Countermeasures for loss of situation awareness: Error filtering strategies for 3-dimensional audio

The U. S. Army has proposed incorporating 3-dimensional (3-D) audio displays into the cockpit as a technological countermeasure for spatial disorientation (SD). We evaluated the effectiveness of a 3-D audio display and found localization accuracy for discrete, short-duration signals fell short of previously published literature. The present investigation explored the extent to which this discrepancy was due to differences in error filtering strategies used during the data screening process. Error filtering is the common practice of eliminating from analysis trials with fore/aft reversals. We applied conservative and liberal error filtering to two comparable datasets from experiments on 3-D audio cue localization accuracy. Conservative error filtering consisted of eliminating from analysis any trial characterized by a 180-degree reversal for the fore and aft positions. Liberal error filtering eliminated trials characterized by hemispheric reversals (i.e., fore/aft), inclusive of central (0/180 degrees) and offset (45 degrees left or right of center) positions. One experiment had been conducted in a quiet environment; the other was conducted in the presence of loud, operationally-relevant ambient noise. Compared to raw, unfiltered performance data, conservative and liberal error filtering led to a 2 and 4% difference, respectively, in accuracy. A significant interaction between error filter type and ambient environment was found (p <; .01). Conservative error filtering resulted in a significant increase in apparent accuracy (over unfiltered data) for localization data from the quiet environment (p <; .05). Liberal error filtering significantly improved apparent accuracy for data from the noisy environment. Error filtering caused the appearance of better performance than was actually present in the data. Further, even these small increases in mean accuracy are sufficient to alter the outcome of null-hypothesis-significance-testing, potentially leading to faulty conclusions through Type I error. Based upon this evidence, we strongly support an approach that retains all fore/aft reversals for experiments on 3-D audio localization accuracy. Doing so will improve the accuracy and validity of such investigations. Instead of focusing on error-screened data, we advocate actually improving 3-D audio localization performance through redundant vibrotactile cueing, where possible, particularly for noisy environments.