This Review of the Literature is divided into two sections. In the first and primary section, literature related to characteristics of the individual which may be predictive of sickness is reviewed. In the second section, literaturerelated to previous attempts at prediction of simulator sickness is reviewed.
Related to age is experience with the real-world task, which plays a critical role in the sensory conflict theory of simulator sickness: conflicts are thought to occur between the actual pattern of stimuli and the expected pattern of stimuli. The expected patterns likely result from repeated experiences, which Reason and Brand (1975) suggest may follow the same long-term learning pattern seen with other types of learning. Age and experience are correlated and, as is discussed shortly, experience with the real-world task is positively correlated with sickness.
Because age appears to be closely related to sickness susceptibility and was easily obtainable, it was investigated in this study as a predictor of sickness in virtual environments.
Because of the lack of any kind of formal metric to assess concentration level, this characteristic was not investigated in this study.
Because of the largely unestablished nature of the relationship between ethnicity and sickness and because of the difficulty which may have been associated with obtaining an appropriate number of Asian individuals to fully investigate this characteristic within the sample population, ethnicity was not investigated in this study.
With VR applications, the relationship between the "real-world" and simulatedtasks is likely to not be as clear cut as the relationship between those tasks in a flight simulator. Specifically, with VEs used for entertainment purposes, there may be no analogous real-world task. Nevertheless, for VR applications which have a real-world counterpart, the simulator sickness literature suggests that increased experience with that real-world task may be associated with increased sickness.
The task performed in this experiment consisted of a video game with no analogous real-world counterpart. Thus, experience with the real-world task was assumed to be zero for all participants in this study.
Due to tradeoffs between the number of variables included in the analysis and the number of research participants needed, VR experience was not investigated in this study. However, attempts were made to control for adaptation from any previous VR exposure by only including participants who had not been exposed to a VR system within 30 days prior to their experimental session. Thus, participants in this study were assumed to have little or no adaptation.
Thorough investigation of the flicker fusion frequency threshold should include observations of sickness for each individual at several times during the day to address diurnal changes. It should also include some measure of the threshold at these times. Lacking both the equipment necessary to measure the threshold and the resources to obtain multiple measurements on each participant, this characteristics was not included in this study.
Because gender appears to be an important aspect of sickness susceptibility and was easily obtainable, it was investigated in this study as a predictor of sickness in virtual environments.
Illness was not specifically investigated in this study, but experimental pre-screening attempted to eliminate those participants with conditions noted above. Thus, illness was assumed to be low or non-existent for the participants in this study.
Parker and Harm (1992) proposed that mental rotation is important for efficient goal-directed locomotion - a common task in a VE - given that a person must orient in order to locomote efficiently. They cited several examples which support their claim that the ability to perform mental rotation is important for competent function and the reduction of motion sickness. They have also conducted their own research on mental rotation ability and space motion sickness using the Device for Orientation and Motion Environments Preflight Adaptation Trainer (DOME-PAT), a microgravity simulator.
The first study involved cosmonauts during a Soviet space mission. These cosmonauts were trained in a mental rotation procedure prior to flight and, during the mission, they significantly improved their performance on the procedure. Parker and Harm (1992) argued that, during the mission, the cosmonauts learned to locomote in their microgravity environment, a task which required mental rotation ability. They suggested that by improving their performance on this complex task, performance on the easier mental rotation procedure was also improved.
As a second piece of supporting evidence, Parker and Harm (1992) pointed out that different astronauts appear to have different methods of dealing with the sensory disturbances experienced in microgravity. Astronauts who appear to deal with the absence of gravity by paying more attention to internally-generated orientation vectors - especially the one associated with their Z-body axis (up-down) - are termed Type IZ. Parker and Harm suggested that these astronauts, who generally report little or no motion sickness during their space flights, are able to ignore visual cues for upright.
These results have several implications for VEs (Parker & Harm, 1992). First, mental rotation tests could be employed to identify individuals who may be less likely to experience sickness in VEs. Second, VR users could receive training to improve their mental rotation abilities. Third, mental rotation skills learned in one virtual environment appear to transfer to other virtual environments.
Although Parker and Harm (1992) conducted their investigation using a sophisticated simulator, mental rotation ability can also be assessed with paper-and-pencil tests. Because of the availability of such a test and because of the implications of mental rotation ability to performance in a VE as discussed above, this characteristic was investigated in this study as a predictor of sickness in virtual environments.
Several tests can be used to classify an individual's perceptual style. The classical test is the Rod and Frame Test (RFT). This test measures the accuracy with which an individual can adjust a rod to the true vertical position under conditions of visual-kinesthetic conflict. Another test of perceptual style is the Embedded Figures Test (EFT), which measures the subject's ability to extract a geometric pattern from a complex pattern. Based on the performance on such tests, an individual's perceptual style is classified as either field-independent or field-dependent. "Field-independent" individuals are able to perceive items as separate from a surrounding field - such individuals are able to adjust the rod to its true vertical with high accuracy and can successfully extract geometric patterns from the complex patterns. The perception of "field-dependent" individuals, however, is strongly dominated by the surrounding field - such individuals are unable to accurately adjust the rod to its true vertical and have difficulty discerning geometric patterns from complex patterns.
Barrett and Thornton (1968) suggested that field-independent individuals are more sensitive to body cues than are field-dependent individuals. Because of this sensitivity and the conflict between static body cues and dynamic visual cues in a moving display, they predicted that field-independent individuals should be more susceptible to simulator sickness than field-dependent individuals. Barrett and Thornton (1968) and Barrett, Thornton, and Cabe (1969, 1970) investigated the possible relationship between perceptual style and simulator sickness.
Barrett and Thornton (1968) found that all of the extremely field-independent subjects left the simulator ill and, even though some field-dependent subjects also became ill in the simulator, they concluded that the results supported their prediction that field-independent individuals would experience more discomfort. Barrett, Thornton, and Cabe (1969), however, found no relationship between simulator sickness and perceptual style as measured with the EFT. Barrett, Thornton, and Cabe (1970) investigated the relationship between perceptual style and cue conflict induced by a "haunted swing"-like device. This amusement park device consisted of a large immobile swing inside of a moving room. The results indicated that, although many subjects did experience discomfort in the swing when the room moved, it was the field-dependent individuals who experienced the most discomfort - opposite the Barrett and Thornton (1968) hypothesis.
Although, as Long, Ambler, and Guedry (1975) pointed out, there would be great theoretical and practical value if a predictive relationship could be found between perceptual-style and susceptibility to sickness, the only clear result that can be discerned from these studies is that there is no clear result. Frank and Casali (1986) reviewed additional studies examining the relationship between perceptual style and simulator sickness. They also concluded that little convincing evidence exists to support the theory that field-independent individuals are more susceptible than field dependent-individuals.
Several points can be made about the perceptual style literature. Frank and Casali (1986) noted that, in order for perceptual style to be a meaningful predictor, the entire range of the perceptual style continuum must be considered. Many studies, however, focus only on the extremes of field independence and dependence. In addition, Ebenholtz (1977) suggested that the visual system of field-dependent subjects may be peripheral-dominant whereas the visual system of field-independent subjects may be foveal-dominant. Because the periphery is more sensitive to motion and, furthermore, because perception of motion in the periphery may induce vection, Ebenholtz's suggestion implies that field-dependent subjects would be more likely to experience a conflict between visual and proprioceptive stimuli. By this reasoning, field-dependent individuals should be more susceptible to simulator sickness, a prediction which is opposite the hypothesis proposed by Barrett and Thornton (1968).
Clearly, the relationship between perceptual style and simulator sickness, if one exists, is not an obvious one. Because of the inconsistent findings in the literature, as well as tradeoffs between the number of variables included in the analysis and the number of research participants needed, perceptual style was not investigated in this study.
Kolasinski et al. (1994) hypothesized that individuals who are less posturally stable will be more likely to experience simulator sickness or will experience more severe sickness; conversely, individuals who are more posturally stable will be less likely to experience simulator sickness or will experience less severe sickness. To investigate this hypothesis, pre-exposure postural stability and post-exposure sickness data from Navy helicopter pilots were analyzed. It was found that pre-exposure postural stability was most strongly associated with the Nausea and Disorientation subscale scores on the SSQ. This result complements previous results which have shown that poor post-exposure postural stability is related to high Disorientation subscale scores (Jones, Kennedy, Lilienthal, & Berbaum, 1993).
A relationship between pre-exposure postural stability and post-exposure sickness would provide support for the use of postural tests as predictors of simulator sickness. It could also have specific implications for liability issues concerned with VR systems. Decreased postural stability for a given individual might be indicative of illness, drugs, or alcohol. It is highly likely that individuals in such states would be more likely to experience sickness. Thus, some specified level of postural stability could be used as a requirement before an individual would be permitted to use those systems. Similarly, Kennedy and Stanney (1996) have suggested that a video posture measure could be used to compare post-exposure postural stability to pre-exposure postural stability as a means of certifying that users of VEs are safe to return to the natural environment.
It appears that there would be practical benefits if it could be established that postural stability is a predictor of sickness in VEs. Kolasinski et al. (1994), however, only examined the relationship between postural stability and sickness in terms of the strength of the association between them - the use of postural stability as a predictor of sickness was not investigated. This study extended the work of Kolasinski et al. to investigate postural stability as a predictor of sickness in virtual environments. Thus, postural stability was included as an independent variable in this study.
Hettinger, Berbaum, Kennedy, Dunlap, and Nolan (1990) hypothesized that vection must be experienced before sickness can occur in fixed-base flight simulators. From their results, they concluded that visual displays which produce vection are more likely to produce simulator sickness. Although all of the subjects who experienced sickness also experienced vection at some point during the experimental trials, not everyone who experienced vection experienced sickness. In other words, it appears that vection may be necessary but not sufficient for sickness. Extending this result, it is also likely that individuals who are prone to experience vection may be prone to experience sickness.
A vection drum was available to assess susceptibility to vection for this study. However, because of the logistics associated with gathering such data based on the resources available for this research, susceptibility to vection was not investigated in this study.
Although they did not develop a predictive model, Kennedy, Dunlap, and Fowlkes (1990) provided some quantitative information which may aid in the investigation of predicting simulator sickness. They reviewed over 100 studies on motion sickness prediction and conducted a meta-analysis to estimate the percentage of variance accounted for by various possible predictors. They concluded that operational measures, which measure sickness occurring in the actual environment, were estimated to account for the largest proportion of variance (67%). Provocative laboratory tests - designed to induce sickness - and motion sickness history explained the next largest proportions of variance, 38% and 34%, respectively. Perceptual style was estimated to explain roughly one quarter (26%) of the variance. Finally, other individual characteristics such as autonomic nervous system tendency, positive function of the end organs, and personality measures were estimated to each explain less than 10% of the variance. Kennedy et al. (1990) concluded that a composite measure consisting of motion sickness history, physiological variables, and standardized laboratory provocative tests might be the most appropriate predictor set.
Of the literature reviewed, only one which reported development of a predictive model was found. Reschke (1990) reviewed studies which investigated the prediction of space motion sickness based on susceptibility to sickness in KC-135 parabolic flights. Initially, 27 variables were investigated using a factor analysis. These variables consisted of age, gender, performance on several laboratory provocative tests, performance on tests of the Vestibulo-Ocular Reflex, performance on several tests of postural stability, and motion sickness history. Multiple discriminant analysis was used to develop functions to predict sickness and then determine how well those equations predicted. Reschke (1990) reported that the accuracy of the resulting predictions were better than chance.
An attempt was then made to enhance the prediction through the use of a logistic model. The model developed can be found in Reschke (1990). The percentage of cases correctly classified by the linear discriminant analysis model was 66%, whereas the percentage correctly classified by the logistic model was 83%. Thus, the logistic model appeared to perform better than the linear discriminant analysis model.
The goal of this investigation was to determine if age, gender, mental rotation ability, and pre-exposure postural stability could be used to develop a predictive model of sickness. The only definitive a priori hypothesis was that sickness would be successfully modeled on these four characteristics or combinations thereof. Characteristics which entered the model as single variables would be expected to relate to sickness in the manner suggested by the literature, but it was unknown how the characteristics would interact and how they would behave together in a model.
References in the literature to the prediction of simulator sickness were also reviewed. Although much has been said about which variables are likely to be predictive of sickness, it appears that very little actual modeling has taken place. Of the literature reviewed here, the sole exception to this is the research presented by Reschke (1990) which employed both linear discriminant analysis and logistic modeling techniques. Furthermore, there has been an emphasis on the use of provocative laboratory tests, rather than characteristics of the individual, as predictors of sickness. Finally, it does not appear that any attempt has been made to model sickness as measured by the SSQ score.
Thus, prediction of sickness (as defined by the SSQ) using linear regression techniques and based only on characteristics of the individual appears to be a unique contribution to the literature. Furthermore, this research likely represents one of the first attempts at investigating the prediction of sickness as it occurs in virtual environments.