Skip to contents

Detect variable types for dynamic correlation method selection

Source: R/14_correlations.R
detect_variable_types.Rd

Classifies each column as "binary", "ordinal", or "continuous". The ordinal threshold defaults to <=21 unique values, which covers (a) the VADER-shaped sentiment scale [-1, 1] quantized at 0.1 (21 distinct levels), (b) the Likert-style 5/7/9/11-point scales common in survey research, and (c) AI-elicited intensity / confidence scores on a small integer grid. The earlier threshold of 7 silently classified VADER sentiment as continuous on the large run, which then dispatched to Pearson (point-biserial for binary x quantized-sentiment pairs) – methodologically wrong for an ordinal support. Spearman is correct when either variable is rank-orderable but not interval-scaled .

Usage

detect_variable_types(corr_data, ordinal_max = 21L)

Arguments

corr_data

Numeric tibble from prepare_correlation_data()

ordinal_max

Integer; upper bound on distinct values for the ordinal classification. Default 21L. Datasets with finer-grained ordinal scales (e.g. 0-50 Likert) can override.

Value

Named character vector with types per column