- Email: [email protected]
Quantitative analysis of time-course development of motion sickness caused by in-vehicle video watching
Displays 35 (2014) 90–97
Contents lists available at ScienceDirect
Displays journal homepage: www.elsevier.com/locate/displa
Quantitative analysis of time-course development of motion sickness caused by in-vehicle video watching Naoki Isu a,⇑, Takuya Hasegawa b,1, Ichiro Takeuchi b, Akihiro Morimoto a,2 a b
Faculty of Engineering, Mie University, 1577 Kurima-machiya, Tsu, Mie 514-8507, Japan Faculty of Engineering, Nagoya Institute of Technology, Gokiso-cho, Showa-ku, Nagoya, Aichi 466-8555, Japan
a r t i c l e
i n f o
Article history: Received 4 June 2012 Received in revised form 21 March 2013 Accepted 19 January 2014 Available online 28 January 2014 Keywords: Visual–vestibular sensory conflict Susceptibility Statistical analysis Random-component location-scale model Incomplete within-subject design Onboard TV
a b s t r a c t It has been common and popular to watch videos in moving vehicles. An important issue in developing comfortable in-vehicle video watching systems is to understand how passengers get motion sickness. With this in mind, the goals of this paper are (1) to introduce an experimental protocol and a statistical analysis procedure for quantitatively evaluating how motion-sickness is developed during car-driving, and (2) to demonstrate their practical usefulness with a working experimental study. In the experimental protocol, motion sickness was induced to subjects by requiring them to watch an in-vehicle video during 15-min driving, and the time-course development of motion sickness was recorded by asking subjects to evaluate their degree of motion sickness every one minute. A main difficulty in analyzing data from these studies is how to incorporate the individual difference in motion-sickness susceptibility. Since susceptibilities are markedly different among subjects, within-subject design experiments are preferred. However, it is practically difficult to conduct complete set of trials because subjects who are not willing to continue experiment (due to heavy motion sickness) should be able to withdraw from the subsequent series of trials in accordance with ethical requirement. To cope with this incomplete data issue, we introduce a statistical data analysis procedure that enables to estimate and impute missing entries in the within-subject design table. Using a working example, we demonstrated that the protocol and the procedure are useful for quantitative assessment of the time-course motion sickness development. We conducted an in-vehicle motion-sickness study with 31 subjects, where the time-course motion sickness developments of video-watching, book-reading, and normal riding conditions were compared. The results indicate that video-watching brings on 2.7 times more severe motion sickness than normal riding, but 25% less severe than book-reading. Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction Recent advancement of video-screen technology allows us to watch videos in moving vehicles such as cars, boats, and airplanes. An important issue in the development of in-vehicle video systems is to understand how easily and severely passengers get motion sickness. It is well known that reading books or maps in a moving vehicle brings on and aggravates motion sickness. This is known to be the result of sensory conflict [1], that is, the vestibular system conveys sensory information of body movement, while the visual system conveys stationary information creating the sensation of the body being immobile. ⇑ Corresponding author. Tel.: +81 59 231 9458; fax: +81 59 231 9460. E-mail address: [email protected] (N. Isu). Present address: DENSO Corporation, 1-1 Showa-cho, Kariya 448-8611, Japan. Present address: Automotive Systems Company, Panasonic Corporation, 4261 Ikonobe-cho, Tsuduki-ku, Yokohama, Kanagawa 225-8520, Japan. 1 2
http://dx.doi.org/10.1016/j.displa.2014.01.003 0141-9382/Ó 2014 Elsevier B.V. All rights reserved.
Watching a video in a moving vehicle is considered to trigger sensory conflict in the same manner as book-reading [2]. Schoettle and Sivak [3] investigated the frequency and the severity of motion sickness experienced from in-vehicle video usage by a questionnaire survey. The study indicated that video watching in a moving vehicle caused motion sickness less often than book-reading, but more often than normal riding. Kato and Kitazaki [4,5] also investigated the influence of in-vehicle video display on carsickness by their experimental studies. It showed that reading still images (news text) or watching moving images brought on more severe carsick than the case that passengers were allowed to see external view. In order to offer comfortable video-watching in moving vehicles, it has been desired to develop new video systems that can reduce motion sickness. For developing such new video systems, it is essential to establish an experimental protocol that can quantitatively evaluate how motion-sickness is developed during
N. Isu et al. / Displays 35 (2014) 90–97
car-driving. To assess the severity of sickness in series during an experimental trial, it must be briefly evaluated in an instant at every measuring time. A variety of subjective rating methods [e.g., 4–7] and magnitude estimation methods [8] have been frequently used for that purpose. Short Symptoms Checklist [9] is also adopted for evaluating the severity of symptoms at given intervals (e.g., every 5 min) during stimulus periods. Another estimation method, in which subjects respond with a joystick instead of replying with numerical ratings, has been used for continuous measurement [10]. In this paper, we first introduce an experimental protocol and a statistical analysis procedure for quantitative assessment of time-course motion sickness development. Then we demonstrate with a working experimental study that our new approach is practically useful for understanding how motion-sickness is developed during car-driving. In the experimental protocol, motion sickness is induced to subjects by requiring them to watch a video via an in-vehicle video display during 15-min driving along a winding road. The timecourse development of motion sickness is recorded by asking subjects to evaluate their degree of motion sickness every one minute of 15-min driving. The degree of motion sickness is subjectively evaluated with 11 grades (0–10) on a rating scale, which is essentially the same as the well-being score introduced by Reason and Graybiel [11]. One of the difficulties in experimental studies on motion sickness is how to incorporate the individual difference in motion sickness susceptibility which differs markedly among individual subjects. When different stimulus conditions are compared such as video-watching, book-reading, and normal-riding, subjects who have similar susceptibilities should be evenly assigned to each stimulus condition. In some previous studies (e.g., [4–6]), individual susceptibilities are scored by questionnaire such as the Motion Sickness Susceptibility Questionnaire (MSSQ) [12,13] in advance, and the subjects were evenly assigned to each stimulus condition based on the score. However, since motion-sickness susceptibilities are markedly different among individuals, stratification based on questionnaire score is not reliable enough, and the results highly depend on subject assignments. Therefore, it is desirable that every stimulus condition is evenly tested by each subject (within-subject design). In practice, however, it is difficult to carry out the complete set of trials, i.e., same number of trials with every stimulus condition by each subject. It is because subjects are allowed to withdraw from the subsequent series of trials in accordance with ethical requirement (subjects who had severe motion sickness in the previous trial would not be willing to try it again). As the result, the within-subject design table inevitably has several missing entries. Even worse, those missing entries arise with bias because subjects who have more severe motion sickness tend to withdraw more often than those who have no motion sickness. This kind of practical problem is a formidable issue more or less common to experimental motion-sickness study. To handle this incomplete data issue, we introduce a simple statistical model for time-course development of motion sickness. This approach allows us to estimate and impute the missing entries in the within-subject design table, and to compare the motion sickness severities among different stimulus conditions. In this paper, for the purpose of demonstration, we conducted an in-vehicle motion-sickness study with 31 subjects, where the time-course motion sickness developments of videowatching, book-reading, and normal riding conditions were compared. The conclusion of the analysis was consistent with previous studies [3–5], i.e., video watching brings on more severe motion sickness than normal riding, but is less severe than book-reading.
91
2. Experimental protocol In this section, we present the experimental protocol by describing the experimental setup of our working example study on time-course motion sickness. 2.1. Subjects N = 31 subjects (21 males and 10 females), approximately 20 years of age, participated in the experiment. None of the subjects reported any medical problems and all presented with normal vestibular functions. Every subject was briefed on the purpose, procedures, risks, and benefits of the study, and gave written informed consent before participating in the experiment. It was also explained that the subjects could withdraw from the experiment whenever they requested, that is, when they could no longer endure the nausea of motion sickness, they could immediately get out of the car and terminate the experimental trial. The experimental protocol used in this study was approved by the Ethics Committee of the Faculty of Engineering, Mie University. 2.2. Apparatus The experiment was performed using a minivan, which had three rows of seats and a boarding capacity of 7 persons. Subjects, one or two in each experimental trial, were seated in the second row of the car. Two in-vehicle video displays were installed separately behind head rests of the first row of the seats by means of the arms fixed to the side roof. They were located about 60 cm apart in front of the individual subjects. The displays had an 11 inch-type wide LCD panel, 244 mm (width) 138 mm (height), whose resolution was 800 (horizontal) 480 (vertical) pixels. The horizontal and vertical viewing angles for the subjects were about 23 and 13 degrees, respectively. Movies were played on the in-vehicle video displays using a DVD player. Sound was provided by speakers attached on the side doors through an FM radio. 2.3. Stimuli (riding conditions and driving course) The experiment was performed to compare the effect of riding conditions on motion sickness. Subjects rode in the car for T = 15 min under one of the following K = 3 conditions: (1) videowatching, (2) book-reading, and (3) no-task conditions. Under the video-watching condition (k = 1), subjects were required to watch a movie shown on the in-vehicle video displays. The movies used in the experiment did not include any scenes of an intense nature or uncomfortable scenes. They were selected from among romantic and family comedies (‘‘Legally Blonde’’, etc.) that had not been watched by the subjects, so that the subjects could concentrate their attention on the scene. The sound was provided in the native language of the subjects (i.e., Japanese) and subtitles were not superimposed. Under the book-reading condition (k = 2), subjects were required to read a picture book (‘‘Where’s Wally?’’) that had few letters but lots of small fine figures. The task for readers was to look for one or several particular figures which got mixed in among lots of similar ones on each page. In the notask condition (k = 3), subjects were simply asked to remain still and quiet. They were allowed to look outside of the car freely. To ensure the subjects obeying the instructions, their behavior was observed by an experimenter seated in the third row of the car. The driving course used in the experiment was a suburban road with numerous curves and occasional gentle slopes. It had no traffic signals but two stops. There were no houses along the road, and the traffic, if any, was not heavy. The driving course was a 2.1-km circuit, and it took 3 min to drive around. The car went around the course 5
92
N. Isu et al. / Displays 35 (2014) 90–97
times for a total of 15 min in each experimental trial, and was driven by a single driver in order to minimize the variance in driving manner throughout the whole experiment. The driver was instructed to drive along the traffic lane on the road as accurately as possible while keeping a constant speed under the legal limit (50 km/h), and the experimental design and details were not informed to the driver. The driving course had 19 sharp curves in one circuit and the maximal angular velocities of yaw in the individual curves were around 8–35 deg/s (mean ± SD: 14.8 ± 7.2 deg/s). 2.4. Measurement of motion sickness severity During a T = 15-min trial, subjects required to subjectively evaluate their degree of motion sickness in a rating-scale such as wellbeing scores [11]. They were asked to answer it orally according to the experimenter’s request every one minute. The subjective severity was expressed in 11 numerical categories (severity scores) from 0 (normal condition with no sickness) to 10 (limit of subject endurance of severe nausea: endpoint). The subjects were instructed to assign one of the numerical categories as a severity score in proportion to the degree of motion sickness severity. When subjects withdrew in the middle of a trial due to severe sickness, severity scores for the rest of the trial were regarded as the maximal point of 10. 2.5. Within-subject-design and drop-outs As mentioned in Section 1, drop-outs of some subjects are inevitable in within-subject-design motion-sickness studies. Also in our working study, scheduled trials were sometimes canceled since the subjects were free to withdraw from the subsequent series of trials in accordance with ethical requirement. Consequently, 11 subjects withdrew after trials in a single condition, and 13 subjects attended trials in two of the three conditions before their withdrawal. Seven subjects attended trials in the all conditions. Excluding trials by the subjects participated only in a single condition, experimental records of 20 subjects were used for subsequent statistical analysis. They participated in 23, 24 and 34 trials in the video-watching, book-reading and no-task conditions, respectively. It means that our within-subject design table from the 20 subjects contains several missing entries. Those missing entries are estimated and imputed by the statistical analysis described in the following Chapter 3. Every subject participated in the experiment at least one day apart. One or two subjects attended each experimental trial. They were seated in the second row of the minivan. After a dozen of minutes driving (without any task) to get to the starting point of the course, subjects had a rest for 10 min before starting the trial. One of the above-mentioned tasks was assigned to subjects during a 15-min riding period. The subjective severity of motion sickness was recorded at the starting time and every one minute during 15min driving. When subjects felt severe nausea and desired to withdraw, they could get out of the car and discontinue the trial at any time. The numbers of trials terminated in this way were 3, 3 and 0 in the video-watching, book-reading and no-task conditions, respectively. Only in one case out of the 6 terminated trials, two subjects participated in the experiment together and one of them withdrew from it. We restarted the trial with the other subject after letting the former off. The interruption of the driving was shorter than 30 s. 3. Statistical analysis procedure In this section, we present the statistical analysis procedure for analyzing the data obtained from experiments such as the working example in the previous section.
Since individual difference of motion-sickness susceptibilities are very large, within-subject design is best suitable for these studies. However, planned trials are often canceled because subjects should be allowed to drop off from the experiment at any stage in accordance with ethical requirement. Accordingly, the following difficult issues must be addressed in statistical analysis. (1) First, the procedure must have a mechanism to impute the missing entries in the within-subject design table. (2) Second, the procedure must be able to capture the timecourse development of motion sickness, in which both the average and the variability (variance) of the severities have increasing trends with riding time. To handle these issues all together, we introduce a simple statistical model, named a random-component location-scale model. The model allows us to estimate and impute the missing entries in the incomplete within-subject design table. It consists of two statistical models, a random-component model and a location-scale model. The former model is useful for longitudinal data analysis when heterogeneity among individuals is not negligible [14,15]. It contains a group of parameters which explains subject-wise motion sickness susceptibilities. Specifically, for each of N subjects, we have a parameter di, i = 1, ..., N. Large values in di indicate that the subject i tends to have severe motion sickness, while small values indicate that the subject i would not get motion sickness easily. The missing entries can be imputed by modeling individual effects as random components which are independent of stimulus condition effects. The latter model, also called as heteroscedastic model, is frequently used in econometrics [16]. This model contains another group of parameters which represent time-course effects of stimulus conditions. In particular, the average and the standard deviation of the motion sickness severity after t-minutes riding with condition k are expressed by the parameters lk,t and rk,t, respectively, where t = 1, ..., T, and k = 1, ..., K. It enables us to analyze time-course data whose variances as well as averages vary with the time by modeling them as functions of time. 3.1. Random-component location-scale model Let us denote yk,t,i be the severity score given by subject i after tminutes riding with condition k. Here, we use the subscript k e {1, , K} to distinguish K stimulus conditions. Similarly, we use the subscript t e {0,1, , T} to specify the riding time. The subscript i e {1,2, , N} represents the subject index. The random-component location-scale model is formulated as
yk;t;i ¼ lk;t þ rk;t di þ ek;t;i ;
ð1Þ
where lk,t and rk,t P 0 are the average (i.e., location) and the standard deviation (i.e., scale) of the motion sickness severity scores over subjects at riding time t with stimulus condition k, respectively. Furthermore, di denotes the random component indicating the susceptibility to motion sickness of subject i, and we assume, for identifiability, that fdi gNi¼1 have zero mean and unit variance. Finally, ek,t,i is the Normally distributed error term independently and identically distributed for all (k, t, i) with the mean zero. The main goal of the analysis is to estimate the average flk;t gTt¼0 and variability frk;t gTt¼0 of the motion sickness severities for each stimulus condition k e {1, , K}. In the model (Eq. (1)) we do not assume any functional relationship between the average severities lk,t and the riding time t because the scores are ordinal scale (not distance scale). Instead, we assume that the average severities are monotonically increasing along the riding time t, i.e.,
0 lk;0 lk;1 lk;T 10; for all k:
ð2Þ
N. Isu et al. / Displays 35 (2014) 90–97
In addition, we introduce the restrictions on the lower and the upper bounds of the expected severities:
0 lk;t þ rk;t di 10; for all k; t; and i
ð3Þ
because the range of the severity scores is limited to [0, 10]. Note that the motion sickness susceptibility di is independent of stimulus condition k. It roughly implies that a subject who is prone to get severe motion sickness in one stimulus condition tends to have severe motion sickness in other conditions as well. This separability of the susceptibility parameters di and stimulus condition parameters (lk,t, rk,t) enables us to impute the missing entries in our within-subject design table. Namely, hypothetical motion sickness trend for a subject i with stimulus condition k can be estin oT lk;t ; rk;t into the randommated by substituting di and t¼1
component location-scale model. The parameters of the model are estimated using least-squares method (i.e., maximum likelihood estimation under the Gaussian noise). The sum of the squared errors between the observed and the estimated severity scores are minimized under the aforementioned constrains (Eqs. (2) and (3)). The problem is mathematically formulated as
X n
o2
93
sampling. A hypothetical experimental data under the null hypothesis is generated by randomly permuting the stimulus condition T index of yk;t;i t¼1 between k and k0 as if they were gotten in either condition of the two jumbled together. Then we fit the randomcomponent location-scale model (Eq. (1)) with the randomized hypothetical data and compute the test statistic in Eq. (6). By repeating the randomized sampling process many times (10,000 times in our analysis), we obtain the null distribution of the test statistic, and p-values can be simply computed by
p :¼
B 1X ðbÞ I Dsk;k0 Dsk;k0 ; B b¼1
ð7Þ
where B is the number of randomized sampling (B = 10,000 in the ðbÞ current analysis), Dsk;k0 is the test static computed from the b-th randomized data, and I is the index function defined as
IðzÞ ¼
1; if z is true; 0; if z is false:
ð8Þ
4. Results
ð4aÞ
In this section, we first report the experimental results obtained from the working experiment, and then show the results of the statistical analysis based on our new model-based analysis.
subject to 0 lk;0 lk;1 lk;T 10; for all k;
ð4bÞ
4.1. Subjective ratings of motion sickness severity
rk;t 0; for all k and t;
ð4cÞ
0 lk;t þ rk;t di 10; for all k; t; and i;
ð4dÞ
N X di ¼ 0;
ð4eÞ
The subjective severity of motion sickness increased gradually during the car ride in every condition. The time-course development of motion sickness severity is shown in histograms of Fig. 1 at intervals of 3 min. The ordinate indicates the ratio of trials rather than frequency so that the histograms can be directly compared among the three conditions. The severity was 0 (no sickness) at the beginning of car-riding (0 min) in almost all trials. It gradually increased in most trials under the video-watching and book-reading conditions. The distribution of severity was shifted toward higher scores along with the riding time. The maximal severity reached 10 (endpoint) at 13 min with video-watching, while at 7 min with book-reading. In about 10% of trials under both conditions, the severity reached 10 in the middle of trials. On the other hand, the severity remained at 0 (no sickness) in about 40% of trials under the no-task condition. It mostly remained below a level of mild discomfort and never exceeded 6 through the entire period of car-riding in all trials. The first increase of severity score (from 0 to 1 in most trials) was observed before 6 min in over 80% of trials with video-watching and those with book-reading, while it took more than 7 min in half of trials with no-task. Histograms of the onset times are shown in Fig. 2a. Since the onset times under each condition had an asymmetrical distribution that extended toward higher values, their median rather than mean was obtained. The medians of the onset time for the no-task, video-watching, and book-reading conditions were 7.5, 3, and 3 min, respectively. It is known that some slight symptoms such as stomach awareness appear around a well-being score of 3–4 [11]. Since the severity score used in this study is equivalent to the well-being score of Reason and Graybiel [11], the instant of exceeding score 3 can be considered to indicate the onset time of slight symptoms. The onset times were distributed as shown in Fig. 2b. The medians were 9 and 8.5 min for the video-watching and book-reading conditions, respectively. The severity score did not exceed 3 in over a half of trials under the no-task condition so that the median of the onset was not obtained within 15 min of riding period. From the individual time-course development of severity reported in each trial, we observed that the severity scores, on aver-
min
flk;t ;rk;t g8k;8t ;fdi g8i
yk;t;i
lk;t þ rk;t di
ðk;t;iÞ2D
i¼1 N 1 X d2 ¼ 1; N 1 i¼1 i
ð4fÞ
where D in Eq. (4a) represents the set of all possible triplets (k, t, i). The optimization problem is solved by a simple iterative procedure using Matlab CVX package (http://cvxr.com/cvx/). The Matlab code is available by request. 3.2. Summary statistic of motion sickness and the statistical test After fitting the model and estimating these parameters, we compute severity index sk as a summary statistic of motion sickness severity with stimulus condition k. It is defined as
sk :¼
T 1X l^ k;t ; T t¼1
ð5Þ
^ k;t is the estimated value of lk,t and T is the total riding time. where l We examine the statistical significance of the difference in the severity indices among different stimulus conditions. Consider testing the difference between two stimulus conditions k and k0 . The test statistic is simply defined as
Dsk;k0 :¼ sk0 sk :
ð6Þ
Here, the null hypothesis is that there is no difference in motion sickness between the conditions k and k0 , and the alternative hypothesis is that the stimulus condition k0 yields more severe sickness than the condition k. To compute the statistical significance (p-values), we use permutation-based approach [17]. Permutation-based approach is a general procedure to compute null distribution and statistical significance based on randomized
N. Isu et al. / Displays 35 (2014) 90–97
Video-Watching Book-Reading No-Task
60 40 20 0
0
2
4
6
8
80 60 40 20 0
10
0
2
9 min
60 40 20 0
0
2
4
6
4
6
8
60 40 20 0
10
6 min
80
0
2
Severity Score R tio of Trials [% Rat %]
R tio of Trials [% Rat %]
Severity Score
8
40 30 20 10 0
10
12 min
50
0
2
4
6
6
8
10
8
10
15 min
50 40 30 20 10 0
0
2
Severity Score
Severity Score
4
Severity Score R tio of Trials [% Rat %]
80
3 min
100
R tio of Trials [% Rat %]
0 min
100
R tio of Trials [% Rat %]
R tio of Trials [% Rat %]
94
4
6
8
10
Severity Score
Fig. 1. Histograms of severity scores of motion sickness developing with riding time. Histograms of severity scores are shown at intervals of 3 min for the three riding conditions: video-watching, book-reading, and no-task conditions. The abscissa indicates the severity scores, and the ordinate indicates the ratio of trials rather than frequency.
R tio of Trials [% Rat %]
(a)
Onset time of the 1st. increase of severity
60 50
Video- Watching Book-Reading No-Task
40 30 20 10 0
0
3
6
9
12
15
0
As summary indices of motion sickness severity induced by individual trials, accumulated scores were calculated by adding severity scores for 15 min. Accumulated scores under the videowatching condition were distributed from 0 to 75, and those under the book-reading condition spread more widely from 6 to 114 as shown in Fig. 3. In contrast, those under the no-task condition were remarkably lower than the other conditions; they were 0 in a third of trials, and even the maximal score was 36. The medians for the no-task, video-watching, and book-reading conditions were 6.5, 27, and 34.5, respectively. These results indicate that video-watching and book-reading aggravate carsickness compared with ordinary ride with no task, and the former seems to affect it less than the latter.
Onset Time [min] 4.2. Statistical model analysis of motion sickness severity development
Onset time of exceeding score 3
70
In order to explain the above observations of time-course development of carsickness severity, we use a random-component location-scale model that can describe the time-dependent changes of
R tio of Trials [% Rat %]
60
Video -Watching Book-Reading No-Task
50 40
40
30 20 10 0
0
3
6
9
12
15
0
Onset Time [min] Fig. 2. Histograms of onset times of the first severity increase and slight symptoms. (a) Onset times of the first increase of severity score (score 1 in most trials) are shown in histograms for the three riding conditions. (b) As the onset time of slight symptoms, elapsed times until the severity score exceeded 3 are shown in histograms. The ordinate indicates the ratio (%) of trials in which the onset times stood between a pair of elapsed times indicated at the divisions of the abscissa.
age, monotonically increased with riding time. In addition, the subject-wise variability also increased with riding time, namely, as time passed, motion sickness severities of some subjects got worse and worse, but those for some other subjects did not increase so rapidly.
R tio of Trials [% Rat %]
(b)
Video-Watching Book-Reading No-Task
30
20
10
0 0
15
30
45
60
75
90
105
120
Accululated Severity Score Fig. 3. Histograms of accumulated severity scores. Histograms of accumulated severity scores for each riding period are shown for the three conditions. In cases subjects withdrew in the middle of a trial, severity scores for the rest period were regarded as the maximal score of 10. The ordinate indicates the ratio (%) of trials in which the accumulated severity scores stood between a pair of scores indicated at the divisions of the abscissa, except that the accumulated scores were 0 in the leftmost section.
95
N. Isu et al. / Displays 35 (2014) 90–97
the mean and the variance of the motion sickness distribution. The model assumes monotonic increase of average severity along the riding time as formulated in Eq. (2), which was satisfied with the observations in each riding condition. The parameters of the model formulated in Eq. (1) were estimated by the least-squares method under the constrains given in Eq. (4). Fig. 4 shows the estimated motion sickness trends under the three conditions. The sequential lines indicate the progression of ^ k;t of motion sickness severity, and the verthe estimated average l tical lines attached topthe ffiffiffiffi average indicate the standard errors ^ k;t = N , where N = 20 is the sample size. The approximated by r severity increased almost in proportion to the time over the 15 min of car-riding with every condition. Motion sickness was the severest with the book-reading condition and the second severest with the video-watching condition at any point of the car ride. As a summary statistic of sickness severity, severity index sk defined in Eq. (5), i.e., average severity for 15 min of riding period, was calculated. The severity indices sk for the no-task, videowatching, and book-reading conditions were 0.80, 2.19, and 2.84, respectively; the ratio of these indices was 1:2.7:3.6 in the above order. Thus, it was shown that the motion sickness severity was increased about three times by watching a video and that the videowatching severity was one quarter less than that caused by reading a book. These results substantially agreed with those obtained with accumulated scores in the above section, although they made a slight difference in the ratio of severity among the conditions. To examine the statistical significance of the difference in the motion sickness trends among different stimulus conditions, permutation-based statistical significance tests were performed in accordance with the procedure described in the Section 3.2. The p-value of the difference between no-task and video-watching conditions was 0.003, that between no-task and book-reading conditions was 0.001, and that between video-watching and bookreading conditions was 0.064. Thus, the statistical tests corrobo-
7
Video-Watching Video Watching Book-Reading Book Reading No-Task No Task
6
5
rated highly significant differences in motion sickness aggravation by video-watching and book-reading compared with ordinary carriding. 4.3. Distribution of subject-wise motion sickness susceptibility Subject-wise motion sickness susceptibility di, which is another parameter of the model formulated in Eq. (1), was estimated to^ k;t and r ^ k;t in the above section. The distribution of gether with l the susceptibility is illustrated in a histogram of Fig. 5. It shows the numbers of subjects (in the ordinate) having the susceptibility values in the abscissa. The origin of the abscissa (di = 0) indicates the mean susceptibility among the subjects participated in this study. The susceptibility values are expressed taking their standard deviation as the unit (di = 1). The histogram indicates that the distribution of the individual susceptibility is widespread (rather than having a single peak like the Normal distribution). In addition, there seems to be a small group of individuals who have very high susceptibility (those with susceptibility >1.0). We can see its reflection on the histograms of severity scores in Fig. 1. The histograms at the last stages of the riding period (12 and 15 min) have small bumps at high severity scores, indicating that the group of subjects with high susceptibility got sever sickness. Histograms of accumulated scores and onset times failed to show any influence resulted from the dichotomous distribution of subject-wise susceptibility (see Figs. 2 and 3). 5. Discussion When the severity of motion sickness is compared among different stimulus conditions, every condition must be tested by subjects evenly balanced for their susceptibility to motion sickness. In general, however, it is hardly possible to balance the susceptibility evenly because it is markedly different among individual subjects. Although it is desirable to conduct within-subject design experiment in which every stimulus is evenly tested by each subject, balanced assignment would not always be carried out in practice because subjects must be free to withdraw from the experiment at any stage. The unbalanced susceptibility due to missing trials makes it difficult to compare the motion sickness severities among different stimulus conditions. The present study has provided a solution to this formidable problem by introducing a statistical model called a random-component location-scale model. The model (see Eq. (1)) has two sets of parameters: one for stimulus condition effects and another for
4 7 6
3
5 4
2
3 2
1
1 0 -2.4 -1.8 -1.2 -0.6
0
0
5
10
15
Riding Time (min) Fig. 4. Motion sickness trends under the three conditions. The sequential lines ^ k;t of motion sickness severity indicate the progression of the estimated average l under the three riding conditions. Vertical lines attached to the average indicate the pffiffiffiffi ^ k;t = N . standard errors approximated by r
0
0.6
1.2
1.8
2.4
3
Susceptibility Index δ i Fig. 5. Distribution of the individual susceptibility to motion sickness. Individual susceptibility to motion sickness was estimated by the random-component location-scale model on 20 subjects. The ordinate of the histogram indicates the number of subjects who had motion sickness susceptibility di plotted as the abscissa.
96
N. Isu et al. / Displays 35 (2014) 90–97
subject-wise motion sickness susceptibilities. Note that the latter susceptibility parameters are independent of the three stimulus conditions. This makes it possible to estimate the average severity scores (the first term in Eq. (1)) in each stimulus condition even though the susceptibility is not exactly balanced among the conditions. Furthermore, this model captures increasing trends of the mean and the variance of the motion sickness distribution. Note that a standard regression model can only describe the change of the mean and implicitly assumes that the variance is constant. The statistical model analysis introduced in this study provides a considerable advantage to quantitative and comparative studies on time-course development of motion sickness. Fig. 4 shows that the severity of motion sickness increases in proportion to the car-riding period. The same tendency was observed in previous studies; the severity gradually increased in proportion to the duration of horizontal body oscillation [18], optokinetic stimulation [19,20], or combined oscillation of body and vision [18] for 30 min. When intensive stimuli were imposed, the increase in severity slowed down after 10–15 min [7,21]. Since proportional relationships were recognized in the progression of motion sickness severity under all conditions in this study, the severity index defined by Eq. (5) was used to present the aggravating effect of a riding condition on motion sickness. As a result, then, watching a video in a moving vehicle aggravated motion sickness about three times as severely as the no-task condition. It was 25% less severe than that caused by reading a book. Substantially identical results were obtained by using the accumulated severity scores without analysis adopting the statistical model. The ratio of the medians of accumulated scores for the no-task, video-watching, and book-reading conditions was 1:4.2:5.1, which is somewhat higher than that obtained by the statistical analysis (i.e., 1:2.7:3.6). It could be considered that this difference mainly derived from distorted balance of the susceptibility among the conditions caused by missing trials. Since the statistical model analysis can compensate for the missing trials caused by withdrawal of subjects, it likely provides us with more reliable estimation. Now, let us consider similar experiments conducted by Kato and Kitazaki [4,5]. They balanced the subject-wise susceptibility to motion sickness by evenly grouping subjects for each stimulus in accordance with MSSQ. Each group consisted of 17–20 subjects and was involved in a single stimulus. Although they did not describe the severity difference quantitatively in text, their illustrations indicate that watching moving images on an in-vehicle video display induced carsickness 2.9 times as severely as an ordinary ride having external view, and that reading still images (two lines of 15 characters) on the display induced sickness of 3.3 times severity. Their results were quite similar to the present statistical analysis, although we need to notice that their stimulus conditions were not identical with those in the present experiment. Here, note that the number of subjects could be saved in the present experiment thanks to the statistical analysis while they needed a fairly large number of subjects in order to balance the susceptibility among the stimulus conditions. Subject-wise motion sickness susceptibility showed dichotomous distribution in the present study (see Fig. 5). This differs from a simple distribution of motion sickness susceptibility investigated with MSSQ [12,13,22]. However, note that the histogram of Fig. 5 shows the distribution among the subjects participated in the present experiment voluntarily. It is known that higher susceptible individuals have less interest to attend experiments concerning motion sickness [22]. Therefore it is not likely that the histogram represents the susceptibility distribution of the whole people in general. Since subjects cannot be randomly selected from the population, it is difficult to investigate the general distribution of motion sickness in experimental studies.
It has been shown in previous studies that a passenger’s vision affects motion sickness to a great extent. Turner and Griffin [23,24] revealed an important role of an external forward view in inhibiting motion sickness by extensive surveys in public road transport. Poor forward visibility was found to increase motion sickness. Griffin and Newman [21] obtained results that a forward view reduced sickness even if it comprised a narrow visual field. In contrast, it has been shown that visual input from an internal view that seems stationary for moving subjects is more nauseogenic than having no visual input at all [18,25,26]. It is considered that an internal view of the vehicle generates visual–vestibular sensory conflict, which causes motion sickness, and that an external forward view diminishes the conflict. Now, let us consider the riding conditions in the present study. Subjects were seated in the second row and could see an external forward view through a front window in the notask condition, although an internal view of the vehicle was also provided. On the other hand, subjects watched a movie via an invehicle video display in front of them under the video-watching condition. The 11 inch-type in-vehicle video displays were placed 60 cm apart in front of the subjects, who had a partial external forward view behind the displays. Under the book-reading condition, subjects read a picture book through orientating their gaze somewhat downward. Their visual field around the periphery of the book was limited to the inside of the car, which did not give any information on their body motion. Thus, aggravation by watching a video and reading a book was considered to have resulted from an increase in visual–vestibular sensory conflict. The severity difference between the two conditions seems to result primarily from the fact that a partial frontal view was given behind the in-vehicle video display under the video-watching condition. An additional possibility of the severity difference is a difference in the head posture of subjects. The head must be kept upright when watching a video, while it must be tilted downward when reading a book. It has been revealed that the direction of linear acceleration relative to the body affects the incidence of motion sickness. Vogel et al. [27] added horizontal, linear acceleration to individual subjects in a car, and observed that subjects sitting upright were more susceptible to motion sickness than those who lying supine. Golding et al. [28] also derived the same conclusion that linear acceleration parallel to the sagittal axis (x-axis) is more nauseogenic than that parallel to the longitudinal axis (z-axis). Considering that an acceleration on ‘‘the averaged plane of the utricular macula’’ (25–30 degrees forward-up from the horizontal plane) is the most effective for producing a kinesthetic sensation [29,30], their results could be attributed to the fact that the sagittal axis forms a smaller angle with this plane than the longitudinal axis. If that is the case, the nosedown posture of the head that subjects held when reading a book placed ‘‘the average plane of the utricular macula’’ near the horizontal, which might have made motion sickness worse than that when watching a video. Unfortunately, however, we did not measure the head posture in the present experiment. This interpretation should be confirmed in a future study. The present study quantitatively revealed the aggravation of motion sickness due to watching an in-vehicle video display in an automobile. Now, the open question is how we can reduce motion sickness aggravated by in-vehicle video watching. One reasonable solution is to provide a visual sensation equivalent to the vestibular sensation that conveys information of vehicle motion. Optokinetic stimuli corresponding to vehicle motion should be provided via an in-vehicle video display along with a movie [5,31]. Considering that in-vehicle videos have recently become popular for passengers’ entertainment in vehicles, it is necessary to develop an in-vehicle video designed to reduce motion sickness.
N. Isu et al. / Displays 35 (2014) 90–97
6. Conclusions In motion sickness studies, an important issue is how to treat the individual difference in motion sickness susceptibility. Although the issue can be circumvented by completing withinsubject design experiment, it is difficult in practice to carry out the complete set of trials because subjects are allowed to withdraw from the subsequent trials in accordance with ethical requirement. In this study, we introduce an experimental protocol and a statistical analysis procedure to handle such incomplete experimental data, and demonstrate that the protocol and the procedure are useful for quantitative assessment of the time-course development of motion sickness severity. The present experimental study evaluated how severely invehicle video watching in a moving vehicle brings motion sickness on passengers in comparison with book-reading and ordinary carriding. Time-course development of motion sickness severity was measured every one minute during 15-min driving along a winding road under the three riding conditions. The severity increased almost in proportion to the time of car-riding under every condition. Watching a video aggravated motion sickness 2.7 times as much as ordinary car-riding, and 25% less severely than reading a book. We can say with fair certainty that the aggravation of motion sickness is attributable to a lack of external view which increases visual–vestibular sensory conflict. It is most likely that an external frontal view, partially visible while watching a video, lessened motion sickness compared with that while book-reading. The head posture might also affect the degree of sickness. Acknowledgments The authors wish to thank Daisuke Ioku and Hitoshi Asano for assistance in data acquisition in the experiment. This study was conducted with the support of the Automotive Systems Company, Panasonic Corporation, Yokohama, Japan. References [1] J.T. Reason, J.J. Brand, Motion Sickness, Academic Press, London, 1975. pp. 102– 134. [2] C. Diels, Carsickness – preventive measures, Transport Res. Lab. Proj. Rep. CPR 130 (2008) 1–8. [3] B. Schoettle, M. Sivak, In-Vehicle Video and Motion Sickness, Technical Report UMTRI-2009-6, University of Michigan Transportation Research Institute, Ann Arbor, MI, 2009. [4] K. Kato, S. Kitazaki, A study for understanding carsickness based on the sensory conflict theory, SAE Technical Paper 2006-01-0096, 2006, pp. 1–7. [5] K. Kato, S. Kitazaki, Improvement of ease of viewing images on an in-vehicle display and reduction of carsickness, SAE Technical Paper 2008-01-0565, 2008, pp. 1–6. [6] P.A. Howarth, S.G. Hodder, Characteristics of habituation to motion in a virtual environment, Displays 29 (2008) 117–123.
97
[7] M.M.C. Bijveld, A.M. Bronstein, J.F. Golding, M.A. Gresty, Nauseogenicity of offvertical axis rotation vs. equivalent visual motion, Aviat. Space Environ. Med. 79 (2008) 661–665. [8] O.L. Bock, C.M. Oman, Dynamics of subjective discomfort in motion sickness as measured with a magnitude estimation method, Aviat. Space Environ. Med. 53 (1982) 773–777. [9] S.C. Nichols, S.V.G. Cobb, J.R. Wilson, Health and safety implications of virtual environments: measurement issues, Presence: Teleoperat. Virtual Environ. 6 (1997) 667–675. [10] M. Emoto, M. Sugawara, Y. Nojiri, Viewing angle dependency of visuallyinduced motion sickness in viewing wide-field images by subjective and autonomic nervous indices, Displays 29 (2008) 90–99. [11] J.T. Reason, A. Graybiel, Changes in subjective estimates of well-being during the onset and remission of motion sickness symptomatology in the slow rotation room, Aerospace Med. 41 (1970) 166–171. [12] J.F. Golding, Motion sickness susceptibility questionnaire revised and its relationship to other forms of sickness, Brain Res. Bull. 47 (1998) 507–516. [13] J.F. Golding, Predicting individual differences in motion sickness susceptibility by questionnaire, Pers. Individ. Dif. 41 (2006) 237–248. [14] G. Verbeke, G. Molenberghs, Linear Mixed Models for Longitudinal Data, Springer, New York, 2000. pp. 77–89. [15] P. Heagerty, K.Y. Liang, S.L. Zeger, P.J. Diggle, Analysis of Longitudinal Data, Oxford University Press, Oxford, 2002. pp. 169–189. [16] W.H. Greene, Econometric Analysis, Prentice-Hall, Upper Saddle River, 2000. pp. 499–524. [17] P. Good, Permutation, Parametric, and Bootstrap Tests of Hypotheses, Springer, New York, 1994. pp. 33–67. [18] K.L. Mills, M.J. Griffin, Effect of seating, vision and direction of horizontal oscillation on motion sickness, Aviat. Space Environ. Med. 71 (2000) 996– 1002. [19] N.A. Webb, M.J. Griffin, Optokinetic stimuli: motion sickness, visual acuity, and eye movements, Aviat. Space Environ. Med. 73 (2002) 351–358. [20] N.A. Webb, M.J. Griffin, Eye movement, vection, and motion sickness with foveal and peripheral vision, Aviat. Space Environ. Med. 74 (2003) 622–625. [21] M.J. Griffin, M.M. Newman, Visual field effects on motion sickness in cars, Aviat. Space Environ. Med. 75 (2004) 739–748. [22] J.M. Lentz, W.E. Collins, Motion sickness susceptibility and related behavioral characteristics in men and women, Aviat. Space Environ. Med. 48 (1977) 316– 322. [23] M. Turner, M.J. Griffin, Motion sickness in public road transport: passenger behavior and susceptibility, Ergonomics 42 (1999) 444–461. [24] M. Turner, M.J. Griffin, Motion sickness in public road transport: the relative importance of motion, vision and individual differences, Br. J. Psychol. 90 (1999) 519–530. [25] T. Probst, S. Krafczyk, W. Büchele, T. Brandt, Visuelle prävention der bewegungskrankheit im auto, (visual prevention of motion sickness in cars), Arch. Psychiatr. Nervenkr. 231 (1982) 409–421 (German). [26] J.E. Bos, S.N. MacKinnon, A. Patterson, Motion sickness symptoms in a ship motion simulator: effects of inside, outside, and no view, Aviat. Space Environ. Med. 76 (2005) 1111–1118. [27] H. Vogel, R. Kohlhaas, R.J. von Baumgarten, Dependence of motion sickness in automobiles on the direction of linear acceleration, Eur. J. Appl. Physiol. 48 (1982) 399–405. [28] J.F. Golding, H.M. Markey, J.R.R. Stott, The effects of motion direction, body axis, and posture on motion sickness induced by low frequency linear oscillation, Aviat. Space Environ. Med. 66 (1995) 1046–1051. [29] H. Schöne, On the role of gravity in human spatial orientation, Aerospace Med. 35 (1964) 764–772. [30] H. Jaslow, Human tilt perception in a dynamic environment, Aviat. Space Environ. Med. 50 (1979) 594–598. [31] A. Morimoto, N. Isu, T. Okumura, Y. Araki, A. Kawai, F. Masui, Screen design of onboard displays for reducing car sickness, in: Proc. of FISITA 2008 World Automotive Congress (CD-ROM), F2008-02-23, 2008.
Contents lists available at ScienceDirect
Displays journal homepage: www.elsevier.com/locate/displa
Quantitative analysis of time-course development of motion sickness caused by in-vehicle video watching Naoki Isu a,⇑, Takuya Hasegawa b,1, Ichiro Takeuchi b, Akihiro Morimoto a,2 a b
Faculty of Engineering, Mie University, 1577 Kurima-machiya, Tsu, Mie 514-8507, Japan Faculty of Engineering, Nagoya Institute of Technology, Gokiso-cho, Showa-ku, Nagoya, Aichi 466-8555, Japan
a r t i c l e
i n f o
Article history: Received 4 June 2012 Received in revised form 21 March 2013 Accepted 19 January 2014 Available online 28 January 2014 Keywords: Visual–vestibular sensory conflict Susceptibility Statistical analysis Random-component location-scale model Incomplete within-subject design Onboard TV
a b s t r a c t It has been common and popular to watch videos in moving vehicles. An important issue in developing comfortable in-vehicle video watching systems is to understand how passengers get motion sickness. With this in mind, the goals of this paper are (1) to introduce an experimental protocol and a statistical analysis procedure for quantitatively evaluating how motion-sickness is developed during car-driving, and (2) to demonstrate their practical usefulness with a working experimental study. In the experimental protocol, motion sickness was induced to subjects by requiring them to watch an in-vehicle video during 15-min driving, and the time-course development of motion sickness was recorded by asking subjects to evaluate their degree of motion sickness every one minute. A main difficulty in analyzing data from these studies is how to incorporate the individual difference in motion-sickness susceptibility. Since susceptibilities are markedly different among subjects, within-subject design experiments are preferred. However, it is practically difficult to conduct complete set of trials because subjects who are not willing to continue experiment (due to heavy motion sickness) should be able to withdraw from the subsequent series of trials in accordance with ethical requirement. To cope with this incomplete data issue, we introduce a statistical data analysis procedure that enables to estimate and impute missing entries in the within-subject design table. Using a working example, we demonstrated that the protocol and the procedure are useful for quantitative assessment of the time-course motion sickness development. We conducted an in-vehicle motion-sickness study with 31 subjects, where the time-course motion sickness developments of video-watching, book-reading, and normal riding conditions were compared. The results indicate that video-watching brings on 2.7 times more severe motion sickness than normal riding, but 25% less severe than book-reading. Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction Recent advancement of video-screen technology allows us to watch videos in moving vehicles such as cars, boats, and airplanes. An important issue in the development of in-vehicle video systems is to understand how easily and severely passengers get motion sickness. It is well known that reading books or maps in a moving vehicle brings on and aggravates motion sickness. This is known to be the result of sensory conflict [1], that is, the vestibular system conveys sensory information of body movement, while the visual system conveys stationary information creating the sensation of the body being immobile. ⇑ Corresponding author. Tel.: +81 59 231 9458; fax: +81 59 231 9460. E-mail address: [email protected] (N. Isu). Present address: DENSO Corporation, 1-1 Showa-cho, Kariya 448-8611, Japan. Present address: Automotive Systems Company, Panasonic Corporation, 4261 Ikonobe-cho, Tsuduki-ku, Yokohama, Kanagawa 225-8520, Japan. 1 2
http://dx.doi.org/10.1016/j.displa.2014.01.003 0141-9382/Ó 2014 Elsevier B.V. All rights reserved.
Watching a video in a moving vehicle is considered to trigger sensory conflict in the same manner as book-reading [2]. Schoettle and Sivak [3] investigated the frequency and the severity of motion sickness experienced from in-vehicle video usage by a questionnaire survey. The study indicated that video watching in a moving vehicle caused motion sickness less often than book-reading, but more often than normal riding. Kato and Kitazaki [4,5] also investigated the influence of in-vehicle video display on carsickness by their experimental studies. It showed that reading still images (news text) or watching moving images brought on more severe carsick than the case that passengers were allowed to see external view. In order to offer comfortable video-watching in moving vehicles, it has been desired to develop new video systems that can reduce motion sickness. For developing such new video systems, it is essential to establish an experimental protocol that can quantitatively evaluate how motion-sickness is developed during
N. Isu et al. / Displays 35 (2014) 90–97
car-driving. To assess the severity of sickness in series during an experimental trial, it must be briefly evaluated in an instant at every measuring time. A variety of subjective rating methods [e.g., 4–7] and magnitude estimation methods [8] have been frequently used for that purpose. Short Symptoms Checklist [9] is also adopted for evaluating the severity of symptoms at given intervals (e.g., every 5 min) during stimulus periods. Another estimation method, in which subjects respond with a joystick instead of replying with numerical ratings, has been used for continuous measurement [10]. In this paper, we first introduce an experimental protocol and a statistical analysis procedure for quantitative assessment of time-course motion sickness development. Then we demonstrate with a working experimental study that our new approach is practically useful for understanding how motion-sickness is developed during car-driving. In the experimental protocol, motion sickness is induced to subjects by requiring them to watch a video via an in-vehicle video display during 15-min driving along a winding road. The timecourse development of motion sickness is recorded by asking subjects to evaluate their degree of motion sickness every one minute of 15-min driving. The degree of motion sickness is subjectively evaluated with 11 grades (0–10) on a rating scale, which is essentially the same as the well-being score introduced by Reason and Graybiel [11]. One of the difficulties in experimental studies on motion sickness is how to incorporate the individual difference in motion sickness susceptibility which differs markedly among individual subjects. When different stimulus conditions are compared such as video-watching, book-reading, and normal-riding, subjects who have similar susceptibilities should be evenly assigned to each stimulus condition. In some previous studies (e.g., [4–6]), individual susceptibilities are scored by questionnaire such as the Motion Sickness Susceptibility Questionnaire (MSSQ) [12,13] in advance, and the subjects were evenly assigned to each stimulus condition based on the score. However, since motion-sickness susceptibilities are markedly different among individuals, stratification based on questionnaire score is not reliable enough, and the results highly depend on subject assignments. Therefore, it is desirable that every stimulus condition is evenly tested by each subject (within-subject design). In practice, however, it is difficult to carry out the complete set of trials, i.e., same number of trials with every stimulus condition by each subject. It is because subjects are allowed to withdraw from the subsequent series of trials in accordance with ethical requirement (subjects who had severe motion sickness in the previous trial would not be willing to try it again). As the result, the within-subject design table inevitably has several missing entries. Even worse, those missing entries arise with bias because subjects who have more severe motion sickness tend to withdraw more often than those who have no motion sickness. This kind of practical problem is a formidable issue more or less common to experimental motion-sickness study. To handle this incomplete data issue, we introduce a simple statistical model for time-course development of motion sickness. This approach allows us to estimate and impute the missing entries in the within-subject design table, and to compare the motion sickness severities among different stimulus conditions. In this paper, for the purpose of demonstration, we conducted an in-vehicle motion-sickness study with 31 subjects, where the time-course motion sickness developments of videowatching, book-reading, and normal riding conditions were compared. The conclusion of the analysis was consistent with previous studies [3–5], i.e., video watching brings on more severe motion sickness than normal riding, but is less severe than book-reading.
91
2. Experimental protocol In this section, we present the experimental protocol by describing the experimental setup of our working example study on time-course motion sickness. 2.1. Subjects N = 31 subjects (21 males and 10 females), approximately 20 years of age, participated in the experiment. None of the subjects reported any medical problems and all presented with normal vestibular functions. Every subject was briefed on the purpose, procedures, risks, and benefits of the study, and gave written informed consent before participating in the experiment. It was also explained that the subjects could withdraw from the experiment whenever they requested, that is, when they could no longer endure the nausea of motion sickness, they could immediately get out of the car and terminate the experimental trial. The experimental protocol used in this study was approved by the Ethics Committee of the Faculty of Engineering, Mie University. 2.2. Apparatus The experiment was performed using a minivan, which had three rows of seats and a boarding capacity of 7 persons. Subjects, one or two in each experimental trial, were seated in the second row of the car. Two in-vehicle video displays were installed separately behind head rests of the first row of the seats by means of the arms fixed to the side roof. They were located about 60 cm apart in front of the individual subjects. The displays had an 11 inch-type wide LCD panel, 244 mm (width) 138 mm (height), whose resolution was 800 (horizontal) 480 (vertical) pixels. The horizontal and vertical viewing angles for the subjects were about 23 and 13 degrees, respectively. Movies were played on the in-vehicle video displays using a DVD player. Sound was provided by speakers attached on the side doors through an FM radio. 2.3. Stimuli (riding conditions and driving course) The experiment was performed to compare the effect of riding conditions on motion sickness. Subjects rode in the car for T = 15 min under one of the following K = 3 conditions: (1) videowatching, (2) book-reading, and (3) no-task conditions. Under the video-watching condition (k = 1), subjects were required to watch a movie shown on the in-vehicle video displays. The movies used in the experiment did not include any scenes of an intense nature or uncomfortable scenes. They were selected from among romantic and family comedies (‘‘Legally Blonde’’, etc.) that had not been watched by the subjects, so that the subjects could concentrate their attention on the scene. The sound was provided in the native language of the subjects (i.e., Japanese) and subtitles were not superimposed. Under the book-reading condition (k = 2), subjects were required to read a picture book (‘‘Where’s Wally?’’) that had few letters but lots of small fine figures. The task for readers was to look for one or several particular figures which got mixed in among lots of similar ones on each page. In the notask condition (k = 3), subjects were simply asked to remain still and quiet. They were allowed to look outside of the car freely. To ensure the subjects obeying the instructions, their behavior was observed by an experimenter seated in the third row of the car. The driving course used in the experiment was a suburban road with numerous curves and occasional gentle slopes. It had no traffic signals but two stops. There were no houses along the road, and the traffic, if any, was not heavy. The driving course was a 2.1-km circuit, and it took 3 min to drive around. The car went around the course 5
92
N. Isu et al. / Displays 35 (2014) 90–97
times for a total of 15 min in each experimental trial, and was driven by a single driver in order to minimize the variance in driving manner throughout the whole experiment. The driver was instructed to drive along the traffic lane on the road as accurately as possible while keeping a constant speed under the legal limit (50 km/h), and the experimental design and details were not informed to the driver. The driving course had 19 sharp curves in one circuit and the maximal angular velocities of yaw in the individual curves were around 8–35 deg/s (mean ± SD: 14.8 ± 7.2 deg/s). 2.4. Measurement of motion sickness severity During a T = 15-min trial, subjects required to subjectively evaluate their degree of motion sickness in a rating-scale such as wellbeing scores [11]. They were asked to answer it orally according to the experimenter’s request every one minute. The subjective severity was expressed in 11 numerical categories (severity scores) from 0 (normal condition with no sickness) to 10 (limit of subject endurance of severe nausea: endpoint). The subjects were instructed to assign one of the numerical categories as a severity score in proportion to the degree of motion sickness severity. When subjects withdrew in the middle of a trial due to severe sickness, severity scores for the rest of the trial were regarded as the maximal point of 10. 2.5. Within-subject-design and drop-outs As mentioned in Section 1, drop-outs of some subjects are inevitable in within-subject-design motion-sickness studies. Also in our working study, scheduled trials were sometimes canceled since the subjects were free to withdraw from the subsequent series of trials in accordance with ethical requirement. Consequently, 11 subjects withdrew after trials in a single condition, and 13 subjects attended trials in two of the three conditions before their withdrawal. Seven subjects attended trials in the all conditions. Excluding trials by the subjects participated only in a single condition, experimental records of 20 subjects were used for subsequent statistical analysis. They participated in 23, 24 and 34 trials in the video-watching, book-reading and no-task conditions, respectively. It means that our within-subject design table from the 20 subjects contains several missing entries. Those missing entries are estimated and imputed by the statistical analysis described in the following Chapter 3. Every subject participated in the experiment at least one day apart. One or two subjects attended each experimental trial. They were seated in the second row of the minivan. After a dozen of minutes driving (without any task) to get to the starting point of the course, subjects had a rest for 10 min before starting the trial. One of the above-mentioned tasks was assigned to subjects during a 15-min riding period. The subjective severity of motion sickness was recorded at the starting time and every one minute during 15min driving. When subjects felt severe nausea and desired to withdraw, they could get out of the car and discontinue the trial at any time. The numbers of trials terminated in this way were 3, 3 and 0 in the video-watching, book-reading and no-task conditions, respectively. Only in one case out of the 6 terminated trials, two subjects participated in the experiment together and one of them withdrew from it. We restarted the trial with the other subject after letting the former off. The interruption of the driving was shorter than 30 s. 3. Statistical analysis procedure In this section, we present the statistical analysis procedure for analyzing the data obtained from experiments such as the working example in the previous section.
Since individual difference of motion-sickness susceptibilities are very large, within-subject design is best suitable for these studies. However, planned trials are often canceled because subjects should be allowed to drop off from the experiment at any stage in accordance with ethical requirement. Accordingly, the following difficult issues must be addressed in statistical analysis. (1) First, the procedure must have a mechanism to impute the missing entries in the within-subject design table. (2) Second, the procedure must be able to capture the timecourse development of motion sickness, in which both the average and the variability (variance) of the severities have increasing trends with riding time. To handle these issues all together, we introduce a simple statistical model, named a random-component location-scale model. The model allows us to estimate and impute the missing entries in the incomplete within-subject design table. It consists of two statistical models, a random-component model and a location-scale model. The former model is useful for longitudinal data analysis when heterogeneity among individuals is not negligible [14,15]. It contains a group of parameters which explains subject-wise motion sickness susceptibilities. Specifically, for each of N subjects, we have a parameter di, i = 1, ..., N. Large values in di indicate that the subject i tends to have severe motion sickness, while small values indicate that the subject i would not get motion sickness easily. The missing entries can be imputed by modeling individual effects as random components which are independent of stimulus condition effects. The latter model, also called as heteroscedastic model, is frequently used in econometrics [16]. This model contains another group of parameters which represent time-course effects of stimulus conditions. In particular, the average and the standard deviation of the motion sickness severity after t-minutes riding with condition k are expressed by the parameters lk,t and rk,t, respectively, where t = 1, ..., T, and k = 1, ..., K. It enables us to analyze time-course data whose variances as well as averages vary with the time by modeling them as functions of time. 3.1. Random-component location-scale model Let us denote yk,t,i be the severity score given by subject i after tminutes riding with condition k. Here, we use the subscript k e {1, , K} to distinguish K stimulus conditions. Similarly, we use the subscript t e {0,1, , T} to specify the riding time. The subscript i e {1,2, , N} represents the subject index. The random-component location-scale model is formulated as
yk;t;i ¼ lk;t þ rk;t di þ ek;t;i ;
ð1Þ
where lk,t and rk,t P 0 are the average (i.e., location) and the standard deviation (i.e., scale) of the motion sickness severity scores over subjects at riding time t with stimulus condition k, respectively. Furthermore, di denotes the random component indicating the susceptibility to motion sickness of subject i, and we assume, for identifiability, that fdi gNi¼1 have zero mean and unit variance. Finally, ek,t,i is the Normally distributed error term independently and identically distributed for all (k, t, i) with the mean zero. The main goal of the analysis is to estimate the average flk;t gTt¼0 and variability frk;t gTt¼0 of the motion sickness severities for each stimulus condition k e {1, , K}. In the model (Eq. (1)) we do not assume any functional relationship between the average severities lk,t and the riding time t because the scores are ordinal scale (not distance scale). Instead, we assume that the average severities are monotonically increasing along the riding time t, i.e.,
0 lk;0 lk;1 lk;T 10; for all k:
ð2Þ
N. Isu et al. / Displays 35 (2014) 90–97
In addition, we introduce the restrictions on the lower and the upper bounds of the expected severities:
0 lk;t þ rk;t di 10; for all k; t; and i
ð3Þ
because the range of the severity scores is limited to [0, 10]. Note that the motion sickness susceptibility di is independent of stimulus condition k. It roughly implies that a subject who is prone to get severe motion sickness in one stimulus condition tends to have severe motion sickness in other conditions as well. This separability of the susceptibility parameters di and stimulus condition parameters (lk,t, rk,t) enables us to impute the missing entries in our within-subject design table. Namely, hypothetical motion sickness trend for a subject i with stimulus condition k can be estin oT lk;t ; rk;t into the randommated by substituting di and t¼1
component location-scale model. The parameters of the model are estimated using least-squares method (i.e., maximum likelihood estimation under the Gaussian noise). The sum of the squared errors between the observed and the estimated severity scores are minimized under the aforementioned constrains (Eqs. (2) and (3)). The problem is mathematically formulated as
X n
o2
93
sampling. A hypothetical experimental data under the null hypothesis is generated by randomly permuting the stimulus condition T index of yk;t;i t¼1 between k and k0 as if they were gotten in either condition of the two jumbled together. Then we fit the randomcomponent location-scale model (Eq. (1)) with the randomized hypothetical data and compute the test statistic in Eq. (6). By repeating the randomized sampling process many times (10,000 times in our analysis), we obtain the null distribution of the test statistic, and p-values can be simply computed by
p :¼
B 1X ðbÞ I Dsk;k0 Dsk;k0 ; B b¼1
ð7Þ
where B is the number of randomized sampling (B = 10,000 in the ðbÞ current analysis), Dsk;k0 is the test static computed from the b-th randomized data, and I is the index function defined as
IðzÞ ¼
1; if z is true; 0; if z is false:
ð8Þ
4. Results
ð4aÞ
In this section, we first report the experimental results obtained from the working experiment, and then show the results of the statistical analysis based on our new model-based analysis.
subject to 0 lk;0 lk;1 lk;T 10; for all k;
ð4bÞ
4.1. Subjective ratings of motion sickness severity
rk;t 0; for all k and t;
ð4cÞ
0 lk;t þ rk;t di 10; for all k; t; and i;
ð4dÞ
N X di ¼ 0;
ð4eÞ
The subjective severity of motion sickness increased gradually during the car ride in every condition. The time-course development of motion sickness severity is shown in histograms of Fig. 1 at intervals of 3 min. The ordinate indicates the ratio of trials rather than frequency so that the histograms can be directly compared among the three conditions. The severity was 0 (no sickness) at the beginning of car-riding (0 min) in almost all trials. It gradually increased in most trials under the video-watching and book-reading conditions. The distribution of severity was shifted toward higher scores along with the riding time. The maximal severity reached 10 (endpoint) at 13 min with video-watching, while at 7 min with book-reading. In about 10% of trials under both conditions, the severity reached 10 in the middle of trials. On the other hand, the severity remained at 0 (no sickness) in about 40% of trials under the no-task condition. It mostly remained below a level of mild discomfort and never exceeded 6 through the entire period of car-riding in all trials. The first increase of severity score (from 0 to 1 in most trials) was observed before 6 min in over 80% of trials with video-watching and those with book-reading, while it took more than 7 min in half of trials with no-task. Histograms of the onset times are shown in Fig. 2a. Since the onset times under each condition had an asymmetrical distribution that extended toward higher values, their median rather than mean was obtained. The medians of the onset time for the no-task, video-watching, and book-reading conditions were 7.5, 3, and 3 min, respectively. It is known that some slight symptoms such as stomach awareness appear around a well-being score of 3–4 [11]. Since the severity score used in this study is equivalent to the well-being score of Reason and Graybiel [11], the instant of exceeding score 3 can be considered to indicate the onset time of slight symptoms. The onset times were distributed as shown in Fig. 2b. The medians were 9 and 8.5 min for the video-watching and book-reading conditions, respectively. The severity score did not exceed 3 in over a half of trials under the no-task condition so that the median of the onset was not obtained within 15 min of riding period. From the individual time-course development of severity reported in each trial, we observed that the severity scores, on aver-
min
flk;t ;rk;t g8k;8t ;fdi g8i
yk;t;i
lk;t þ rk;t di
ðk;t;iÞ2D
i¼1 N 1 X d2 ¼ 1; N 1 i¼1 i
ð4fÞ
where D in Eq. (4a) represents the set of all possible triplets (k, t, i). The optimization problem is solved by a simple iterative procedure using Matlab CVX package (http://cvxr.com/cvx/). The Matlab code is available by request. 3.2. Summary statistic of motion sickness and the statistical test After fitting the model and estimating these parameters, we compute severity index sk as a summary statistic of motion sickness severity with stimulus condition k. It is defined as
sk :¼
T 1X l^ k;t ; T t¼1
ð5Þ
^ k;t is the estimated value of lk,t and T is the total riding time. where l We examine the statistical significance of the difference in the severity indices among different stimulus conditions. Consider testing the difference between two stimulus conditions k and k0 . The test statistic is simply defined as
Dsk;k0 :¼ sk0 sk :
ð6Þ
Here, the null hypothesis is that there is no difference in motion sickness between the conditions k and k0 , and the alternative hypothesis is that the stimulus condition k0 yields more severe sickness than the condition k. To compute the statistical significance (p-values), we use permutation-based approach [17]. Permutation-based approach is a general procedure to compute null distribution and statistical significance based on randomized
N. Isu et al. / Displays 35 (2014) 90–97
Video-Watching Book-Reading No-Task
60 40 20 0
0
2
4
6
8
80 60 40 20 0
10
0
2
9 min
60 40 20 0
0
2
4
6
4
6
8
60 40 20 0
10
6 min
80
0
2
Severity Score R tio of Trials [% Rat %]
R tio of Trials [% Rat %]
Severity Score
8
40 30 20 10 0
10
12 min
50
0
2
4
6
6
8
10
8
10
15 min
50 40 30 20 10 0
0
2
Severity Score
Severity Score
4
Severity Score R tio of Trials [% Rat %]
80
3 min
100
R tio of Trials [% Rat %]
0 min
100
R tio of Trials [% Rat %]
R tio of Trials [% Rat %]
94
4
6
8
10
Severity Score
Fig. 1. Histograms of severity scores of motion sickness developing with riding time. Histograms of severity scores are shown at intervals of 3 min for the three riding conditions: video-watching, book-reading, and no-task conditions. The abscissa indicates the severity scores, and the ordinate indicates the ratio of trials rather than frequency.
R tio of Trials [% Rat %]
(a)
Onset time of the 1st. increase of severity
60 50
Video- Watching Book-Reading No-Task
40 30 20 10 0
0
3
6
9
12
15
0
As summary indices of motion sickness severity induced by individual trials, accumulated scores were calculated by adding severity scores for 15 min. Accumulated scores under the videowatching condition were distributed from 0 to 75, and those under the book-reading condition spread more widely from 6 to 114 as shown in Fig. 3. In contrast, those under the no-task condition were remarkably lower than the other conditions; they were 0 in a third of trials, and even the maximal score was 36. The medians for the no-task, video-watching, and book-reading conditions were 6.5, 27, and 34.5, respectively. These results indicate that video-watching and book-reading aggravate carsickness compared with ordinary ride with no task, and the former seems to affect it less than the latter.
Onset Time [min] 4.2. Statistical model analysis of motion sickness severity development
Onset time of exceeding score 3
70
In order to explain the above observations of time-course development of carsickness severity, we use a random-component location-scale model that can describe the time-dependent changes of
R tio of Trials [% Rat %]
60
Video -Watching Book-Reading No-Task
50 40
40
30 20 10 0
0
3
6
9
12
15
0
Onset Time [min] Fig. 2. Histograms of onset times of the first severity increase and slight symptoms. (a) Onset times of the first increase of severity score (score 1 in most trials) are shown in histograms for the three riding conditions. (b) As the onset time of slight symptoms, elapsed times until the severity score exceeded 3 are shown in histograms. The ordinate indicates the ratio (%) of trials in which the onset times stood between a pair of elapsed times indicated at the divisions of the abscissa.
age, monotonically increased with riding time. In addition, the subject-wise variability also increased with riding time, namely, as time passed, motion sickness severities of some subjects got worse and worse, but those for some other subjects did not increase so rapidly.
R tio of Trials [% Rat %]
(b)
Video-Watching Book-Reading No-Task
30
20
10
0 0
15
30
45
60
75
90
105
120
Accululated Severity Score Fig. 3. Histograms of accumulated severity scores. Histograms of accumulated severity scores for each riding period are shown for the three conditions. In cases subjects withdrew in the middle of a trial, severity scores for the rest period were regarded as the maximal score of 10. The ordinate indicates the ratio (%) of trials in which the accumulated severity scores stood between a pair of scores indicated at the divisions of the abscissa, except that the accumulated scores were 0 in the leftmost section.
95
N. Isu et al. / Displays 35 (2014) 90–97
the mean and the variance of the motion sickness distribution. The model assumes monotonic increase of average severity along the riding time as formulated in Eq. (2), which was satisfied with the observations in each riding condition. The parameters of the model formulated in Eq. (1) were estimated by the least-squares method under the constrains given in Eq. (4). Fig. 4 shows the estimated motion sickness trends under the three conditions. The sequential lines indicate the progression of ^ k;t of motion sickness severity, and the verthe estimated average l tical lines attached topthe ffiffiffiffi average indicate the standard errors ^ k;t = N , where N = 20 is the sample size. The approximated by r severity increased almost in proportion to the time over the 15 min of car-riding with every condition. Motion sickness was the severest with the book-reading condition and the second severest with the video-watching condition at any point of the car ride. As a summary statistic of sickness severity, severity index sk defined in Eq. (5), i.e., average severity for 15 min of riding period, was calculated. The severity indices sk for the no-task, videowatching, and book-reading conditions were 0.80, 2.19, and 2.84, respectively; the ratio of these indices was 1:2.7:3.6 in the above order. Thus, it was shown that the motion sickness severity was increased about three times by watching a video and that the videowatching severity was one quarter less than that caused by reading a book. These results substantially agreed with those obtained with accumulated scores in the above section, although they made a slight difference in the ratio of severity among the conditions. To examine the statistical significance of the difference in the motion sickness trends among different stimulus conditions, permutation-based statistical significance tests were performed in accordance with the procedure described in the Section 3.2. The p-value of the difference between no-task and video-watching conditions was 0.003, that between no-task and book-reading conditions was 0.001, and that between video-watching and bookreading conditions was 0.064. Thus, the statistical tests corrobo-
7
Video-Watching Video Watching Book-Reading Book Reading No-Task No Task
6
5
rated highly significant differences in motion sickness aggravation by video-watching and book-reading compared with ordinary carriding. 4.3. Distribution of subject-wise motion sickness susceptibility Subject-wise motion sickness susceptibility di, which is another parameter of the model formulated in Eq. (1), was estimated to^ k;t and r ^ k;t in the above section. The distribution of gether with l the susceptibility is illustrated in a histogram of Fig. 5. It shows the numbers of subjects (in the ordinate) having the susceptibility values in the abscissa. The origin of the abscissa (di = 0) indicates the mean susceptibility among the subjects participated in this study. The susceptibility values are expressed taking their standard deviation as the unit (di = 1). The histogram indicates that the distribution of the individual susceptibility is widespread (rather than having a single peak like the Normal distribution). In addition, there seems to be a small group of individuals who have very high susceptibility (those with susceptibility >1.0). We can see its reflection on the histograms of severity scores in Fig. 1. The histograms at the last stages of the riding period (12 and 15 min) have small bumps at high severity scores, indicating that the group of subjects with high susceptibility got sever sickness. Histograms of accumulated scores and onset times failed to show any influence resulted from the dichotomous distribution of subject-wise susceptibility (see Figs. 2 and 3). 5. Discussion When the severity of motion sickness is compared among different stimulus conditions, every condition must be tested by subjects evenly balanced for their susceptibility to motion sickness. In general, however, it is hardly possible to balance the susceptibility evenly because it is markedly different among individual subjects. Although it is desirable to conduct within-subject design experiment in which every stimulus is evenly tested by each subject, balanced assignment would not always be carried out in practice because subjects must be free to withdraw from the experiment at any stage. The unbalanced susceptibility due to missing trials makes it difficult to compare the motion sickness severities among different stimulus conditions. The present study has provided a solution to this formidable problem by introducing a statistical model called a random-component location-scale model. The model (see Eq. (1)) has two sets of parameters: one for stimulus condition effects and another for
4 7 6
3
5 4
2
3 2
1
1 0 -2.4 -1.8 -1.2 -0.6
0
0
5
10
15
Riding Time (min) Fig. 4. Motion sickness trends under the three conditions. The sequential lines ^ k;t of motion sickness severity indicate the progression of the estimated average l under the three riding conditions. Vertical lines attached to the average indicate the pffiffiffiffi ^ k;t = N . standard errors approximated by r
0
0.6
1.2
1.8
2.4
3
Susceptibility Index δ i Fig. 5. Distribution of the individual susceptibility to motion sickness. Individual susceptibility to motion sickness was estimated by the random-component location-scale model on 20 subjects. The ordinate of the histogram indicates the number of subjects who had motion sickness susceptibility di plotted as the abscissa.
96
N. Isu et al. / Displays 35 (2014) 90–97
subject-wise motion sickness susceptibilities. Note that the latter susceptibility parameters are independent of the three stimulus conditions. This makes it possible to estimate the average severity scores (the first term in Eq. (1)) in each stimulus condition even though the susceptibility is not exactly balanced among the conditions. Furthermore, this model captures increasing trends of the mean and the variance of the motion sickness distribution. Note that a standard regression model can only describe the change of the mean and implicitly assumes that the variance is constant. The statistical model analysis introduced in this study provides a considerable advantage to quantitative and comparative studies on time-course development of motion sickness. Fig. 4 shows that the severity of motion sickness increases in proportion to the car-riding period. The same tendency was observed in previous studies; the severity gradually increased in proportion to the duration of horizontal body oscillation [18], optokinetic stimulation [19,20], or combined oscillation of body and vision [18] for 30 min. When intensive stimuli were imposed, the increase in severity slowed down after 10–15 min [7,21]. Since proportional relationships were recognized in the progression of motion sickness severity under all conditions in this study, the severity index defined by Eq. (5) was used to present the aggravating effect of a riding condition on motion sickness. As a result, then, watching a video in a moving vehicle aggravated motion sickness about three times as severely as the no-task condition. It was 25% less severe than that caused by reading a book. Substantially identical results were obtained by using the accumulated severity scores without analysis adopting the statistical model. The ratio of the medians of accumulated scores for the no-task, video-watching, and book-reading conditions was 1:4.2:5.1, which is somewhat higher than that obtained by the statistical analysis (i.e., 1:2.7:3.6). It could be considered that this difference mainly derived from distorted balance of the susceptibility among the conditions caused by missing trials. Since the statistical model analysis can compensate for the missing trials caused by withdrawal of subjects, it likely provides us with more reliable estimation. Now, let us consider similar experiments conducted by Kato and Kitazaki [4,5]. They balanced the subject-wise susceptibility to motion sickness by evenly grouping subjects for each stimulus in accordance with MSSQ. Each group consisted of 17–20 subjects and was involved in a single stimulus. Although they did not describe the severity difference quantitatively in text, their illustrations indicate that watching moving images on an in-vehicle video display induced carsickness 2.9 times as severely as an ordinary ride having external view, and that reading still images (two lines of 15 characters) on the display induced sickness of 3.3 times severity. Their results were quite similar to the present statistical analysis, although we need to notice that their stimulus conditions were not identical with those in the present experiment. Here, note that the number of subjects could be saved in the present experiment thanks to the statistical analysis while they needed a fairly large number of subjects in order to balance the susceptibility among the stimulus conditions. Subject-wise motion sickness susceptibility showed dichotomous distribution in the present study (see Fig. 5). This differs from a simple distribution of motion sickness susceptibility investigated with MSSQ [12,13,22]. However, note that the histogram of Fig. 5 shows the distribution among the subjects participated in the present experiment voluntarily. It is known that higher susceptible individuals have less interest to attend experiments concerning motion sickness [22]. Therefore it is not likely that the histogram represents the susceptibility distribution of the whole people in general. Since subjects cannot be randomly selected from the population, it is difficult to investigate the general distribution of motion sickness in experimental studies.
It has been shown in previous studies that a passenger’s vision affects motion sickness to a great extent. Turner and Griffin [23,24] revealed an important role of an external forward view in inhibiting motion sickness by extensive surveys in public road transport. Poor forward visibility was found to increase motion sickness. Griffin and Newman [21] obtained results that a forward view reduced sickness even if it comprised a narrow visual field. In contrast, it has been shown that visual input from an internal view that seems stationary for moving subjects is more nauseogenic than having no visual input at all [18,25,26]. It is considered that an internal view of the vehicle generates visual–vestibular sensory conflict, which causes motion sickness, and that an external forward view diminishes the conflict. Now, let us consider the riding conditions in the present study. Subjects were seated in the second row and could see an external forward view through a front window in the notask condition, although an internal view of the vehicle was also provided. On the other hand, subjects watched a movie via an invehicle video display in front of them under the video-watching condition. The 11 inch-type in-vehicle video displays were placed 60 cm apart in front of the subjects, who had a partial external forward view behind the displays. Under the book-reading condition, subjects read a picture book through orientating their gaze somewhat downward. Their visual field around the periphery of the book was limited to the inside of the car, which did not give any information on their body motion. Thus, aggravation by watching a video and reading a book was considered to have resulted from an increase in visual–vestibular sensory conflict. The severity difference between the two conditions seems to result primarily from the fact that a partial frontal view was given behind the in-vehicle video display under the video-watching condition. An additional possibility of the severity difference is a difference in the head posture of subjects. The head must be kept upright when watching a video, while it must be tilted downward when reading a book. It has been revealed that the direction of linear acceleration relative to the body affects the incidence of motion sickness. Vogel et al. [27] added horizontal, linear acceleration to individual subjects in a car, and observed that subjects sitting upright were more susceptible to motion sickness than those who lying supine. Golding et al. [28] also derived the same conclusion that linear acceleration parallel to the sagittal axis (x-axis) is more nauseogenic than that parallel to the longitudinal axis (z-axis). Considering that an acceleration on ‘‘the averaged plane of the utricular macula’’ (25–30 degrees forward-up from the horizontal plane) is the most effective for producing a kinesthetic sensation [29,30], their results could be attributed to the fact that the sagittal axis forms a smaller angle with this plane than the longitudinal axis. If that is the case, the nosedown posture of the head that subjects held when reading a book placed ‘‘the average plane of the utricular macula’’ near the horizontal, which might have made motion sickness worse than that when watching a video. Unfortunately, however, we did not measure the head posture in the present experiment. This interpretation should be confirmed in a future study. The present study quantitatively revealed the aggravation of motion sickness due to watching an in-vehicle video display in an automobile. Now, the open question is how we can reduce motion sickness aggravated by in-vehicle video watching. One reasonable solution is to provide a visual sensation equivalent to the vestibular sensation that conveys information of vehicle motion. Optokinetic stimuli corresponding to vehicle motion should be provided via an in-vehicle video display along with a movie [5,31]. Considering that in-vehicle videos have recently become popular for passengers’ entertainment in vehicles, it is necessary to develop an in-vehicle video designed to reduce motion sickness.
N. Isu et al. / Displays 35 (2014) 90–97
6. Conclusions In motion sickness studies, an important issue is how to treat the individual difference in motion sickness susceptibility. Although the issue can be circumvented by completing withinsubject design experiment, it is difficult in practice to carry out the complete set of trials because subjects are allowed to withdraw from the subsequent trials in accordance with ethical requirement. In this study, we introduce an experimental protocol and a statistical analysis procedure to handle such incomplete experimental data, and demonstrate that the protocol and the procedure are useful for quantitative assessment of the time-course development of motion sickness severity. The present experimental study evaluated how severely invehicle video watching in a moving vehicle brings motion sickness on passengers in comparison with book-reading and ordinary carriding. Time-course development of motion sickness severity was measured every one minute during 15-min driving along a winding road under the three riding conditions. The severity increased almost in proportion to the time of car-riding under every condition. Watching a video aggravated motion sickness 2.7 times as much as ordinary car-riding, and 25% less severely than reading a book. We can say with fair certainty that the aggravation of motion sickness is attributable to a lack of external view which increases visual–vestibular sensory conflict. It is most likely that an external frontal view, partially visible while watching a video, lessened motion sickness compared with that while book-reading. The head posture might also affect the degree of sickness. Acknowledgments The authors wish to thank Daisuke Ioku and Hitoshi Asano for assistance in data acquisition in the experiment. This study was conducted with the support of the Automotive Systems Company, Panasonic Corporation, Yokohama, Japan. References [1] J.T. Reason, J.J. Brand, Motion Sickness, Academic Press, London, 1975. pp. 102– 134. [2] C. Diels, Carsickness – preventive measures, Transport Res. Lab. Proj. Rep. CPR 130 (2008) 1–8. [3] B. Schoettle, M. Sivak, In-Vehicle Video and Motion Sickness, Technical Report UMTRI-2009-6, University of Michigan Transportation Research Institute, Ann Arbor, MI, 2009. [4] K. Kato, S. Kitazaki, A study for understanding carsickness based on the sensory conflict theory, SAE Technical Paper 2006-01-0096, 2006, pp. 1–7. [5] K. Kato, S. Kitazaki, Improvement of ease of viewing images on an in-vehicle display and reduction of carsickness, SAE Technical Paper 2008-01-0565, 2008, pp. 1–6. [6] P.A. Howarth, S.G. Hodder, Characteristics of habituation to motion in a virtual environment, Displays 29 (2008) 117–123.
97
[7] M.M.C. Bijveld, A.M. Bronstein, J.F. Golding, M.A. Gresty, Nauseogenicity of offvertical axis rotation vs. equivalent visual motion, Aviat. Space Environ. Med. 79 (2008) 661–665. [8] O.L. Bock, C.M. Oman, Dynamics of subjective discomfort in motion sickness as measured with a magnitude estimation method, Aviat. Space Environ. Med. 53 (1982) 773–777. [9] S.C. Nichols, S.V.G. Cobb, J.R. Wilson, Health and safety implications of virtual environments: measurement issues, Presence: Teleoperat. Virtual Environ. 6 (1997) 667–675. [10] M. Emoto, M. Sugawara, Y. Nojiri, Viewing angle dependency of visuallyinduced motion sickness in viewing wide-field images by subjective and autonomic nervous indices, Displays 29 (2008) 90–99. [11] J.T. Reason, A. Graybiel, Changes in subjective estimates of well-being during the onset and remission of motion sickness symptomatology in the slow rotation room, Aerospace Med. 41 (1970) 166–171. [12] J.F. Golding, Motion sickness susceptibility questionnaire revised and its relationship to other forms of sickness, Brain Res. Bull. 47 (1998) 507–516. [13] J.F. Golding, Predicting individual differences in motion sickness susceptibility by questionnaire, Pers. Individ. Dif. 41 (2006) 237–248. [14] G. Verbeke, G. Molenberghs, Linear Mixed Models for Longitudinal Data, Springer, New York, 2000. pp. 77–89. [15] P. Heagerty, K.Y. Liang, S.L. Zeger, P.J. Diggle, Analysis of Longitudinal Data, Oxford University Press, Oxford, 2002. pp. 169–189. [16] W.H. Greene, Econometric Analysis, Prentice-Hall, Upper Saddle River, 2000. pp. 499–524. [17] P. Good, Permutation, Parametric, and Bootstrap Tests of Hypotheses, Springer, New York, 1994. pp. 33–67. [18] K.L. Mills, M.J. Griffin, Effect of seating, vision and direction of horizontal oscillation on motion sickness, Aviat. Space Environ. Med. 71 (2000) 996– 1002. [19] N.A. Webb, M.J. Griffin, Optokinetic stimuli: motion sickness, visual acuity, and eye movements, Aviat. Space Environ. Med. 73 (2002) 351–358. [20] N.A. Webb, M.J. Griffin, Eye movement, vection, and motion sickness with foveal and peripheral vision, Aviat. Space Environ. Med. 74 (2003) 622–625. [21] M.J. Griffin, M.M. Newman, Visual field effects on motion sickness in cars, Aviat. Space Environ. Med. 75 (2004) 739–748. [22] J.M. Lentz, W.E. Collins, Motion sickness susceptibility and related behavioral characteristics in men and women, Aviat. Space Environ. Med. 48 (1977) 316– 322. [23] M. Turner, M.J. Griffin, Motion sickness in public road transport: passenger behavior and susceptibility, Ergonomics 42 (1999) 444–461. [24] M. Turner, M.J. Griffin, Motion sickness in public road transport: the relative importance of motion, vision and individual differences, Br. J. Psychol. 90 (1999) 519–530. [25] T. Probst, S. Krafczyk, W. Büchele, T. Brandt, Visuelle prävention der bewegungskrankheit im auto, (visual prevention of motion sickness in cars), Arch. Psychiatr. Nervenkr. 231 (1982) 409–421 (German). [26] J.E. Bos, S.N. MacKinnon, A. Patterson, Motion sickness symptoms in a ship motion simulator: effects of inside, outside, and no view, Aviat. Space Environ. Med. 76 (2005) 1111–1118. [27] H. Vogel, R. Kohlhaas, R.J. von Baumgarten, Dependence of motion sickness in automobiles on the direction of linear acceleration, Eur. J. Appl. Physiol. 48 (1982) 399–405. [28] J.F. Golding, H.M. Markey, J.R.R. Stott, The effects of motion direction, body axis, and posture on motion sickness induced by low frequency linear oscillation, Aviat. Space Environ. Med. 66 (1995) 1046–1051. [29] H. Schöne, On the role of gravity in human spatial orientation, Aerospace Med. 35 (1964) 764–772. [30] H. Jaslow, Human tilt perception in a dynamic environment, Aviat. Space Environ. Med. 50 (1979) 594–598. [31] A. Morimoto, N. Isu, T. Okumura, Y. Araki, A. Kawai, F. Masui, Screen design of onboard displays for reducing car sickness, in: Proc. of FISITA 2008 World Automotive Congress (CD-ROM), F2008-02-23, 2008.