Statistical analyses were conducted using Stata/IC 14 for Macintosh (Stata Corp, College Station, TX). We examined the impact of viewing condition, VA, and order on IA scores with a linear mixed model with age, gender, and education as covariates, and participant and video clip as fully-crossed random effects. We examined the effects of viewing condition on head motion using a linear mixed model that included VA and order as fixed factors, and age, gender, and education as covariates, and participant and video clip as fully crossed random effects. For that analysis, we used the logarithm of head rotation, which approximately normalized (Shapiro-Wilks test; z = 2.14; P = 0.02) the otherwise skewed distribution. We examined the effects of viewing condition on responses to the discomfort questionnaire in a series of mixed effects, ordered logistic regressions that included head motion and order as fixed factors, and age, gender, and education as covariates, and participant as a random effect. Viewing condition was included in the models described elsewhere in this article as the full 2 × 2 factorial, with screen size, zoom magnification, and their interaction.
Linear mixed models are robust to certain missing data, and the random effects account for repeated measures by including terms for individual differences between subjects (e.g., some people are more loquacious or more observant) and between video clips (e.g., some clips are harder to describe or have less material to report). Because we have previously found that IA scores can vary with age, education, and gender,
4,17–19 we included these demographic factors as covariates. To determine whether subjects were aware of their ability (and, conversely, their limitations), we fit a mixed-effects ordered logistic model to the data from question 4 that had IA score, VA, and order as fixed factors, and participant as a random effect. As a compromise between the risks of type I (multiple comparisons problem) and type II (small sample size) errors, we accepted a
P value of 0.01 or less as statistically significant, and report terms with 0.10 ≥
P > 0.01 as trends.