3.2. Comparison of externally randomized intervals
The coverage fluctuations and expected length of the externally randomized split sample Wilson, U(−1/2, 1/2) Wilson
and Stevens intervals are compared in Fig. 3. The U(−1/2, 1/2) interval improves upon the split sample interval about as
much as the split sample interval improves upon the standard Wilson interval. Unlike the split sample and U(−1/2, 1/2)
intervals, the Stevens interval has coverage exactly equal to 1−α for all values of p, so that there are no coverage oscillations
at all. The difference in expected length between the intervals is quite small for all combinations of p and n and is decreasing
in n. The Stevens interval is shorter when p is close to 0 or 1, but wider when p is close to 1/2. When n = 20 the difference
in expected length is at most 0.018 and when n = 100 it is at most 0.002.
3.2. Comparison of externally randomized intervals
The coverage fluctuations and expected length of the externally randomized split sample Wilson, U(−1/2, 1/2) Wilson
and Stevens intervals are compared in Fig. 3. The U(−1/2, 1/2) interval improves upon the split sample interval about as
much as the split sample interval improves upon the standard Wilson interval. Unlike the split sample and U(−1/2, 1/2)
intervals, the Stevens interval has coverage exactly equal to 1−α for all values of p, so that there are no coverage oscillations
at all. The difference in expected length between the intervals is quite small for all combinations of p and n and is decreasing
in n. The Stevens interval is shorter when p is close to 0 or 1, but wider when p is close to 1/2. When n = 20 the difference
in expected length is at most 0.018 and when n = 100 it is at most 0.002.
การแปล กรุณารอสักครู่..
