Figure 1
From: Interim analyses of data as they accumulate in laboratory experimentation
An example of matched-pairs sequential analysis [23]. The experiment was designed to test the relative efficacy of: (A) a solution of inorganic chemicals dissolved in rainwater (Kaltaler solution); and (B) rainwater (tankwater), on the vase-life of roses. The end-point was time to first petal-fall. The graphical design relied on 2α = 0.05 or 0.01, 1 - β (power) = 0.95. θ was set at 0.90. Left panel: 2α set at 0.05. Kaltaler solution was superior to rainwater because the upper boundary was crossed at N = 11. A preference for B occurred only at preference number 5. Right panel: 2α set at 0.01. There was no significant difference because the right boundary was crossed at N = 23. A preference for B occurred only at preference numbers 5, 12, 17, 18 and 23. Confirmatory post hoc exact, two-sided, tests of the null hypothesis that for a single binomial p = 0.50 gave P = 0.012 for the left-hand panel (ie. P = 0.05), and P = 0.011 (ie. P > 0.01) for the right-hand panel (StatXact 5, Cytel Software Corporation, Cambridge MA).