Dear Applied Statistics Workshop Community,
Our next meeting will be on October 25 (12:00 EST). Melody Huang presents
"Towards Credible Causal Inference under Real-World Complications:
Sensitivity Analysis for Generalizability"
<When>
October 25, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up inside CGIS K354.
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Randomized controlled trials (RCT’s) allow researchers to estimate causal
effects in an experimental sample with minimal identifying assumptions.
However, to generalize or transport a causal effect from an RCT to a target
population, researchers must adjust for a set of treatment effect
moderators. In practice, it is impossible to know whether the set of
moderators has been properly accounted for. In the following talk, I
propose a two parameter sensitivity analysis for generalizing or
transporting experimental results using weighted estimators. The
contributions in the paper are two-fold. First, I show that the sensitivity
parameters are scale-invariant and standardized. Unlike existing
sensitivity analyses in external validity, the proposed framework allows
researchers to simultaneously account for the bias in their estimates from
omitting a moderator, as well as potential changes to their inference.
Second, I propose several tools researchers can use to perform sensitivity
analysis: (1) graphical and numerical summaries for researchers to assess
how robust an estimated effect is to changes in magnitude as well as
statistical significance; (2) a formal benchmarking approach for
researchers to estimate potential sensitivity parameter values using
existing data; and (3) an extreme scenario analysis. While sensitivity
tools for routine reporting have been introduced for sensitivity frameworks
for outcome modeling approaches, these tools do not yet exist for weighted
estimators. Thus, the talk introduces a collection of methods that provide
much needed interpretability to sensitivity analyses, and a framework for
researchers to transparently and quantitatively argue about the robustness
in their estimated effects.
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Our next meeting will be on October 18 (12:00 EST). Dae Woong Ham presents
"Design-Based Confidence Sequences: A General Approach to Risk Mitigation
in Online Experimentation."
<When>
October 18, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up inside CGIS K354.
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Randomized experiments have become the standard method for companies to
evaluate the performance of new products or services. In addition to
augmenting managers' decision-making, experimentation mitigates risk by
limiting the proportion of customers exposed to innovation. Since many
experiments are on customers arriving sequentially, a potential solution is
to allow managers to ``peek'' at the results when new data becomes
available and stop the test if the results are statistically significant.
Unfortunately, peeking invalidates the statistical guarantees for standard
statistical analysis and leads to uncontrolled type-1 error. Our paper
provides valid design-based confidence sequences, sequences of confidence
intervals with uniform type-1 error guarantees over time for various
sequential experiments in an assumption-light manner. In particular, we
focus on finite-sample estimands defined on the study participants as a
direct measure of the incurred risks by companies. Our proposed confidence
sequences are valid for a large class of experiments, including multi-arm
bandits, time series, and panel experiments. We further provide a variance
reduction technique incorporating modeling assumptions and covariates.
Finally, we demonstrate the effectiveness of our proposed approach through
a simulation study and three real-world applications from Netflix. Our
results show that by using our confidence sequence, harmful experiments
could be stopped after only observing a handful of units; for instance, an
experiment that Netflix ran on its sign-up page on 30,000 potential
customers would have been stopped by our method on the first day before 100
observations.
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Our next meeting will be on October 11 (12:00 EST). Soichiro Yamauchi
presents "Statistical Analysis with Machine Learning Predicted Variables."
<When>
October 11, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up inside CGIS K354.
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Scholars in the social sciences are increasingly relying on machine
learning (ML) techniques to construct data from large corpora of text and
images. The ML-generated variables are subsequently utilized in statistical
analysis to address substantive questions through regression and hypothesis
testing. However, this approach can introduce substantial bias and lead to
incorrect inferences due to prediction errors during the machine learning
stage. In this paper, we present an approach that incorporates ML-generated
variables into regression analysis while ensuring consistency and
asymptotic normality. The proposed approach leverages a small-scale
human-coded sample to capture the bias in the naive estimator, without the
need for strict assumptions about the structure of prediction errors.
Furthermore, we have developed diagnostic tools to assess whether
additional human coding can further reduce variance in the main analysis.
We illustrate the effectiveness of our method by revisiting a study on the
sources of election fraud with ballot image data and regression analysis.
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Our next meeting will be on October 4 (12:00 EST). Michael Lingzhi Li
presents "Statistical Performance Guarantee for Selecting Those Predicted
to Benefit Most from Treatment."
<When>
October 4, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up inside CGIS K354.
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Across a wide array of disciplines, many researchers use modern machine
learning algorithms to identify a subgroup of individuals, called
exceptional responders, who are likely to be helped by a treatment the
most. A common approach is to first estimate the conditional average
treatment effect (CATE) or its proxy given a set of pre-treatment
covariates and then optimize a cutoff of the resulting treatment
prioritization score to prioritize who should receive the treatment.
Unfortunately, since these estimated scores are often biased and noisy in
practice, naive reliance on them can lead to misleading inference.
Furthermore, practitioners often utilize the same set of data to optimize
the cutoff and evaluate the performance of the resulting subset, causing a
multiple testing problem. In this paper, we propose a methodology that has
a uniform statistical performance guarantee for selecting such exceptional
responders regardless of the cutoff optimization. Specifically, we develop
a uniform confidence interval for experimentally evaluating the group
average treatment effect (GATE) among the individuals whose estimated score
is at least as high as any given quantile value. This uniform confidence
interval enables researchers to utilize arbitrary methods to choose the
quantile of estimated score, including optimizing over the lower confidence
bound of the estimated GATE among the selected individuals. The proposed
methodology provides this statistical performance guarantee without
suffering from multiple testing problems, and also generalizes to a generic
class of statistics beyond GATE. Importantly, the validity of our
methodology depends solely on randomization of treatment and random
sampling of units and does not require modeling assumptions or resampling
methods. Consequently, our methodology is applicable to any machine
learning algorithm and is computationally efficient.
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Our next meeting will be on September 27 (12:00 EST). Tyler Simko presents
"Title: School Desegregation by Redrawing District Boundaries."
<When>
September 27, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up inside CGIS K354.
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Schools in the United States remain heavily segregated by race and income.
Previous work demonstrates districts can reduce segregation between their
schools with policies like redrawing attendance zones. Yet, the promise of
such policies in many areas is limited by the fact that most school
segregation occurs between school districts, and not between schools in the
same district. I adapt Markov Chain Monte Carlo (MCMC) algorithms from
political redistricting methodology to redraw school district boundaries
that decrease segregation while maintaining desirable criteria like
distance to school and using only existing school facilities. Focusing on
New Jersey, where the segregation of Black and Hispanic students from White
and Asian students is among the worst in the country, I demonstrate that
redrawing school districts could reduce nearly 40% of existing segregation
in the median New Jersey county, compared to less than 5% for redrawing
attendance zones alone. Finally, I show how my proposed methodology can be
applied to as few as two districts to reduce segregation in proposed
“mergers,” a consolidation of small districts into one large district.
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Our next meeting will be on September 20 (12:00 EST). Larry Han presents
"Promises and Perils of Multiply Robust Federated and Transfer Learning to
Estimate Causal Effects."
<When>
September 20, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up at 11:45 (CGIS K354).
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Abstract: Federated or multi-site studies have distinct advantages over
single-site studies, including increased generalizability, the ability to
study underrepresented populations, and the opportunity to study rare
exposures and outcomes. However, these studies are challenging due to the
need to preserve the privacy of each individual's data and the
heterogeneity in their covariate distributions. We propose a novel
federated approach to derive valid causal inferences for a target
population using multi-site data. We adjust for covariate shift and
covariate mismatch between sites by developing multiply-robust and
privacy-preserving nuisance function estimation. Our methodology
incorporates transfer learning to estimate ensemble weights to combine
information from source sites. We show that these learned weights are
efficient and optimal under different scenarios. We showcase the finite
sample advantages of our approach in terms of efficiency and robustness
compared to existing approaches. Finally, we showcase the utility of our
methodology for estimating COVID-19 vaccine efficacy (Moderna vs. Pfizer)
across geographic regions, and variations in congenital heart surgery
quality across racial/ethnic groups. Our findings have implications for the
efficient allocation of scarce resources.
Paper 1: https://arxiv.org/abs/2112.09313
Paper 2: https://arxiv.org/abs/2203.00768
Paper 3: In progress (will update with the Arxiv link soon)
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Our next meeting will be on September 13 (12:00 EST). Davide Viviano
presents "Policy Targeting under Network Interference."
<When>
September 13, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up at 11:30 (CGIS K354).
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Abstract: This paper studies the problem of optimally allocating treatments
in the presence of spillover effects, using information from a
(quasi-)experiment. I introduce a method that maximizes the sample analog
of average social welfare when spillovers occur. I construct
semi-parametric welfare estimators with known and unknown propensity scores
and cast the optimization problem into a mixed-integer linear program,
which can be solved using off-the-shelf algorithms. I derive a strong set
of guarantees on regret, i.e., the difference between the maximum
attainable welfare and the welfare evaluated at the estimated policy. The
proposed method presents attractive features for applications: (i) it does
not require network information of the target population; (ii) it exploits
heterogeneity in treatment effects for targeting individuals; (iii) it does
not rely on the correct specification of a particular structural model; and
(iv) it accommodates constraints on the policy function. An application for
targeting information on social networks illustrates the advantages of the
method.
Paper: https://dviviano.github.io/projects/main_text_NEWM_Jan2023.pdf
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
Dear Applied Statistics Workshop Community,
Welcome back! Our first meeting of the semester will be on September 6
(12:00 EST). Keyon Vafa presents "Decomposing Changes in the Gender Wage
Gap over Worker Careers."
<When>
September 6, 12:00 to 1:30 PM, EST
Lunch will be available for pick-up at 11:30 (CGIS K354).
<Where>
In-person: CGIS K354
Zoom:
https://harvard.zoom.us/j/93217566507?pwd=elBwYjRJcWhlVE5teE1VNDZoUXdjQT09
<Abstract>
Abstract: A large literature in labor economics seeks to decompose gender
wage gaps into different sources, including portions explained by
cross-gender differences in education and occupation. While career
histories contain valuable information about sources of gender wage
disparities, they are too high-dimensional to include in standard
econometric techniques. This talk presents new machine learning methods for
decomposing gender wage gaps over worker careers. We develop a "foundation
model" of career trajectories to summarize worker histories with
low-dimensional representations. We show how to fine-tune the foundation
model on small survey datasets while ensuring that the representations do
not omit features of history whose exclusion would bias decompositions. On
data from the Panel Study of Income Dynamics, our method explains more of
the gender wage gap than standard techniques. Finally, we propose a new
decomposition of the change in gender wage gaps over workers careers into
two sources: gender differences in initial characteristics and gender
differences in worker transitions. Using representations from the
foundation model, we show that early in careers, the gender wage gap
widens, driven by males transitioning to higher-paying characteristics than
females; meanwhile, later in careers, the gender wage gap narrows, driven
by female initial characteristics setting up workers for more wage growth
than those of males.
<2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/u/0?cid=Y18zdjkzcGF2OWZqa2tsZHJidTlzbm…
Best,
Jialu
--
Jialu Li
Department of Government
Harvard University
jialu_li(a)g.harvard.edu
To the Harvard Gov participants,
The Government department is seeking graduate student volunteers to fill
committee and workshop coordinator roles for the next academic year. If
you're interested in serving on a departmental committee or as a workshop
coordinator next year, please indicate your interest on this form (
https://forms.gle/aJBvpucEkwffH4E1A) by the coming Tuesday, August 1. As
you might guess, this is a relatively time-sensitive issue since speaker
series and department committees need to schedule their activities in
advance.
Best,
Shusei
Dear Applied Statistics Workshop Community,
Our next meeting of the semester will be on April 26 (12:00 EST). Dean Knox
and Guilherme Duarte will present "Optimal Allocation of Data-Collection
Resources."
<Where>
CGIS K354
Bagged lunches are available for pick-up at 11:45 (CGIS K354).
Zoom:
https://harvard.zoom.us/j/99181972207?pwd=Ykd3ZzVZRnZCSDZqNVpCSURCNnVvQT09
<Abstract>
Complications in applied work often prevent researchers from obtaining
unique point estimates of target quantities using cheaply available data—at
best, ranges of possibilities, or sharp bounds, can be reported. To make
progress, researchers frequently collect more information by (1)
re-cleaning existing datasets, (2) gathering secondary datasets, or (3)
pursuing entirely new designs. Common examples include manually correcting
missingness, recontacting attrited units, validating proxies with
ground-truth data, finding new instrumental variables, and conducting
follow-up experiments. These auxiliary tasks are costly, forcing tradeoffs
with (4) larger samples from the original approach. Researchers'
data-collection strategies, or choices over these tasks, are often based on
convenience or intuition. In this work, we show how to provably identify
the most cost-efficient data-collection strategy for a given research
problem.
We quantify the quality of existing data using the width of the confidence
regions on the sharp bounds, which captures two sources of uncertainty:
statistical uncertainty due to finite samples of the variables measured,
and fundamental uncertainty because some variables are not measured at all.
We then show how to compute the expected information gain, defined as the
expected amount by which each data-collection task will narrow these bounds
by addressing one or both sources of uncertainty. Finally, we select the
task with the greatest information efficiency, or gain per unit cost.
Leveraging recent advances in automatic bounding (Duarte et al., 2022), we
prove efficiency is computable for essentially any discrete causal system,
estimand, and auxiliary data task.
Based on this theoretical framework, we develop a method for optimal
adaptive allocation of data-collection resources. Users first input a
causal graph, estimand, and past data. They then enumerate distributions
from which future samples can be drawn, fixed and per-sample costs, and any
prior beliefs. Our method automatically derives and sequentially updates
the optimal data-collection strategy.
<2022-2023 Schedule>
GOV 3009 Website:
https://projects.iq.harvard.edu/applied.stats.workshop-gov3009
Calendar:
https://calendar.google.com/calendar/embed?src=c_3v93pav9fjkkldrbu9snbhned8…
Best,
Shusei