|
Activity Number:
|
389
- Improving Survey Data Quality with Machine Learning Techniques
|
|
Type:
|
Invited
|
|
Date/Time:
|
Tuesday, July 31, 2018 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Survey Research Methods Section
|
|
Abstract #326562
|
Presentation
|
|
Title:
|
Dynamic, Personalized Instruments via Responsive Matrix Sampling with High-Dimensional Covariates
|
|
Author(s):
|
Sean Taylor and Curtiss Cobb and Chelsea Zhang*
|
|
Companies:
|
Facebook and Facebook and UC Berkeley
|
|
Keywords:
|
surveys;
matrix sampling;
matrix completion;
web surveys;
active learning;
side information
|
|
Abstract:
|
In a landscape of declining response rates, researchers seek more meaningful data from fewer respondents willing to answer long surveys. Increasingly, survey platforms are leveraging rich sets of covariates from large data sources to build sophisticated outcome models. We show how these models can be used to reduce respondent burden without decreasing information, via an adaptive matrix sampling procedure that downsamples a longer survey instrument into the most informative subset of questions for each respondent. The question selection method optimizes for variance reduction while incorporating side information. As a respondent answers questions, their predictions are updated in an online fashion. The efficiency gains of our approach enable the researcher to terminate sampling early when desired precision is reached or adapt the instrument over time.
|
Authors who are presenting talks have a * after their name.
