Activity Number:
|
487
- Novel Causal Inference Methods for Epidemiology Studies
|
Type:
|
Contributed
|
Date/Time:
|
Wednesday, August 10, 2022 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Section on Statistics in Epidemiology
|
Abstract #323292
|
|
Title:
|
The Functional Synthetic Control Method
|
Author(s):
|
Aaron Shev* and Andrew Farris and Chris McCort and Veronica Pear and Hannah Laqueur and Rose Kagawa
|
Companies:
|
University of California, Davis and University of California, Davis and University of California, Davis and University of California, Davis and University of California, Davis and University of California, Davis
|
Keywords:
|
Synthetic control method;
Functional data analysis;
Epidemiologic methods;
Quasi-experimental design
|
Abstract:
|
We propose a new approach to the Synthetic Control Method (SCM) using the Principal Analysis by Conditional Expectation (PACE) algorithm for functional data analysis. Our proposed method improves upon the SCM by taking a functional data analysis approach that improves the performance of the classic SCM. This approach considers observed data on any given unit to comprise samples from an underlying random trajectory by estimating continuous functions for the treated unit and those units in the donor pool. It then then conducts a synthetic control analysis on these estimates. This provides several benefits. Fitting continuous curves allows the method to inherently handle missing or irregularly sampled data. Noise reduction from using smooth curves, as opposed to the observed values, allows for good pre-treatment fit in cases where poor fit may have made using SCM inadvisable. The proposed method eliminates the step of choosing arbitrary summaries of pre-treatment data for optimizing weights by using functional principal components loadings instead. Finally, PACE uses information from the mean trajectory of units to estimate unit trajectories even when units are sparsely sampled.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2022 program
|