Data from all 30 participants were analyzed using generalized mixed-effects models in R \cite{Bates2014} with a declared binomial distribution, given that the dependent variable was binary (hit or miss). The mean hit rates per phonological feature are shown in Figure \ref{fig:acc}.
A model was created with Phonological Feature as the fixed factor and Participant and Contrast (the phoneme pair) as random factors. A random slope for Phonological Feature was included in the random factor Participant, but not for Contrast as the two are inherently correlated (each contrast is a difference in one phonological feature). Differences in manner were detected significantly more often than differences in both voicing (\(\beta = -0.80, SE = 0.22, z = -3.58, p < 0.001\)) and place (\(\beta = -0.66, SE = 0.19, z = -3.58, p < 0.001\)). However, no difference in accuracy was observed between voicing and place (\(z < 1\)).