Using Simpson's Paradox to Discover Interesting Patterns in Behavioral Data
By Nazanin Alipourfard, Peter G. Fennell, Kristina Lerman. 2018
22 citations recorded by the source corpus.
We describe a data-driven discovery method that leverages Simpson's paradox to uncover interesting patterns in behavioral data. Our method systematically disaggregates data to identify subgroups within a population whose behavior deviates significantly from the rest of the population. Given an outcome of interest and a set of covariates, the method follows three steps. First, it disaggregates data into subgroups, by conditioning on a particular covariate, so as minimize the variation of the outcome within the subgroups. Next, it models the outcome as a linear function of another covariate, both in the subgroups and in the aggregate data. Finally, it compares trendsto identify disaggregations that produce subgroups with different behaviors from the aggregate.We illustrate the method by applying it to three real-world behavioral datasets, including Q\&A site Stack Exchange and online learning platforms Khan Academy and Duolingo.
Subfields: Artificial Intelligence, Information Systems, Statistics and Probability
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Fred Morstatter
- Aram Galstyan
- Ninareh Mehrabi
- Jundong Li
- Huan Liu
- Ruocheng Guo
- Nripsuta Ani Saxena
- Kristina Lerman
In Information Systems
- Fred Morstatter
- Aram Galstyan
- Huan Liu
- Ninareh Mehrabi
- Nripsuta Ani Saxena
- Kristina Lerman
- Jundong Li
- Gilles Trédan
In Statistics and Probability
- Aram Galstyan
- Fred Morstatter
- Nripsuta Ani Saxena
- Ninareh Mehrabi
- Lu Cheng
- P. Richard Hahn
- Kristina Lerman
- Jundong Li
Citations in the corpus
- A Survey on Bias and Fairness in Machine Learning (2019, methodology): “For example, [68] has proposed methods to test for cases of Simpson’s paradox in the data, and [2, 3] proposed methods to discover Simpson’s paradoxes in data automatically.”
- A Survey of Learning Causality with Data: Problems and Methods (2018, background): “Though existing work addressed this problem by showing difference between groups [3] in terms of causal effects.”
- The Bouncer Problem: Challenges to Remote Explainability (2019, background): “For instance, researchers seek to detect the Simpson’s Paradox [8] in the data [4].”