Statscraft 0.15.0, release v0.15.0
WEIGHTED KAPPA
Cohen’s weighted kappa: how well two raters agree on an ordered scale.
Syntax
WEIGHTED KAPPA varlist | rowlist BY columnlist
[/MISSING CLASSMISSING=TABLE | INCLUDE]
[/CRITERIA [ASYMPTOTICCILEVEL=95] [WEIGHTING=LINEAR | QUADRATIC] [IGNORECASE=FALSE | TRUE]]
[/PRINT [TABULATION(ALL | rowlist BY columnlist)] [CASE(UPPER | LOWER)]
[ORDER(ASCENDING | DESCENDING)]].
One list pairs each variable with every later one, the earlier as the row; two lists pair every
row variable with every column variable, never a variable with itself. Both variables of a pair
must be numeric, or both string. The k categories of a pair are the values either rater used,
numbers by size and strings alphabetically, and two categories d places apart get the agreement
weight 1 - d/(k - 1) (LINEAR) or 1 - d^2/(k - 1)^2 (QUADRATIC). Each pair drops the cases missing
on either of its variables; INCLUDE keeps user-missing values, each as a category of its own.
The kappa table has a row per pair: N, the kappa, its asymptotic standard error, z with its
two-sided significance (from the standard error under chance agreement) and the confidence
interval at ASYMPTOTICCILEVEL percent. IGNORECASE=TRUE compares strings regardless of case and
shows them in the case CASE names. TABULATION adds the table of counts of every pair, or of the
pairs the two subsets make, its categories in ORDER. Case weights are rounded to whole numbers
(0.5 or less drops the case). A pair with fewer than two cases or only one category is skipped
with a warning. KEYWORD(value) works as KEYWORD=value; ASYMPTOTIC_CILEVEL and IGNORE_CASE are
other spellings.
Example
WEIGHTED KAPPA nurse_a nurse_b BY doctor /CRITERIA WEIGHTING=QUADRATIC ASYMPTOTICCILEVEL=90
/PRINT TABULATION(ALL).
statscraft help WEIGHTED KAPPA prints the same text in the terminal. All commands