172017-08-31 12:27:03PeterWatson162017-08-31 12:25:23PeterWatson152017-08-31 12:15:23PeterWatson142017-08-31 12:13:57PeterWatson132017-08-31 12:10:57PeterWatson122017-08-31 12:10:40PeterWatson112017-08-31 11:17:33PeterWatson102017-08-31 11:15:49PeterWatson92013-03-08 10:17:09localhostconverted to 1.6 markup82008-04-04 14:00:30PeterWatson72008-04-04 13:59:04PeterWatson62008-04-04 13:58:49PeterWatson52008-04-04 13:32:42PeterWatson42008-04-04 11:30:50PeterWatson32008-04-04 11:30:26PeterWatson22008-04-04 10:46:19PeterWatson12008-04-04 10:45:15PeterWatson

This spreadsheet computes the total sample size for a one sample t-test with given power. This can also be done using this on-line calculator. Both these programs use the more liberal normal quantiles as opposed to t quantiles for sample size evaluation. The web calculator here computes the number of subjects for a paired t-test using the formula on page 221 of Rosner B. Fundamentals of Biostatistics. 4th ed. Duxbury Press; 1995. Page 221. See also here. Note This formula is also incorrectly given on page 39 of Kraemer HC and Thiemann S (1987) How many subjects? Statistical power analysis in research. Sage.:London. (There should be a copy of this book in the CBU library). The formula should be (z(a) + z(P)) and not (z(a) - z(P) as in Kraemer and Thiemann (1987) for type I error, a, and power, P and z-value, z. The formula is: N for paired t-test = (1.96+0.842)2 / (difference in paired means / sd of the DIFFERENCE in the means)2 for 80% power and two-tailed type I error of 5% so Example Using the above formula if the raw paired subject difference is 0.303 and the sd of this difference equals 0.266 then we require 7.85 / (0.303/0.266)2 = 6 subjects.