Methodological approach
Analyses relate stimulus conditions, feature assessments, visual judgements and — depending on the task — subjective confidence.
In corresponding observer studies, stimuli can be presented under controlled conditions and responses, including subjective confidence, can be recorded. Repeated measurements can then be used to examine repeatability, agreement between observers, and — for tasks with objectively defined accuracy — the relation between confidence and accuracy.
Controlling the stimulus
Faces change their appearance with view, projection, image quality, illumination, expression, and temporal distance. These conditions are part of the question, not secondary details.
3D/2D material makes it possible to derive views systematically and to examine which cues remain visible across representations and which are weakened, emphasised, or no longer discernible because of representation and perspective. Metric frequency and visual salience must be distinguished: in a published 3D analysis, for example, eye-axis skew was frequent but clearly visible in only 4.1% of the examined individuals.
Recording the judgement
Observer studies can collect age estimates, similarity ratings, same/different decisions, assessability ratings, and confidence. Repeated tasks and independent observer groups make it possible to see whether a judgement is stable, shared, or strongly observer-dependent.
Same/different means deciding whether two images show the same or different people. Confidence is the observer’s stated certainty.
Analysing variability
Differences between judgements may originate from the face, the view, image quality, the task, or the observer. These sources have to be separated analytically if visual judgements are to be examined systematically.
Descriptive statistics, measures of repeatability and agreement, regression and mixed models, and confidence-calibration analyses can be used for this purpose. The aim is to measure variability and examine its sources. A published observer assessment also showed that measurement itself can depend on the feature: mean interobserver deviation was 0.19 mm for vertical ear position and 0.61 mm for the more difficult lateral projection of the nose.
Show psychological analysis perspectives
Observer data make it possible to describe how strongly judgements vary between individuals and how stable they remain across repeated measurements. With an appropriate design, it is also possible to examine how confidence and accuracy are related and what proportions of variance are explained by stimulus, perspective, image quality, task, and observer.
This also raises a psychophysical question: how are controlled stimulus conditions related to visual judgements and their confidence?
Possible levels of analysis include visible features, observer differences and — depending on the task — decision criteria as well as the calibration of confidence and accuracy.
Uncertainty and decision thresholds
Visual decisions are often made under incomplete information. A cautious or uncertain judgement may therefore be appropriate. High confidence, however, does not necessarily imply high accuracy.
For tasks with objectively defined accuracy, the relation between stated confidence and actual accuracy can be examined.