Keywords

Pillai scores and multidimensional vowel overlap, sample-size effects in Pillai-based merge detection, simulation of merged vowel classes in bivariate space

Abstract

Pillai scores in Sociolinguistics

A way to quantify overlap in multiple dimensions (e.g., F1, F2, duration) (Hay et al., 2006; Nycz & Hall-Lew 2013)
Some concern about unequal vowel categories sizes (Johnson 2015; convo on Twitter June 2, 2021)
Our solution: a big simulation to show exactly what impact sample size has on Pillai scores

The Simulation

1. Start with a single bivariate normal distribution.

2. Sample two “vowel classes” from this distribution. – Ground Truth is they’re merged.

3. Calculate the Pillai score – Should be close to 0 because they’re merged!

4. Repeat many, many times.

Original Publication Citation

Joseph A. Stanley & Betsy Sneller. “How Sample Size Impacts Pillai Scores – and What Sociophoneticians Should Do About It.” New Ways of Analyzing Variation 50. San Jose, California. October 13–15.