Skip to contents

Linear Discriminant Analysis for morphometric data

Source: R/stat_lda.R
stat_lda.Rd

Perform LDA on coefficient data from morphometric analyses. Supports both bare response names (auto-detects all coe columns) and full formula syntax.

Usage

stat_lda(data, formula, ...)

Arguments

data

A tibble with coefficient columns

formula

A formula or bare column name specifying response and predictors:

  • Bare name: species → auto-detects all coe columns

  • Formula: species ~ . → all coe columns

  • Formula: species ~ . + length → all coe + covariate

  • Formula: species ~ coe1 → specific coe column

  • Formula: species ~ coe1 + coe2 + length → multiple specific

...

Additional arguments passed to MASS::lda()

Value

An object of class c("stat_lda", "momstats") containing:

  • data: Original tibble (unchanged)

  • model: List containing both the LDA model and CV predictions

  • method: "lda"

  • call: The function call

  • formula: Formula used

  • response_col: Name of response column

  • predictor_cols: All predictor column names (coe + others)

  • kept_predictor_cols: Predictor columns after removing collinear

  • n_groups: Number of groups

  • n_ld: Number of discriminant functions

Details

stat_lda() provides a unified interface for LDA on morphometric coefficient data. It handles:

  • Flexible formula syntax: Bare response names or full formulas

  • Automatic coefficient detection: Finds all coe columns via class

  • Coefficient unfolding: Expands list-columns to individual harmonic columns

  • Multiple coefficient types: Can combine different coefficient columns

  • Covariates: Add non-morphometric predictors (size, environmental variables)

  • Cross-validation: Always runs with CV=TRUE for reliable predictions

Formula syntax

The formula specifies response and predictors:

  • species → Response only, auto-detect all coe columns

  • species ~ . → All coe columns (explicit)

  • species ~ coe1 → Specific coe column

  • species ~ coe1 + coe2 → Multiple specific coe columns

  • species ~ . + length → All coe plus a covariate

Centering and scaling

LDA internally centers and scales the data as part of the algorithm. No manual preprocessing is needed.

Cross-validation

The function runs LDA twice:

  1. CV = FALSE: Builds the full model (discriminant functions, loadings)

  2. CV = TRUE: Generates cross-validated predictions

Both are stored in $model for easy access.

Getting results

Use collect() to add predictions to your data:

lda <- boteft %>% stat_lda(type)
boteft_pred <- collect(lda)  # Adds pred, prob columns

See also

MASS::lda(), collect.stat_lda(), plot.stat_lda()

Examples

if (FALSE) { # \dontrun{
# Basic LDA - auto-detects all coe columns
lda1 <- boteft %>% stat_lda(type)

# Explicit formula
lda2 <- boteft %>% stat_lda(type ~ .)

# Specific coe column
lda3 <- boteft %>% stat_lda(type ~ coe)

# With covariate
lda4 <- boteft %>% stat_lda(type ~ . + length)

# Add predictions to data
boteft_pred <- collect(lda1)  # Adds pred, prob

# Add LD scores too
boteft_ld <- collect(lda1, retain = 2)  # Adds pred, prob, LD1, LD2

# Plot results
plot(lda1)  # Discrimination plot (default)
plot(lda1, type = "loadings")
plot(lda1, color = type)  # Color by true groups
plot(lda1, labels = pred)  # Label by predictions
} # }