Within-subject, repeated-measures designs are frequently used in studies in experimental psychology. They achieve higher statistical power than between-subject designs, because the effects of the experimental manipulation can be analysed within each participant, which allows to discount variance between participants. However, this comes with the challenge of how to display in scientific graphs the within-subject variance for the comparison of paired measurements and the between-subject variance for each measurement separately. In traditional bar-like graphs, where the different levels of the paired measurements are separated on the abscissa, and the dependent variable is shown on the ordinate, one can display either between- or within-subject confidence intervals.
We describe a method how to display both, between and within-subject confidence intervals in scatter plots. A diagonal confidence interval can be used to infer the statistical significance of the difference between two paired measures by comparing it against the positive main diagonal in the scatter plot. We evaluated such diagonal confidence intervals by asking scientists to interpret between- and within-subject effects in four different types of graphs: scatter plots with classical between-subject confidence intervals along the horizontal and vertical axes, scatter plots with additional diagonal, within-subject confidence intervals and traditional bar-like graphs with either between- or within-subject confidence intervals. The results showed that between- and within-subject effects can be interpreted with high certainty and accuracy in scatter plots. Thus, scatter plots with diagonal within-subject confidence intervals are easy to generate and can be interpreted intuitively.