schedule of the workshop*:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Looking forward to seeing you all on Wednesday!
Best,
Soichiro
--
Soichiro Yamauchi
PhD candidate
Harvard University
URL: https://soichiroy.github.io/
Hi all,
Our next virtual meeting will be at 12pm (EST) Wednesday, March 10, where
Adeline Lo (University of Wisconsin-Madison) presents research on
"Mixed-Membership Stochastic Blockmodels for Bipartite Networks:
Application to Cosponsorship in the US Senate."
*Abstract*:
Many networks in political and social research are naturally bipartite —
with two distinct types of actors (nodes), and edges connecting exclusively
across the actor types. An example of such networks is the one that results
from cosponsorship decisions, in which legislators are connected to the
bills they support. Typically, researchers who wish to study these networks
are forced to "project" or marginalize over one node type, in an effort to
form the kinds of unipartite networks that standard models can handle. This
can result in aggregation bias and loss of relevant information about the
node type that is averaged over. To avoid these issues, we propose an
extension of the mixed-membership stochastic blockmodel that operates
directly on the bipartite network structure, incorporating both node and
dyad-level covariates. We design and implement a fast, scalable stochastic
variational algorithm to obtain estimates of latent variables and
hyper-parameters, and illustrate our model using data from the 107th and
108th sessions of the U.S. Senate. We find that using our model allows us
to uncover interesting groups of bills, explore the effects of individual
ideology on the likelihood of bipartisan cosponsorship, and evaluate the
norms of reciprocity in the legislative context.
*Zoom link*:
https://harvard.zoom.us/j/97787602526?pwd=Uzh3bVVVS0F4TEVYQTJlV3BQNjcydz09
*Schedule of the workshop*:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Looking forward to seeing you all on Wednesday!
Best,
Soichiro
--
Soichiro Yamauchi
PhD candidate
Harvard University
URL: https://soichiroy.github.io/
Hi all,
Our next virtual meeting will be at 12pm (EST) Wednesday, March 3, where
Arthur Yu (University of Chicago) presents research on "Beyond LATE:
Identification of ATEs of Always-Takers and Never-Takers."
*Abstract*:
In the presence of heterogeneous treatment effects, instrumental variable
(IV) estimation point identifies the local average treatment effect, an
average treatment effect (ATE) for compliers. This paper provides a set of
identification results that extrapolate the LATE to the ATEs of
always-takers and never-takers. We first show that the ATEs of
always-takers and never-takers can be written as the weighted average of
marginal treatment effect (MTE) functions. We then demonstrate that, under
additional parametric assumptions on these MTE functions, we can point
identify the ATEs of always-takers and never-takers. In the absence of
these parametric assumptions we can construct bounds for the ATEs of
always-takers and never-takers by linear programming developed in Mogstad
et al. (2018), which performs better than the competing partial
identification strategies. We illustrate the proposed methodology using a
simulation study and an application based on Kern and Hainmueller (2009).
We find that exposure to West German television reduces support for
communism among never-takers. These never-takers, who would not watch West
German TV even if they had improved access, act as-if they anticipate the
effect of watching West German TV and thus opt out of exposure.
*Zoom link*:
https://harvard.zoom.us/j/97787602526?pwd=Uzh3bVVVS0F4TEVYQTJlV3BQNjcydz09
*Schedule of the workshop*:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Looking forward to seeing you all on Wednesday!
Best,
Soichiro
--
Soichiro Yamauchi
PhD candidate
Harvard University
URL: https://soichiroy.github.io/