Visuospatial Training Does Not Reduce Motion Sickness Susceptibility

An a priori power analysis using G*Power (Faul et al., 2007) showed that to detect a difference between the visuospatial-training group and the no-training group in the reduction in motion sickness from the first simulated drive to the second simulated drive of size η_p² = .15 given α = β = .05, N = 78 participants were needed. Participants were recruited on campus at Heinrich Heine University Düsseldorf. Our goal was to recruit about 100 participants to still have at least the required sample size available in case a substantial dropout due to motion sickness occured. We were able to recruit 111 participants in the period during which we had a laboratory at our disposal, but 6 participants decided to terminate the experiment prematurely or did not show up for the second drive. The data of 14 of the remaining participants could not be analyzed because they did not fill out the Mental Rotation Test properly (they systematically selected only one of the two correct solutions) or because they were given the wrong form of the Mental Rotation Test (for instance, Form A before the second drive when they had already filled out Form A before the first drive). With data of N = 91 participants remaining (n₁ = 46 in the visuospatial-training group, n₂ = 45 in the no-training group), the sensitivity of the critical statistical test still was even better than we had planned. All other things being equal, we were able to detect a difference of size η_p² = .13 between the visuospatial-training group and the no-training group in the change in motion sickness from the first to the second drive.

The median age of the 67 female and 24 male participants was 22 years (between 18 and 54 years). Most of them (80) were right-handed.

Participants gave written informed consent prior to participation. They knew that they could terminate their participation in the experiment at any time without any negative consequences for them. The experiments were conducted in line with the Declaration of Helsinki. Approval for the experiment was obtained from the ethics committee of the Faculty of Mathematics and Natural Sciences at Heinrich Heine University Düsseldorf (Case ID ZI01-2021-02).

Every potential participant first read a description of the study and the consent form which they signed if they wished to participate. Participants were randomly assigned either to the visuospatial-training group or to the no-training control group. Their self-reported age and gender and their measured handedness (Oldfield, 1971) were recorded and a private code was generated such that data from the first and the second drive and from the various questionnaire measures could later be matched without reference to the participants’ identity.

Next, participants received Form A or Form B (randomly determined) of the Mental Rotation Test (Peters et al., 1995; Vandenberg & Kuse, 1978). Subsequently, participants filled out the Simulator Sickness Questionnaire (Kennedy et al., 1993). Participants then received an introduction to the virtual reality equipment after which their first virtual 30-minute drive began.

During the drive, participants were sitting on a chair wearing an HTC Vive Pro headset (with a 1400 × 1600 pixels OLED display per eye and a 90 Hz refresh rate) and holding a Vive motion controller in their right hand. The experiment was controlled by an Intel®️ i7-based PC with an NVIDIA®️ RTX 3070 graphics card. The virtual environment was rendered using a custom-made program created in Unreal Engine 4.26 and extensively optimized to run smoothly at 90 frames per second. Head and hand movements were tracked via the headset and the motion controller using the standard SteamVR tracking system.

The drive covered 8.5 km of curvy rural roads with altitude differences of 250 m (15 minutes) followed by about 3.5 km of urban streets with numerous 90-degree turns (15 minutes, Figure 1). The virtual environment was designed to give the impression of sitting in the front passenger’s seat of an autonomous left-hand drive passenger car and included features to make this experience convincing such as suitable steering wheel rotations. Once every minute a tone signaled to the participants that they had to rate their motion sickness symptoms using the Fast Motion Sickness Scale (Keshavarz & Hecht, 2011) by selecting, with their index finger, a value between 0 (no sickness at all) and 20 (frank sickness) on a scale displayed on the car’s dashboard. After their first response to the Fast Motion Sickness Scale, participants initiated the drive by pressing a start-stop button displayed in the center console of their virtual car. Participants knew that this button could also be used to terminate the drive, for instance in case they experienced motion sickness symptoms beyond the level they were willing to tolerate. Given this, we did not expect motion-sickness symptoms as serious as vomiting. Nevertheless, a bucket was placed to the left of the participants. This bucket was tracked and virtually rendered—it appeared to be placed on the driver’s seat—so that participants could locate and use it quickly without needing to remove the headset. None of the participants had to use the bucket.

Immediately after the first drive, participants again filled out the Simulator Sickness Questionnaire. This segment of the experiment took about 45 minutes.

Participants in the visuospatial-training group, but not participants in the no-training control group, subsequently received 14 emails, each on one of the 14 days following the day of the first drive. Each email contained a set of visuospatial problems (the problems listed in Table 1 of Smyth et al., 2021, but translated into German, typeset and edited by the present authors) together with the instruction, parallel to that used by Smyth et al. (2021), to work on the set on the day it had been received for 15 minutes; if participants needed less time to find the solutions to the visuospatial problems they were to fill the remaining time studying their solutions, as in the original study. Participants were instructed to turn in the completed sets with their solutions by email on the same day on which they had solved the problems. The problems of Set 14 were to be solved on the day of the second drive before showing up in the laboratory. Of the 46 (participants) × 14 (days) = 644 sets handed out to the participants, 2 (.003 %) were not turned in at all and 138 (21 %) were turned in on an incorrect day (usually a day too late and then together with the set of that day).

Participants returned to the laboratory 14 days after their first drive. If an interval of exactly 14 days was not feasible for a particular participant, then an appointment was made for the weekday following the day that had originally been planned for that participant. The sequence of events was identical to that of the first drive with two exceptions. First, participants started immediately by filling out the Mental Rotation Test (Form B if they had received Form A before or Form A if they had received Form B before). Second, at the end of the experiment participants were informed about the purpose of the study and were either paid (90 € if they were in the visuospatial-training group and 20 € if they were in the no-training control group) or received course credit for their participation. This segment of the experiment also took about 45 minutes.

The results of the Mental Rotation Test are displayed in Figure 2. A 2 × 2 analysis of variance with test occasion (before the first drive vs. before the second drive) as within-subject independent variable and training (visuospatial training vs. no training) as between-subjects independent variable showed a main effect of test occasion, F(1, 89) = 44.733, p < .001, η_p² = .334, no main effect of training, F(1, 89) = 0.352, p = .554, η_p² = .004, and no interaction between these variables, F(1, 89) = 0.541, p = .464, η_p² = .006. Thus, there was the expected practice effect in that participants improved substantially from the first to the second time they took one of the two versions of the Mental Rotation Test (Peters et al., 1995) but the critical interaction between the test-occasion variable and the training variable was not significant: The improvement in the Mental Rotation Test score was only descriptively, but not significantly, larger in the visuospatial-training group than in the no-training group. Unfortunately, we cannot compare this statistical test result to the corresponding result from Smyth et al. (2021) because they did not report the critical statistical interaction test. However, a comparison of the results at a descriptive level can be done and is in fact quite instructive. We found that the visuospatial-training group and the no-training group improved from the first test occasion before the first drive to the second test occasion before the second drive by 49.8 % and 35.4 %, respectively. Smyth et al. (2021, p. p. 5) reported improvements of 45.8 % and 29.2 % for their visuospatial-training group and their no-training group, respectively. Given the striking similarity of these data combined with the fact that Smyth et al. (2021) had only n₁ = 15 and n₂ = 7 participants in their visuospatial-training group and their no-training group, respectively (and, thus, very little statistical power in their test of an interaction between the test occasion and training variables), it seems safe to extrapolate that the non-reported critical interaction based on the data of Smyth et al. (2021) was most likely not statistically significant.

An obvious question is why there was not more of a training effect on the Mental Rotation Test score in the visuospatial-training group apart from the to-be-expected practice effect itself (Peters et al., 1995). This question will be dealt with in the Discussion section. To anticipate, it seems intuitively plausible that it should be quite difficult to augment an effect as large as the practice effect observed with the Mental Rotation Test. Also, a closer look at the visuospatial training problems used by Smyth et al. (2021) as well as here reveals that there is surprisingly little overlap between what is practiced in these visuospatial training problems and what the Mental Rotation Test measures.

Next, we analyzed whether participants developed motion sickness during the simulated drives. For that purpose, the total score of the Simulator Sickness Questionnaire before each drive was subtracted from the total score of the Simulator Sickness Questionnaire after each drive. Positive differences indicate an increase in motion sickness during the drive. Figure 3 shows, and a statistical test confirms, that collectively these differences were significantly above zero, F(1, 89) = 70.392, p < .001, η_p² = .442. In other words, participants developed motion sickness during the simulated drives. A 2 × 2 analysis of variance with test occasion (during the first drive vs. during the second drive) as within-subject independent variable and training (visuospatial training vs. no training) as between-subjects independent variable showed a main effect of test occasion, F(1, 89) = 19.392, p < .001, η_p² = .179, no main effect of training, F(1, 89) = 2.052, p = .156, η_p² = .023, and, importantly, no interaction between these variables, F(1, 89) = 0.016, p = .900, η_p² < .001.

In addition, we analyzed participants’ responses to the Fast Motion Sickness Scale. The average Fast Motion Sickness Scale score is presented in Figure 4. A 2 × 2 analysis of variance with test occasion (during the first drive vs. during the second drive) as within-subject independent variable and training (visuospatial training vs. no training) as between-subjects independent variable showed a main effect of drive, F(1, 89) = 44.426, p < .001, η_p² = .333, no main effect of training, F(1, 89) = 0.414, p = .522, η_p² = .005, and, importantly, no interaction between these variables, F(1, 89) = 1.349, p = .249, η_p² = .015. These results consistently show that participants habituated considerably to the motion sickness inducing drive in the virtual reality environment from the first drive to the second drive. However, this habituation was not modulated by whether participants had performed 14 days of visuospatial training or not.

The data of the experiment and the virtual scenario as well as supplementary Bayesian analyses pertaining to the interactions between test occasion (first drive vs. second drive) as within-subject independent variable and training (visuospatial training vs. no training) as between-subjects independent variable for each of the three dependent variables considered above are available at the OSF project page (see the Data Acessibility Statement at the end of the article). In essence, the results of the Bayesian analyses amount to clear evidence in favor of the null hypothesis of no interaction between test occasion and training with respect to the Mental Rotation Test score, BF₀₁ = 11.131, the total score of the Simulator Sickness Questionnaire, BF₀₁ = 6.747, and the Fast Motion Sickness Scale score, BF₀₁ = 7.075.

The results of the present conceptual replication of the study by Smyth et al. (2021) are unequivocal. Whereas both the typical practice effect in the Mental Rotation Test (Peters et al., 1995) and the typical habituation of the susceptibility to motion sickness (Keshavarz & Golding, 2022) were observed, none of the key findings of Smyth et al. (2021) were replicated: The visuospatial training affected neither the Mental Rotation Test score nor the susceptibility to motion sickness. This result is remarkable considering that the sample size and, thus, the sensitivity of the statistical tests was much larger in the present study (N = 91) than in that of Smyth et al. (2021) (N = 22).

The failure to replicate the key findings of Smyth et al. (2021) must be seen in the context of several unexpected features of the data they reported, as discussed in the introduction. First, Smyth et al. (2021) reported an improvement of 45.8 % and 29.2 % in their visuospatial-training group and their no-training control group, respectively, but they did not report the critical statistical test of whether the difference in improvement between groups is statistically significant. Our analysis of this pattern (presented in the Results section) has demonstrated that this difference was most likely not statistically significant. Second, in the visuospatial-training group the effect of the visuospatial training was smaller on the visuospatial performance itself (a 45.8 % increase) than on motion sickness (a 58.5 % decrease). It is unexpected that the effect of training an ability was smaller on the trained ability itself than on another, less directly related variable. Third, Smyth et al. (2021) had only n₁ = 15 participants in the visuospatial-training group and even only n₂ = 7 participants in the no-training control group in the critical part of their study. It is well-known that the probability of obtaining sample statistics that differ considerably from the underlying population parameters is larger with such a small sample than with larger samples. In addition, no justification was presented as to why an element of inefficiency was introduced into the design by having less than half the number of participants in the no-training control group relative to the number of participants in the visuospatial-training group. Fourth, there is no reason as to why visuospatial skill should be related, let alone causally related, to motion sickness. Smyth et al. (2021, p. p. 9) have conceded this fact which means that there is currently no reason to assume that manipulations of the cognitive skills needed for mental rotation should have an effect on the susceptibility to motion sickness. The present data are compatible with the conclusion that this effect does not exist.

A possible problem common to the study reported here and that of Smyth et al. (2021) is that the improvement in the Mental Rotation Test was only descriptively, but not significantly larger in the visuospatial-training group relative to the no-training group. As mentioned above, in the Mental Rotation Test “practice effects are dramatic” (Peters et al., 1995, p. 35) both in general and in the data reported here. It seems intuitively plausible that it should be quite difficult to augment an effect that is already dramatic. What is more, the Mental Rotation Test consists of variants of the rotation problems used by Shepard and Metzler (1971), that is, of two-dimensional presentations of three-dimensional objects composed of ‘wireframe’ cubes that need to be rotated and then matched to other two-dimensional presentations of three-dimensional objects composed of ‘wireframe’ cubes. It is well-known that mentally rotating such objects can be improved when the training task involves rotating these very same objects (e.g., Heil et al., 1998; Jost & Jansen, 2021; Wiedenbauer et al., 2007; Wiedenbauer & Jansen-Osmann, 2008). Importantly, a closer look at the training problem sets used by Smyth et al. (2021, see their Table 1) and, therefore, in the present conceptual replication study reveals that none of these problems resemble the problems used in the Mental Rotation Test. In fact, six problem sets (Sets 1, 2, 6, 9, 12 and 13) do not even contain a single three-dimensional problem; the problems in these sets are exclusively two-dimensional. Two more sets (Sets 5 and 11) consist of both two-dimensional and three-dimensional problems and six sets (Sets 3, 4, 7, 8, 10 and 14) consist of three-dimensional problems. In other words, the mental routines necessary to solve the training problems overlap only partially with the mental routines necessary to solve the problems of which the Mental Rotation Test consists. It should thus not come as a surprise that there is very little positive transfer from these training problems to the Mental Rotation Test. In sum, then, the fact that the so-called visuospatial training did not affect performance in the Mental Rotation Test more than it did is not surprising at all but rather is to be expected.

A deviation of the present study from that of Smyth et al. (2021) is that the drives occurred in a virtual-reality environment whereas the participants in the original study were driven around. However, from a pragmatic point of view it seems important to note that the critical construct here is motion sickness and, as the data reported above show, participants developed substantial motion sickness. From a theoretical point of view, sensory conflict theory implies that motion sickness is primarily caused by a mismatch between the expected and the actual interplay among the visual, vestibular and somatosensory senses (Cha et al., 2021; Keshavarz & Golding, 2022). Simply put, in the driving task used by Smyth et al. (2021) in which participants were driven around while reading text in the backseat of a car, the visual signal implied little motion of the body whereas the vestibular and somatosensory signals implied motion, creating a mismatch compared to the normal interplay between these senses. In the simulated drive in a virtual environment used here, the vestibular and somatosensory signals implied less motion of the body than the visual signal, again creating a mismatch compared to the normal interplay between these senses. Thus, the crucial mismatch occurs in both types of situations and, expectedly, the motion sickness symptoms occurring in both types of situations are essentially the same (Cha et al., 2021). The sensory mismatch theory also implies that motion sickness should be reduced as a function of repeated exposure due to habituation, that is, because “the internal model of expected input recalibrates” (Cha et al., 2021, p. 329) and the organism learns, within the limits of its adaptive capacity, to expect the previously unexpected interplay among the visual, vestibular and somatosensory senses.

In sum, Smyth et al. (2021, p. p. 9) have conceded that there is no reason why manipulating the amount of practice in solving two-dimensional and three-dimensional visuospatial matching problems should affect the susceptibility to motion sickness, and we did not observe such an effect despite the fact that the sample size and, hence, the sensitivity of the statistical tests in the present study were much larger than their counterparts in the study by Smyth et al. (2021). Given this, it is probably prudent to assume that the effect does not exist and to continue to rely on the conclusion reached by Keshavarz and Golding (2022, p. p. 107) that “habituation remains the most effective nonpharmacological method to reduce motion sickness”.

Contributed to conception and design: MW, EZ, ES, AB

Contributed to acquisition of data: MW, AB

Contributed to analysis and interpretation of data: MW, EZ, ES, AB

Drafted and/or revised the article: MW, EZ, ES, AB

Approved the submitted version for publication: MW, EZ, ES, AB

We have no competing interests to disclose.

The data of the experiment and the virtual scenario as well as the results of the supplementary Bayesian analyses are available at https://osf.io/4deuq/?view_only=baf13d4ece2d4949b9bea310274280da