Figure 1.

(a) An outline of our sampling approach for estimating split-half reliabilities. The procedure is appropriate for all kinds of data obtained from tests that are composed of multiple items or repeated trials. On any iteration i, the overall set of items or trials was split into two halves of equal length, A_i and B_i, respectively. For each patient j with j = (1, . . ., J), the test score of interest was calculated separately for the two test halves, yielding two scores, A_ij and B_ij for each individual. The Pearson correlation coefficient r(i) was calculated across all A_ij and B_ij pairings, which was finally corrected for test length by the well-known Spearman–Brown formula (r_SB(i); Brown, 1910; Spearman, 1910). The emergent r_SB(i) was stored. Further inferences were based on the distribution of all r_SB(i) for I = 100,000 iterations. The median and the 95% highest density intervals (HDI; shown as horizontal lines in red color in the histograms) of the emergent distribution were computed as estimates of the central tendency and uncertainty of split-half reliability estimates, respectively. (b) For the purpose of our reliability analysis, we considered random and systematic splits of scores that were derived from test halves. Exemplary test halves are shown in black and white for illustration. Random test splits such as the exemplary one shown here were used for the sampling approach (as described in detail in Part (a) of this figure). Systematic test splits included different ways to split the test into halves. With 48 trials, a relatively large number of systematic splits can be construed that differ with regard to their “grain size.” The odd/even split (with a grain size of 1 trial) and the first/second test half split (with a grain size of 24 trials) are most commonly utilized for that purpose. In addition to these two splits, we also divided the test into systematic halves based on grain sizes of 2, 3, 4, 6, 8, and 12 consecutive trials. The emergent reliability estimates are reported.