More rank tests (sign, median, Jonckheere-Terpstra, Kolmogorov-Smirnov)

In brief

Four rank tests besides the default: they are under "No normal distribution (rank test)" in the field "Which rank test?". Statable never switches by itself.

What for

• Sign test (before-after): counts only whether the value rose or fell. Pairs without a difference are dropped. Statable always calculates exactly; SPSS approximates from 25 pairs.
• Median test: share above the overall median in each group; values on the median count as "≤". With two groups Fisher's exact test applies; SPSS shows the Yates χ² from 31 cases.
• Jonckheere-Terpstra: trend across ordered groups (three or more). The order is visible and reversible; two-sided as SPSS, one-sided selectable.
• Kolmogorov-Smirnov (exactly two groups): largest distance between the distributions, so spread and shape too.

Requirements

Independent groups (median, J-T, K-S) or the same people twice (sign). Blocked with a case weight.

Reading the result

Calculated exactly as long as possible (Jonckheere-Terpstra up to 2,000,000 assignments, K-S up to n₁ · n₂ = 10,000), otherwise an approximation; the line "Calculated" says so. Tied values stay tied.

Reporting the result

Jonckheere-Terpstra test, JT = 40.5, exact two-sided p = .013, N = 12

Limits

PSPP calculates Jonckheere-Terpstra wrongly with tied values, the median test only approximately and K-S for two samples not at all; Statable calculates all four itself. The median test has little power when many values lie on the median.

Source

Hollander, M., Wolfe, D. A., & Chicken, E. (2014). Nonparametric Statistical Methods (3rd ed.). Wiley.