Jae-Kwang Kim
Iowa State University, IA·Mathematics & Statistics
Many questions about a population, such as how many people hold a job or have health insurance, are answered with surveys. Traditional surveys pick people at random so the results represent everyone, but they are slow and expensive.
- Fields
Mathematics & Statistics
Computer Science
- Topics
data integration
data fusion
survey sampling
- Grants
- 1 known grant, through Aug 2027
What they work on
- Develop mass imputation methods that use modern machine learning to predict missing survey variables. These methods would let a convenience sample and a probability sample be combined for population estimates.
- Create propensity score weighting methods based on information projection. These methods would estimate and correct the selection bias in convenience data, using an independent probability sample as a calibration reference.
- Develop calibration weighting that adjusts sample weights to match known population totals when there are many (high-dimensional) background variables, and extend it to correct multiple biases together.
- Derive optimal estimators and sampling designs for data fusion, then demonstrate the new methods in applications carried out with statistical agencies.
More about the research
Newer sources like web survey panels, administrative records and other large "big data" collections are cheaper and faster. However, they often miss or overrepresent certain groups, a problem called selection bias. This project develops statistical and machine learning tools for data integration: combining several data sources reliably so that conclusions about the whole population remain trustworthy.
Why it matters
Statistical agencies are under growing pressure to use cheap, fast but unrepresentative data sources. These tools could help government statisticians and survey researchers combine big data, web panels and administrative records with traditional probability surveys. This could make official population estimates more timely while reducing distortion from selection bias. The planned software and courses would help practitioners apply the methods.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Armeen Taeb
University of Washington, WA·Mathematics & Statistics
Develop a framework that organizes complex model classes (partitions, permutations, directed acyclic graphs, subspaces) as partially ordered sets, and use it to define generalized false positive error.
Leonard Schulman
California Institute of Technology, CA·Computer Science
Relax the statistical separation assumption needed when a hidden global confounder takes a limited number of values, aiming instead for model identification measured in Wasserstein distance, a way of comparing probability distributions.
Yao Zheng
University of Connecticut, CT·Mathematics & Statistics·Early-career
Develop recurrent neural network (RNN) models for high-dimensional time series that stay interpretable through Granger causality, a test of whether one series predicts another, to balance accuracy and interpretability.
The school
Iowa State University: 23 labs on PI Pages, $462M a year on research, 327 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.