Abstract

Today's massive datasets make statistical computing and displays even more needed than when the terms "Statistical Computing" and "Statistical Graphics" evolved as disciplines 50 years ago. Because the central goals of data analysis are insight and inference, and because rarely should all data be displayed, we need algorithms and data displays that meet both these objectives. Further, `big data' are even more likely to require robust techniques, due to exotic values, outliers, or mixtures of distributions. Finally, more data does not imply more confidence, especially when they are non-representative of their target populations. Robust statistical methods are essential to these displays, in sampling the dataset, estimating key quantities, and
communicating insights and inferences. This talk will discuss some recently developed displays that demonstrate that statistical methods in the `data science' era remain critical for analyzing `big data.'