Evaluation of two algorithms for counting forward saccades in a reading task

We compared two algorithms, which are used to assess the number of forward saccades in a reading task from records of eye movements. In one algorithm saccades are detected analysing the velocity of eye movements. The third derivate of eye position in time (jerk) is used in the second algorithm for the detection of saccades. Both algorithms were applied on the same set of data, recorded using 24 subjects reading a German text, which was presented on two different displays. Our subjects read the text at a mean reading speed of 258.5 word/min. Both algorithms were found to produce a similar rate of artefacts in the number of detected saccades (2.5%), provided the threshold for detection (velocity or jerk) is set at an appropriate level and the same level of threshold is applied to all data. In both algorithms, the rate of artefacts increases with increasing distance of the threshold from its optimum. Inter-individual variation of the rate of artefacts increases more pronounced in the algorithm based on jerks. Eye blinks were identified as a major source of artefacts. A remedy is proposed, by means of which the rate of artefacts can be reduced.