error_matrix_hair: Hair Color Error/Confusion Matrix

In mispitools: Missing Person Identification Tools

Hair Color Error/Confusion Matrix

Description

Creates a 5x5 error matrix (also known as confusion matrix) that models the probability of observing each hair color given the true hair color. This accounts for observation errors in hair color classification.

The matrix rows represent the true hair color of the missing person, and columns represent the observed hair color. Each row sums to 1, indicating that some color must be observed.

Usage

error_matrix_hair(
  errorModel = c("custom", "uniform")[1],
  ep = 0.01,
  ep12 = 0.01,
  ep13 = 0.005,
  ep14 = 0.01,
  ep15 = 0.003,
  ep23 = 0.01,
  ep24 = 0.003,
  ep25 = 0.01,
  ep34 = 0.003,
  ep35 = 0.003,
  ep45 = 0.01
)

Arguments

errorModel	Character. Type of error model to use: "custom": Use specific error rates for each color pair (default) "uniform": Use a single error rate for all color pairs
ep	Numeric (0-1). Base error rate used when errorModel = "uniform". Represents the probability of confusing any two different colors. Default: 0.01.
ep12	Numeric. Error rate between colors 1 (Black) and 2 (Brown). Default: 0.01.
ep13	Numeric. Error rate between colors 1 (Black) and 3 (Blonde). Default: 0.005.
ep14	Numeric. Error rate between colors 1 (Black) and 4 (Red). Default: 0.01.
ep15	Numeric. Error rate between colors 1 (Black) and 5 (Gray/White). Default: 0.003.
ep23	Numeric. Error rate between colors 2 (Brown) and 3 (Blonde). Default: 0.01.
ep24	Numeric. Error rate between colors 2 (Brown) and 4 (Red). Default: 0.003.
ep25	Numeric. Error rate between colors 2 (Brown) and 5 (Gray/White). Default: 0.01.
ep34	Numeric. Error rate between colors 3 (Blonde) and 4 (Red). Default: 0.003.
ep35	Numeric. Error rate between colors 3 (Blonde) and 5 (Gray/White). Default: 0.003.
ep45	Numeric. Error rate between colors 4 (Red) and 5 (Gray/White). Default: 0.01.

Details

The error rates are symmetric: the probability of confusing color A with color B equals the probability of confusing B with A.

The diagonal elements (correct observations) are calculated to ensure each row sums to 1:

P(i|i) = 1 / (1 + \sum_{j \neq i} ep_{ij})

Lower error rates between dissimilar colors (e.g., black and blonde) and higher rates between similar colors (e.g., brown and red) reflect realistic observation patterns.

Value

A 5x5 numeric matrix where:

• Rows represent true hair colors (1-5)

• Columns represent observed hair colors (1-5)

• Cell (i,j) = P(observed color j | true color i)

• Each row sums to 1

• Diagonal elements are highest (correct observations)

Hair color codes: 1=Black, 2=Brown, 3=Blonde, 4=Red, 5=Gray/White.

References

Marsico FL, et al. (2023). "Likelihood ratios for non-genetic evidence in missing person cases." Forensic Science International: Genetics, 66, 102891. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/j.fsigen.2023.102891")}

See Also

cpt_missing_person which uses this matrix, lr_hair_color for hair color LR calculations.

Examples

# Default custom error model
emat <- error_matrix_hair()
print(round(emat, 4))

# Verify rows sum to 1
rowSums(emat)

# Uniform error model with 2% error rate
emat_uniform <- error_matrix_hair(errorModel = "uniform", ep = 0.02)
print(round(emat_uniform, 4))

# Higher error rates for similar colors
emat_custom <- error_matrix_hair(
  errorModel = "custom",
  ep12 = 0.05,  # Black-Brown confusion more likely
  ep23 = 0.05,  # Brown-Blonde confusion more likely
  ep34 = 0.05   # Blonde-Red confusion more likely
)

mispitools documentation built on Aug. 26, 2026, 1:08 a.m.