Prepare external features for exploratory grouping

Keep one row per Person, rater, or other entity, review feature availability, and optionally attach user-supplied reasons for missing values.

Usage

mfrm_features(data, id, features, missing_reasons = NULL)

# S3 method for class 'mfrm_features'
print(x, ...)

# S3 method for class 'mfrm_features'
summary(object, ...)

Arguments

data: A data frame with one row per entity. Repeated rating rows must be summarized or joined to an entity-level table before calling this function.
id: Name of the unique, nonmissing identifier column. Character, factor, or finite numeric identifiers are stored as character strings without trimming.
features: Explicit character vector of feature column names, excluding the identifier. Numeric, character, factor, ordered factor, and logical columns are supported. Character columns become nominal factors; logical columns represent symmetric binary features. Ordered factor levels retain their declared order. Dates, lists, and matrices must be converted explicitly.
missing_reasons: Optional data frame with columns ID, Feature, and Reason, one row per annotated missing cell. IDs refer to id; features must be selected columns. Reasons for observed or unknown cells are refused. Unannotated missing cells are labelled "Not supplied". Reasons are supplied by the user, not inferred missing-data mechanisms.
x, object: An object returned by the corresponding function.
...: Reserved for method compatibility.

Value

An mfrm_features object containing data, id, features, a feature_summary, a row_summary, and missing (all missing cells and their reasons). No values are imputed or rows discarded.

Details

Numeric NA and NaN are missing. Infinite numeric values and blank categorical labels are refused; replace missing markers with NA explicitly. Constant and entirely missing features remain available for review but cannot be used by mfrm_cluster_pam(). IDs and unselected columns do not enter distances. These functions are intended for external attributes such as training, experience, or specialization. They do not propagate uncertainty from estimated ability, severity, or fit statistics. See vignette("mfrmr-external-features") for a complete rater-attribute example including missingness review and multiple imputation.

Examples

# Fictional rater attributes; experience is measured in completed years.
raters <- data.frame(
  Rater = paste0("R", 1:6),
  ExperienceYears = c(1, NA, 3, 12, 13, 14),
  Specialty = c("Language", "Language", "Language", "Science", "Science", "Science")
)
reasons <- data.frame(ID = "R2", Feature = "ExperienceYears", Reason = "Not recorded")
features <- mfrm_features(raters, "Rater", c("ExperienceYears", "Specialty"),
                          missing_reasons = reasons)
summary(features)
#>           Feature    Type Observed Missing Distinct
#> 1 ExperienceYears Numeric        5       1        5
#> 2       Specialty Nominal        6       0        2
features$missing
#>   ID         Feature       Reason
#> 1 R2 ExperienceYears Not recorded
features$row_summary
#>   ID MissingFeatures Complete
#> 1 R1               0     TRUE
#> 2 R2               1    FALSE
#> 3 R3               0     TRUE
#> 4 R4               0     TRUE
#> 5 R5               0     TRUE
#> 6 R6               0     TRUE