Skip to contents

Calculate the conditional Mahalanobis distance for any variables.

Usage

cond_maha(
  data,
  R,
  v_dep,
  v_ind = NULL,
  v_ind_composites = NULL,
  mu = 0,
  sigma = 1,
  use_sample_stats = FALSE,
  label = NA
)

Arguments

data

Data.frame with the independent and dependent variables. Unless mu and sigma are specified, data are assumed to be z-scores.

R

Correlation among all variables.

v_dep

Vector of names of the dependent variables in your profile.

v_ind

Vector of names of independent variables you would like to control for.

v_ind_composites

Vector of names of independent variables that are composites of dependent variables

mu

A vector of means. A single value means that all variables have the same mean.

sigma

A vector of standard deviations. A single value means that all variables have the same standard deviation

use_sample_stats

If TRUE, estimate R, mu, and sigma from data. Only complete cases are used (i.e., no missing values in v_dep, v_ind, v_ind_composites).

label

optional tag for labeling output

Value

a list with the conditional Mahalanobis distance

  • dCM = Conditional Mahalanobis distance

  • dCM_df = Degrees of freedom for the conditional Mahalanobis distance

  • dCM_p = A proportion that indicates how unusual this profile is compared to profiles with the same independent variable values. For example, if dCM_p = 0.88, this profile is more unusual than 88 percent of profiles after controlling for the independent variables.

  • dM_dep = Mahalanobis distance of just the dependent variables

  • dM_dep_df = Degrees of freedom for the Mahalanobis distance of the dependent variables

  • dM_dep_p = Proportion associated with the Mahalanobis distance of the dependent variables

  • dM_ind = Mahalanobis distance of just the independent variables

  • dM_ind_df = Degrees of freedom for the Mahalanobis distance of the independent variables

  • dM_ind_p = Proportion associated with the Mahalanobis distance of the independent variables

  • v_dep = Dependent variable names

  • v_ind = Independent variable names

  • v_ind_singular = Independent variables that can be perfectly predicted from the dependent variables (e.g., composite scores)

  • v_ind_nonsingular = Independent variables that are not perfectly predicted from the dependent variables

  • data = data used in the calculations

  • d_ind = independent variable data

  • d_inp_p = Assuming normality, cumulative distribution function of the independent variables

  • d_dep = dependent variable data

  • d_dep_predicted = predicted values of the dependent variables

  • d_dep_deviations = d_dep - d_dep_predicted (i.e., residuals of the dependent variables)

  • d_dep_residuals_z = standardized residuals of the dependent variables

  • d_dep_cp = conditional proportions associated with standardized residuals

  • d_dep_p = Assuming normality, cumulative distribution function of the dependent variables

  • R2 = Proportion of variance in each dependent variable explained by the independent variables

  • zSEE = Standardized standard error of the estimate for each dependent variable

  • SEE = Standard error of the estimate for each dependent variable

  • ConditionalCovariance = Covariance matrix of the dependent variables after controlling for the independent variables

  • distance_reduction = 1 - (dCM / dM_dep) (Degree to which the independent variables decrease the Mahalanobis distance of the dependent variables. Negative reductions mean that the profile is more unusual after controlling for the independent variables. Returns 0 if dM_dep is 0.)

  • variability_reduction = 1 - sum((X_dep - predicted_dep) ^ 2) / sum((X_dep - mu_dep) ^ 2) (Degree to which the independent variables decrease the variability the dependent variables (X_dep). Negative reductions mean that the profile is more variable after controlling for the independent variables. Returns 0 if X_dep == mu_dep)

  • mu = Variable means

  • sigma = Variable standard deviations

  • d_person = Data frame consisting of Mahalanobis distance data for each person

  • d_variable = Data frame consisting of variable characteristics

  • label = label slot

Examples

library(unusualprofile)
library(simstandard)

m <- "
Gc =~ 0.85 * Gc1 + 0.68 * Gc2 + 0.8 * Gc3
Gf =~ 0.8 * Gf1 + 0.9 * Gf2 + 0.8 * Gf3
Gs =~ 0.7 * Gs1 + 0.8 * Gs2 + 0.8 * Gs3
Read =~ 0.66 * Read1 + 0.85 * Read2 + 0.91 * Read3
Math =~ 0.4 * Math1 + 0.9 * Math2 + 0.7 * Math3
Gc ~ 0.6 * Gf + 0.1 * Gs
Gf ~ 0.5 * Gs
Read ~ 0.4 * Gc + 0.1 * Gf
Math ~ 0.2 * Gc + 0.3 * Gf + 0.1 * Gs"
# Generate 10 cases
d_demo <- simstandard::sim_standardized(m = m, n = 10)

# Get model-implied correlation matrix
R_all <- simstandard::sim_standardized_matrices(m)$Correlations$R_all

cond_maha(data = d_demo,
          R = R_all,
          v_dep = c("Math", "Read"),
          v_ind = c("Gf", "Gs", "Gc"))
#> Conditional Mahalanobis Distance = 1.0632, df = 2, p = 0.4317 Conditional Mahalanobis Distance = 1.1124, df = 2, p = 0.4614 Conditional Mahalanobis Distance = 0.4195, df = 2, p = 0.0842 Conditional Mahalanobis Distance = 0.1944, df = 2, p = 0.0187 Conditional Mahalanobis Distance = 1.3498, df = 2, p = 0.5979 Conditional Mahalanobis Distance = 1.1014, df = 2, p = 0.4548 Conditional Mahalanobis Distance = 0.8975, df = 2, p = 0.3315 Conditional Mahalanobis Distance = 1.1225, df = 2, p = 0.4674 Conditional Mahalanobis Distance = 2.2328, df = 2, p = 0.9173 Conditional Mahalanobis Distance = 1.3840, df = 2, p = 0.6162