Degree Name

Doctor of Philosophy


University of Wollongong.Department of Psychology


Ocular artifact is a major source of contamination of the EEG. This artifact causes serious difficulties in EEG interpretation, and although methods to overcome these difficulties have been sought since the late 1960s, there is no consensus on how best to do this. A widely used means of removing ocular artifact is termed 'EOG correction', where a portion of EOG is removed from the EEG. There are a number of ways of performing this 'correction'. This thesis is an attempt to resolve these differences.

A review of the literature suggested that the main discrepancies between EOG correction findings (different rates of EOG propagation for different eye-movement types and/or frequencies) could be removed if EOG magnitude was held constant. A simulation study found that low EOG magnitude significantly inflated propagation estimates (Bs), and a second study found that the same inflation pattern pertained subject data. It was then found that if interference was removed, differences between Bs for different eye-movement types could also be removed. Thus it was suggested that propagation does not vary between eye-movement types and/or frequencies.

A means of averaging eye-movements was thus suggested to overcome the effects of interference in EOG and EEG channels (the AAA method). When tested with simulation data, AAA was found to be relatively unaffected by interference. A more easily implemented version of the AAA (NERP) was tested empirically, and found to produce equivalent Bs. It was also found that at least 40 epochs should be included such an averaging procedure. Due to difficulties with correcting blink data with saccade Bs under certain circumstances, the role of the radial EOG channel voltage above and below the eye) was explored. It was found that in order to correctblink data with saccade Bs adequately, the radial channel was required, and a revised version of the AAA (RAAA) was proposed for this purpose. This was tested and found to produce very good correction, and better than that of blinks corrected with blink data Bs, or saccades with saccade data Bs.

It was also found empirically that vertical, horizontal and radial EOG channels are required for good correction, that Bs remain constant for at least ninety minutes, that calibration trials are appropriate for calculating Bs, and that simultaneous regression is the best form of multiple regression for doing this. A number of other issues were addressed theoretically, the main one being the need to make an adjustment for overcorrection, even when Bs are accurately calculated. An adjustment was thus proposed on theoretical grounds, but has not yet been validated. To summarize the various conclusions reached in this thesis, a suggested EOG correction algorithm is presented.