Computes descriptive Spearman correlations over paired observations, using
average ranks for ties. Each observation must occur once per complete group
key. Include the independent sampling unit in groups; no independence,
eligibility, causal interpretation or inferential p value is established.
Missingness is paired: a row is available only when both values are known.
The default suppresses correlation when any pair is unavailable. Explicit
available-case analysis retains the supplied and unavailable pair counts.
Usage
CorrelateGroupedRanks(
data,
groups,
id,
x,
y,
min_pairs = 3L,
missing = c("propagate", "available")
)Arguments
- data
Data frame with group keys, observation IDs and paired values.
- groups
Nonempty character vector of distinct group columns with nonempty character values and no missing keys.
- id
Character column naming observations within each group.
- x, y
Distinct numeric value columns. Finite numbers and NA are allowed; NaN and infinite values are rejected.
- min_pairs
Integer minimum available pair count, at least two.
- missing
Missingness policy:
"propagate"or explicit"available".
Value
Data frame sorted by group keys with n_supplied, n_available,
n_unavailable, available-pair n_unique_x and n_unique_y, rho,
status, and missing_policy. Status precedence is UNAVAILABLE,
INSUFFICIENT, MISSING_PAIRS (propagation), CONSTANT_BOTH, CONSTANT_X,
CONSTANT_Y, then COMPLETE or PARTIAL. Suppressed correlations are NA.
Empty input returns a typed empty table; absent groups are not invented.
Examples
d <- data.frame(sample = "one", cell = letters[1:4],
a = c(1, 1, 3, 4), b = c(4, 3, 2, 1))
CorrelateGroupedRanks(d, "sample", "cell", "a", "b")
#> sample n_supplied n_available n_unavailable n_unique_x n_unique_y rho
#> 1 one 4 4 0 3 4 -0.9486833
#> status missing_policy
#> 1 COMPLETE propagate