We present a method of directly analysing lie detection data, without the need to recode, and aggregate, or incorporate dubious corrections: Generalised Linear Mixed Models with a Probit Link. These translate to Signal Detection Theory models w/o requiring data processing!

This method avoids many of the pitfalls of ANOVA-style analyses, or the need for up to 4 separate (unconnected) analyses! And, it incorporates important sources of variance in the data (from senders and judges), ensuring more realistic effect estimates.