FACES is the dimensional alternative to action-unit coding, and the dominant behavioural measure of blunted affect. Rather than decomposing a face into muscle movements, a coder finds each expression event and records three things about it: whether it was positive or negative, how intense it was, and how long it lasted. Frequency, mean intensity and total duration per valence then fall out of the coding — six summary variables from one pass through the tape.
What counts as one expression. An expression is any change from a neutral face to a non-neutral one and back to neutral. A shift straight from one non-neutral display to another counts as an additional expression rather than a continuation. Getting raters to agree on segmentation is most of the training; agreement on valence and intensity comes comparatively easily.
Valence is binary, and that is deliberate. There is positive and there is negative — no mixed category. A smile that collapses into a frown is two events with two valences, not one ambivalent event. Coders who want a "both" option are usually trying to avoid making the segmentation decision the system is asking for.
Duration comes from the timeline, and this is where Tagaroo helps. The original protocol has coders read a burned-in time-mark off the tape and subtract. Here the annotation's own start and end are the duration: tag the expression from onset to return-to-neutral and the number is recorded for you, which removes the single most tedious and error-prone step in the method.
The third code is a segment-level global. Overall expressiveness is rated once for a whole segment on a 1-to-5 scale, independently of the event counts — a useful sanity check on them, and the variable most comparable to clinical impressions of flat affect.
Reliability in the original work was reported as intraclass correlations (absolute agreement), averaging around .88 across samples after 10 to 20 hours of coder training. Budget the training; this is not a scheme raters can pick up cold.