Day 1: May 23rd, 2022
Session(s) | Time | Description | Presenters | Room |
---|---|---|---|---|
Full Day Session 1 | 9:00am - 4:00pm | AT CAPACITY -Interventionist mediation analysis with single world intervention graphs | Jamie Robins & Ilya Shpitser | BW 1217 |
Full Day Session 2 | 9:00am - 4:00pm | Targeted Machine Learning of the Causal Effects of Dynamic and Shift Interventions with the tlverse R Packages | Mark van der Laan, Alan Hubbard, Jeremy Coyle, Nima Hejazi, Ivana Malenica, & Rachael Phillips | BW 1104 |
Morning Session 1 | 9:00am - 12:00pm | Improving research planning through design declaration and diagnosis in DeclareDesign | Graeme Blair | BW 1215 |
Morning Session 2 | 9:00am - 12:00pm | AT CAPACITY - Machine learning & nonparametric efficiency in causal inference | Edward Kennedy | BW 1102 |
Morning Session 3 | 9:00am - 12:00pm | AT CAPACITY -An introduction to instrumental variable methods | Hyunseung Kang & Luke Keele | BW 1203 |
Afternoon Session 1 | 1:00pm - 4:00pm | AT CAPACITY -Randomization inference for experiments and observational studies: a holistic look | Colin Fogarty, Xinran Li, & Anqi Zhao | BW 1203 |
Afternoon Session 2 | 1:00pm - 4:00pm | Introductory Workshop for Using Propensity Score Weights When Drawing Causal Inferences | Brian Vegetabile | BW 1215 |
Afternoon Session 3 | 1:00pm - 4:00pm | AT CAPACITY -Causal inference for multiple time-point (longitudinal) exposures | Laura Balzer & Lina Montoya | BW 1102 |
about workshops
Attend one full-day workshop or one half-day workshop or two half-day (1 morning and 1 afternoon session) pre-conference workshop. Each workshop has a maximum attendance, so early registration is encouraged. Pre-conference workshops are intended/designed for in-person participation but remote attendance is available if needed. Registration for pre-conference workshops will NOT be available on-site. All workshops are separately ticketed activities. For workshop pricing, please visit ACIC’s Registration page.
Full-day Sessions
Interventionist mediation analysis with single world intervention graphs
May 23, 9:00 am - 4:00 pm
Location:
Berkeley West Room 1217
Presenters
Jamie Robins & Ilya Shpitser
Description
Mediation analysis is a framework for assessing direct, indirect, and pathway specific causal relationships, and has seen applications in psychology, public health, and disparity research. This short course will present a graphical modeling framework of Single World Intervention Graphs for describing counterfactual targets of inference that arise in mediation analysis and causal inference more generally.
Using this framework, we will describe an identification theory for these targets of inference, which include the generalization of the g-formula to mediation settings, and hidden variable settings, and discuss appropriate estimation strategies in important special cases. The morning session will focus on SingleWorld Intervention Graphs (SWIGs) and Interventionist Mediation. The Afternoon Session will focus on Interventionist Mediation In Models With Hidden Variables.
Workshop is at capacity.
Targeted Machine Learning of the Causal Effects of Dynamic and Shift Interventions with the tlverse R Packages
May 23, 9:00 am - 4:00 pm
Location:
Berkeley West Room 1104
Presenters
Mark van der Laan, Alan Hubbard, Jeremy Coyle, Nima Hejazi, Ivana Malenica, & Rachael Phillips
Description
This workshop focuses on the use of Targeted Machine Learning estimation to evaluate the effects of dynamic and stochastic interventions. Participants will learn to:
1. Train a super learner using the sl3 R package to estimate prediction functions, including the conditional mean, conditional probability, or conditional density.
2. Use the tmle3mopttx R package to learn the optimal individualized treatment regime, and to estimate effects under such data-adaptive regimes.
3. Approximate the causal effect of shifting a continuous-valued exposure with the tmle3shift R package.
4. Estimate direct and indirect causal effects with the tmle3mediate R package, which decomposes the total causal effects of static or stochastic interventions
5. Differentiate stochastic, dynamic, optimal dynamic, and static exposure regimes from each other, in terms of their interpretation and the assumptions required for the identification of their respective causal effects from the observed data.
The workshop will incorporate four learning modules, one for each tlverse R package covered, including sl3, tmle3mopttx, tmle3shift, and tmle3mediate. Each module will take place in-person over a span of about 1.5 hours and will incorporate a concise preliminary lecture, vignette-guided live coding exercises, and discussion. Participants will therefore have the opportunity to practice hands-on implementation of these estimators for answering causal questions with real-world cross-sectional data.
Maximum Attendance
10 seats remaining.
Morning Sessions
Machine learning & nonparametric efficiency in causal inference
May 23, 9:00 am - 12:00 pm
Location:
Berkeley West Room 1102
Presenters
Edward Kennedy
Description
This short course covers the basics of efficient nonparametric estimation in causal inference, including estimating equations, TMLE, and double machine learning. It considers nonparametric efficiency bounds for causal estimands, and efficient bias-corrected estimators based on influence functions. Importantly, these estimators yield fast rates of convergence and normal limiting distributions, even in complex nonparametric models where nuisance functions (e.g., propensity scores) are estimated with modern machine learning tools. The estimators are often doubly robust. Background in mathematical statistics is useful but not required. The workshop covers both theory and application, including R code for implementing various methods.
Workshop is at capacity.
An introduction to instrumental variable methods
May 23, 9:00 am - 12:00 pm
Location:
Berkeley West Room 1203
Presenters
Hyunseung Kang & Luke Keele
Description
The instrumental variables (IV) framework offers an alternative approach for estimating causal effects in the presence of unmeasured confounding. The framework requires a variable that is (1) independent of unmeasured confounders, (2) affects the treatment, and (3) affects the outcome only indirectly through its effect on the treatment. The course will begin with the application of IVs to handle non-compliance in RCTs, a classic application of IVs. This portion of the course will introduce concepts related to compliance classes, the Wald estimator, two-stage least squares (2SLS), and local average treatment effects (LATE). The second portion of the course will focus on IVs based on natural experiments. The course will include numerous real-world applications of IVs to illustrate concepts and help course participants understand how to evaluate IV assumptions.
Workshop is at capacity.
Improving research planning through design declaration and diagnosis in DeclareDesign
May 23, 9:00 am - 12:00 pm
Location:
Berkeley West Room 1215
Presenters
Graeme Blair
Description
A research design is a procedure for generating empirical answers to theoretical questions. Research designs can be strong or weak. Assessing whether a design is strong requires having a clear sense of what the question is and knowing whether the answers a study is likely to deliver are reliable. In this session, we will introduce the nuts-and-bolts of using DeclareDesign, which offers a language for describing research designs. We will also introduce an algorithm for selecting among researchers designs using DeclareDesign that involves using Monte Carlo simulation to “diagnose” the properties of the design and redesigning among feasible alternatives. The session will be hands-on: we will provide a live demo and then we will facilitate several activities in which you will declare, diagnose, and redesign stylized studies in small groups. You will leave knowing how to set up a design in DeclareDesign, modify key parts to match your research setting, and use our simulation tools to assess quantities like the power and bias of the design. There will be a short pre-reading from our new book manuscript to enable us to jump right in.
Maximum Attendance
TBD
Afternoon Sessions
Causal inference for multiple time-point (longitudinal) exposures
May 23, 1:00 pm - 4:00 pm
Location:
Berkeley West Room 1102
Presenters
Laura Balzer & Lina Montoya
Description
This workshop applies the Causal Roadmap to estimate the causal effects with multiple intervention variables, such as the cumulative effect of an exposure over time and the effects on survival-type outcomes with right-censoring. We will cover longitudinal causal models, identification in the presence of time-dependent confounding, and estimation of joint treatment effects using G-computation, inverse probability weighting (IPW), and targeted maximum likelihood (or machine learning) estimation (TMLE) with Super Learner. During the workshop session, participants will work through the Roadmap using an applied example and implement these estimators with the ltmle R package.
Prior training in causal inference with a single time-point exposure is recommended.
Workshop is at capacity.
Randomization inference for experiments and observational studies: a holistic look
May 23, 1:00 pm - 4:00 pm
Location:
Berkeley West Room 1203
Presenters
Colin Fogarty, Xinran Li, & Anqi Zhao
Description
The workshop aims to provide a holistic view of the status quo of randomization inference in the analysis of randomized experiments and observational studies. The dominant approach to statistical analysis, often known as model-based inference, models the observed outcomes as a random sample from some possibly hypothetical super-population, and conducts inference based on often unverifiable assumptions on the outcome-generating process. Randomization inference, on the other hand, focuses particularly on the study population in hand, and takes the physical act of randomization as the sole source of randomness. The resulting inference makes no assumptions about the outcome generating process, and provides robust protection against model misspecification. Building on the seminal work of Fisher and Neyman, recent researchers have adapted these ideas to a range of treatment-control experiments and observational studies.
The goals of the workshop are: (1) Understand the basics of randomization inference and its recent developments; (2) Understand the difference between model-based and randomization inference, and be able to choose the right approach based on inference goal; (3) Understand how to perform a sensitivity analysis for hidden bias in observational studies within the randomization inference framework; and (4) Be able to perform basic randomization inference in R.
Workshop is at capacity.
Introductory Workshop for Using Propensity Score Weights When Drawing Causal Inferences
May 23, 1:00 pm - 4:00 pm
Location:
Berkeley West Room 1215
Presenters
Brian Vegetabile
Description
Estimation of causal effects is a primary activity of many studies. Examples include testing whether a substance use treatment program is effective, whether an intervention improves the quality of mental health care, or whether new medicines cure a disease. Controlled, random-assignment experiments are the gold standard for estimating such effects. However, experiments are often infeasible, forcing analysts to rely on observational data in which treatment assignments are out of the control of the researchers. This workshop will provide an introduction to causal modeling using the potential outcomes framework and the use of propensity scores and weighting (i.e., propensity score or inverse probability of treatment weights) to estimate causal effects from observational data.
The goals of the workshop are to increase attendee’s understanding of: (i) how to define and estimate causal effects using the potential outcomes framework; (ii) how to use propensity scores and inverse probability of treatment weights when estimating causal effects; and (iii) how to assess the validity of key assumptions of the proposed methods. We will also provide attendees with step-by-step instructions and guidelines for analyses involving binary treatments. Attendees will gain hands-on experience estimating propensity score weights using boosted models in R; evaluating the quality of those weights; and using them to estimate intervention effects. Additional topics will include methods for conducting sensitivity analyses for unobserved confounding and careful consideration of other candidate estimation methods for estimation of balancing weights.
Attendees should be familiar with linear and logistic regression; no knowledge of propensity scores is expected.
Maximum Attendance
TBD
Useful Links
Registration
Workshop Schedule & Descriptions
Conference Schedule
Data Challenge Closed May 1
Lodging
Transportation
Proposal Portal Closed
Sponsorship Closed
Travel Grants Awarded
ACIC 2022 Committee Members
Poster Presentation Guidelines
Jobs Portal & Careers Lunch Session
ACIC Code of Conduct
Have a Question?
Covid-19 Information & Disclaimer
As members of the campus community, the ACIC Planning Committee is not autonomous in our decision-making, and is subject to UC Office of the President, UC Berkeley, and government guidelines.
The ACIC Planning Committee strongly encourages travelers to purchase refundable or transferable tickets, and is not responsible for travel or other similar costs.
Masks are strongly recommended, but not required, indoors - regardless of vaccination status. N95 masks will be made available.
All participants will be required to adhere to UC Berkeley, UC Office of the President, and local, regional, state, and federal mandates. Failure to comply with COVID-19 safety protocols will result in dismissal from any conference and/or workshop and forfeiture of registration fees.