Hi all,
Our next virtual meeting will be Wednesday April 1, where we will hear Soichiro
Yamauchi present research on “*Difference-in-Differences for Ordinal
Outcomes: Application to the Effect of Mass Shootings on Attitudes towards
Gun Control*”.
*Abstract*: The difference-in-differences (DID) design is widely used in
observational studies to estimate the causal effect of a treatment when
repeated observations over time are available. Yet, almost all existing
methods assume linearity in the potential outcome (parallel trends
assumption) and target the additive effect. In social science research,
however, many outcomes of interest are measured on an ordinal scale. This
makes the linearity assumption inappropriate because the difference between
two ordinal potential outcomes is not well defined. In this paper, I
propose a method to draw causal inferences for ordinal outcomes under the
DID design. Unlike existing methods, the proposed method utilizes the
latent variable framework to handle the non-numeric nature of the outcome,
enabling identification and estimation of causal effects based on the
assumption on the quantile of the latent continuous variable. The paper
also proposes an equivalence-based test to assess the plausibility of the
key identification assumption when additional pre-treatment periods are
available. The proposed method is applied to a study estimating the causal
effect of mass shootings on the public’s support for gun control. I find
that the effect is concentrated on left-leaning respondents who experienced
the shooting for the first time in more than a decade. A copy of the paper
can be found here <https://soichiroy.github.io/files/papers/ordinal_did.pdf>
.
*Zoom link*: https://harvard.zoom.us/j/987462892
*When*: Wednesday, April 1 at 12noon - 1:30pm.
Best,
Georgie
Hi all,
I hope everyone is safe and healthy! Applied Stats will continue for the
remainder of the semester via zoom.
Our first virtual meeting will be *Wednesday March 25*, where we will
hear Weihua
An present research on “*Causal Inference with Networked Treatment
Diffusion*”.
*Abstract:* Causal inference under treatment interference (i.e., one unit’s
potential outcomes depend on other units’ treatment) is a challenging but
important problem. Past studies usually make strong assumptions on the
structure of treatment interference. In this study, I will highlight the
importance of collecting data on actual treatment diffusion in order to
more accurately measure treatment interference. Furthermore, I will show
that with accurate measures of treatment interference, one can identify and
estimate a series of causal effects that are previously unavailable,
including the direct treatment effect, the treatment interference effect,
and the treatment effect on interference. Last, I will use exponential
random graph models to model treatment diffusion networks in order to
reveal covariates and network processes that significantly correlate with
treatment diffusion. I will illustrate the ideas and methods through a case
study of a smoking prevention intervention conducted in six middle schools
in China. The findings provide an empirical basis to evaluate previous
assumptions on the structure of treatment interference, are informative for
imputing treatment diffusion when it is unavailable, and help improve
designs of future interventions that aim to optimize treatment diffusion.
*Zoom link: * https://harvard.zoom.us/j/987462892
*When: *Wednesday, March 25 at 12noon - 1:30pm.
I hope many of you can join! Lunch will sadly not be provided.
Best,
Georgie
Hi all,
Our next meeting will be *Wednesday March 11*, where Reagan Mozer will
present research on* "**New approaches for scaling-up human coding efforts
in randomized trials with text-based outcomes*".
*Abstract: *Text data have a long history in social science and education
research. However, these data are notoriously high-dimensional and
characterized by many nuances of language that lack plausible statistical
models. As a result, analysis of text data typically involves intensive
human coding tasks where particular constructs or features of the text are
first defined, and then a collection of documents are inspected and coded
for the presence or absence of these constructs. While this process may be
feasible in studies with smaller sample sizes, the time and resources
required to train and employ multiple human coders frequently poses a
challenge for large-scale efforts. In this talk, I will consider how to
reliably and efficiently extract meaningful constructs from text documents
for the purposes of drawing causal inferences, with an emphasis on the
context of experimental studies where some outcomes of interest are
features of text generated by the trial’s participants. In particular, I
will describe an approach that combines machine learning and survey
sampling methods to streamline the process of hand-coding in a way that is
automatically verified and validated. To illustrate the proposed methods, I
will present results from a pilot analysis of a randomized trial that used
student-generated essays to evaluate the impact of an educational
intervention on students’ writing abilities.
*Where:* CGIS Knafel Building, Room K354 (see this link
<https://map.harvard.edu/?bld=04471&level=9> for directions).
*When: *Wednesday, March 11 at 12noon - 1:30pm.
All are welcome and lunch will be provided.
Best,
Georgie
Hi all,
Our next meeting will be *Wednesday March 4*, where Professor James Robins
will present research on* “**Causal Mediation Analysis and The Philosophy
of Causation**”.*
*Abstract:* I introduce an interventionist view of mediation analysis
that, unlike standard non-interventionist mediation analysis of Pearl and
Vanderweele, (i) does not assume that the mediator has well defined causal
effects and/or counterfactuals , (ii) preserves the dictum “ no causation
without manipulation” (iii) renders questions of mediation scientific,
because eventually testable, (iv) replaces the current complex definitions
of path specific effects as nested counterfactuals with easily understood
definitions in terms of concrete experimental interventions, (v)
facilitates communication with subject matter experts (vi) makes concrete
the concept of a “recanting witness” and makes testable the hypothesis that
a particular variable is a ‘recanting witness’, (vii) when identified from
data, the identifying formulae under an interventionist view and the
non-interventionist view are identical; however the effect being identified
differs. This talk is based on a joint paper with Ilya Shpitser and Thomas
Richardson that generalizes an earlier paper by myself and Thomas
Richardson.
*Where:* CGIS Knafel Building, Room K354 (see this link
<https://map.harvard.edu/?bld=04471&level=9> for directions).
*When: *Wednesday, March 4 at 12noon - 1:30pm.
All are welcome and lunch will be provided.
Best,
Georgie