From symptoms to causes

Figure 12.3: The symptoms are observed, but the causes are not directly measured. Researchers face an inverse problem, which is to speculate on the causes based on observed symptoms. The trouble is that each symptom may have many possible causes, some of which might not be related to the VR experience.
\begin{figure}\centerline{\psfig{file=figs/symptoms.eps,width=\columnwidth}}\end{figure}

The symptoms are the effect, but what are their causes? See Figure 12.3. The unfortunate problem for the scientist or evaluator of a VR system is that only the symptoms are observable. Any symptom could have any number of direct possible causes. Some of them may be known and others may be impossible to determine. Suppose, for example, that a user has developed mild nausea after $ 5$ minutes of a VR experience. What are the chances that he would have gotten nauseated anyway because he rode his bicycle to the test session and forgot to eat breakfast? What if he has a hangover from alcoholic drinks the night before? Perhaps a few users such as this could be discarded as outliers, but what if there was a large festival the night before which increased the number of people who are fatigued before the experiment? Some of these problems can be handled by breaking them into groups that are expected to have low variability; see Section 12.4. At the very least, one should probably ask them beforehand if they feel nauseated; however, this could even cause them to pay more attention to nausea, which generates a bias.

Even if it is narrowed down that the cause was the VR experience, this determination may not be narrow enough to be useful. Which part of the experience caused it? The user might have had no problems were it not for $ 10$ seconds of stimulus during a $ 15$-minute session. How much of the blame was due to the hardware versus the particular content? The hardware might be as comfortable as an optokinetic drum, which essentially shifts the blame to the particular images on the drum.

Questions relating to cause are answered by finding statistical correlations in the data obtained before, during, and after the exposure to VR. Thus, causation is not determined through directly witnessing the cause and its effect in the way as witnessing the effect of a shattered glass which is clearly caused by dropping it on the floor. Eliminating irrelevant causes is an important part of the experimental design, which involves selecting users carefully and gathering appropriate data from them in advance. Determining more specific causes requires more experimental trials. This is complicated by the fact that different trials cannot be easily applied to the same user. Once people are sick, they will not be able to participate, or would at least give biased results that are difficult to compensate for. They could return on different days for different trials, but there could again be issues because of adaptation to VR, including the particular experiment, and simply being in a different health or emotional state on another occasion.

Steven M LaValle 2020-11-11