bad_classification: Detect poor performers

In kleinschmidt/phonetic-sup-unsup: Supervised/unsupervised adaptation analysis

Description

Basic check: find people who are classifying the endpoints inconsistently or inaccurately. One way to measure quality is by the fitted log-odds of /p/ response at 0 and 70 ms (continuum endpoints, for all intents and purposes). This measures how consistent people are (in a smoothed way) at the ends of the continuum. Intuitively, we want to throw out anyone who doesn't ever consistently get both /b/ and /p/. So we'll set a cutoff of 80 logodds of a /p/ response greater than about 1.4 ('qlogis(0.8)') at the /p/ end of 70ms and less than -1.4 at the /b/ end.

Usage

bad_classification(data_, vots = c(0, 70))

Arguments

data_	Parsed data frame
vots	= c(0,70) where to check predicted classification.

Value

A data.frame of subjects/assignments with column 'exclude80PercentAcc' (whether this assignment is excluded by the criterion of at least 80

kleinschmidt/phonetic-sup-unsup documentation built on May 20, 2019, 12:33 p.m.