How to recover Selection Bias from Sample and Survey Data

In Data Science, more often than not, people think the more data the better. However, despite the vast amount of data, the data collection process could determine whether the analysis is biased. In the talk, Anh will present standard techniques to remove the selection bias. When the necessary assumptions do not hold, she provides considerations about how to exploit domain knowledge to recover partially from selection bias, which in some cases result in informative bounds. The methodology has several advantages: (1) It operates non-parametrically based on the causal graphical model which is less sensitive to model misspecification; and (2) it does not rely on having the selection probability which is not always available. This methodology is very useful especially in extrapolating results from survey samples to the whole population, and in any causal analysis.
Who is this presentation for?
Data Scientists
Prerequisite knowledge:
Probability, Statistics
What you'll learn?
How to remove selection bias

Anh is a passionate Data Scientist and an active public speaker who has a solid international background in data analytics and data management across Asia and Europe. With more than 11+ years of experience in different industries such as finance, ride-hailing, travel technology, manufacturing and wholesale, she is keen on bringing to the audience her unique perspectives, best practices and know-how when it comes to applying disruptive technologies.