Plot ANOVA Relationships Matrix

This function plots the relationship between continuous and categorical variables using ANOVA or Kruskal-Wallis tests. It generates a "heatmap" with points colored and shaped based on statistical significance and effect size. When generalized eta-squared is unavailable, raw p-values are colored with scale_color_pvalue(); the FDR-corrected plot uses adjusted p-values.

Usage

PlotAnovaRelationshipsMatrix(
  data,
  CatVars,
  ContVars,
  covariates = NULL,
  Relabel = TRUE,
  Parametric = TRUE,
  Ordinal = FALSE,
  min_n = 4,
  eps = 1e-08,
  Data = lifecycle::deprecated(),
  fdr_scope = c("matrix", "per_outcome", "per_predictor"),
  Covariates = lifecycle::deprecated()
)

Arguments

data

The data frame containing the variables of interest.

CatVars

Character vector of categorical variable names.

ContVars

Character vector of continuous variable names.

covariates

Optional character vector of covariate names for ANCOVA analysis.

Relabel

Logical indicating whether to relabel variables with their labels (default is TRUE).

Parametric

Logical indicating whether to use parametric (ANOVA) or non-parametric (Kruskal-Wallis) tests (default is TRUE).

Ordinal

Logical, indicating whether ordinal variables should be considered.

min_n

Minimum number of complete observations required for a tested relationship.

eps

Small positive value used to avoid zero-size plotting artifacts.

Data

Deprecated (since 19.15.0). Use data instead.

fdr_scope

Either "matrix" (default) or "per_outcome", passed to ApplyFDRCorrection(). "matrix" corrects across all p-values at once (historical behavior). "per_outcome" corrects separately within each outcome: outcomes are the continuous variables (ContVars).

Covariates

Deprecated (since 19.15.0). Use covariates instead.

Value

A list containing three ggplot objects: p (scatter plot without multiple comparison correction), p_FDR (scatter plot with FDR correction), and pvaltable (data frame of p-values and significance).

Parametric and non-parametric tests

ANOVA assumes roughly normal residuals and similar variance across groups. Biomarker concentrations are often right-skewed, and group sizes are often uneven, both of which make the F-test unreliable.

Parametric = FALSE switches to Kruskal-Wallis, which compares ranks rather than means and assumes neither normality nor equal variance. The matrix is built and read exactly the same way.

Comparing the two p-value tables shows which conclusions depended on the choice of test. Relationships that hold under both are the ones to trust. Where they disagree, the direction is informative: significant only under ANOVA suggests the result is being carried by skew or a few extreme values, while significant only under Kruskal-Wallis suggests a real shift in the bulk of the distribution that outliers were masking from the mean-based test.

Kruskal-Wallis is a rank test and has no ANCOVA form, so covariates requires Parametric = TRUE.

Examples

data(SampleData)
data(SampleVariableTypes)

# Attach labels and factor levels for readable axes
Labelled <- RevalueData(SampleData, SampleVariableTypes)$RevaluedData

result <- PlotAnovaRelationshipsMatrix(
  Labelled,
  CatVars = c("Diagnosis", "sex", "Genotype"),
  ContVars = c("age", "ACE_CD143_Angiotensin_Converti",
               "ACTH_Adrenocorticotropic_Hormon", "AXL", "Adiponectin",
               "Alpha_1_Antichymotrypsin", "Alpha_1_Antitrypsin",
               "Alpha_1_Microglobulin", "Alpha_2_Macroglobulin",
               "Apolipoprotein_A1")
)

# Raw p-value associations
result$Unadjusted$plot


# FDR-adjusted associations
result$FDRCorrected$plot
#> Warning: Using size for a discrete variable is not advised.


# The same matrix using Kruskal-Wallis instead of ANOVA
result_NonParametric <- PlotAnovaRelationshipsMatrix(
  Labelled,
  CatVars = c("Diagnosis", "sex", "Genotype"),
  ContVars = c("age", "ACE_CD143_Angiotensin_Converti",
               "ACTH_Adrenocorticotropic_Hormon", "AXL", "Adiponectin",
               "Alpha_1_Antichymotrypsin", "Alpha_1_Antitrypsin",
               "Alpha_1_Microglobulin", "Alpha_2_Macroglobulin",
               "Apolipoprotein_A1"),
  Parametric = FALSE
)

result_NonParametric$Unadjusted$plot

result_NonParametric$FDRCorrected$plot
#> Warning: Using size for a discrete variable is not advised.


# Where the two tests disagree
cols_Key <- c("CategoricalVariable", "ContinuousVariable", "p")
df_Compare <- merge(
  result$Unadjusted$PvalTable[, cols_Key],
  result_NonParametric$Unadjusted$PvalTable[, cols_Key],
  by = c("CategoricalVariable", "ContinuousVariable"),
  suffixes = c("_ANOVA", "_KruskalWallis")
)
df_Compare$AgreesAt05 <-
  (df_Compare$p_ANOVA < 0.05) == (df_Compare$p_KruskalWallis < 0.05)

htmltools::browsable(htmltools::HTML(as.character(
  FreezeTableHeader(
    dplyr::mutate(
      df_Compare,
      dplyr::across(dplyr::where(is.numeric), \(x) signif(x, 3))
    ),
    height = "320px", full_width = TRUE
  )
)))
 CategoricalVariable 
   
 ContinuousVariable 
   
 p_ANOVA 
   
 p_KruskalWallis 
   
 AgreesAt05 
  
 Diagnosis 
   
 ACE_CD143_Angiotensin_Converti 
   
 8.87e-01 
   
 7.45e-01 
   
 TRUE 
  
 Diagnosis 
   
 ACTH_Adrenocorticotropic_Hormon 
   
 3.04e-01 
   
 1.85e-01 
   
 TRUE 
  
 Diagnosis 
   
 AXL 
   
 2.62e-01 
   
 4.80e-01 
   
 TRUE 
  
 Diagnosis 
   
 Adiponectin 
   
 4.70e-02 
   
 3.03e-02 
   
 TRUE 
  
 Diagnosis 
   
 Alpha_1_Antichymotrypsin 
   
 8.00e-03 
   
 7.97e-03 
   
 TRUE 
  
 Diagnosis 
   
 Alpha_1_Antitrypsin 
   
 1.50e-04 
   
 1.13e-04 
   
 TRUE 
  
 Diagnosis 
   
 Alpha_1_Microglobulin 
   
 4.00e-03 
   
 1.95e-03 
   
 TRUE 
  
 Diagnosis 
   
 Alpha_2_Macroglobulin 
   
 2.70e-02 
   
 3.43e-02 
   
 TRUE 
  
 Diagnosis 
   
 Apolipoprotein_A1 
   
 6.87e-01 
   
 7.96e-01 
   
 TRUE 
  
 Diagnosis 
   
 age 
   
 5.54e-01 
   
 7.06e-01 
   
 TRUE 
  
 Genotype 
   
 ACE_CD143_Angiotensin_Converti 
   
 4.19e-01 
   
 4.14e-01 
   
 TRUE 
  
 Genotype 
   
 ACTH_Adrenocorticotropic_Hormon 
   
 6.09e-01 
   
 5.13e-01 
   
 TRUE 
  
 Genotype 
   
 AXL 
   
 7.27e-01 
   
 6.40e-01 
   
 TRUE 
  
 Genotype 
   
 Adiponectin 
   
 6.46e-01 
   
 6.29e-01 
   
 TRUE 
  
 Genotype 
   
 Alpha_1_Antichymotrypsin 
   
 9.47e-01 
   
 9.57e-01 
   
 TRUE 
  
 Genotype 
   
 Alpha_1_Antitrypsin 
   
 9.72e-01 
   
 9.80e-01 
   
 TRUE 
  
 Genotype 
   
 Alpha_1_Microglobulin 
   
 4.03e-01 
   
 4.39e-01 
   
 TRUE 
  
 Genotype 
   
 Alpha_2_Macroglobulin 
   
 1.40e-01 
   
 1.48e-01 
   
 TRUE 
  
 Genotype 
   
 Apolipoprotein_A1 
   
 3.05e-01 
   
 1.98e-01 
   
 TRUE 
  
 Genotype 
   
 age 
   
 3.05e-01 
   
 5.18e-01 
   
 TRUE 
  
 sex 
   
 ACE_CD143_Angiotensin_Converti 
   
 2.90e-01 
   
 3.97e-01 
   
 TRUE 
  
 sex 
   
 ACTH_Adrenocorticotropic_Hormon 
   
 1.34e-01 
   
 1.82e-01 
   
 TRUE 
  
 sex 
   
 AXL 
   
 6.52e-01 
   
 3.67e-01 
   
 TRUE 
  
 sex 
   
 Adiponectin 
   
 4.36e-01 
   
 4.15e-01 
   
 TRUE 
  
 sex 
   
 Alpha_1_Antichymotrypsin 
   
 1.20e-06 
   
 1.60e-06 
   
 TRUE 
  
 sex 
   
 Alpha_1_Antitrypsin 
   
 3.45e-05 
   
 6.39e-05 
   
 TRUE 
  
 sex 
   
 Alpha_1_Microglobulin 
   
 0.00e+00 
   
 0.00e+00 
   
 TRUE 
  
 sex 
   
 Alpha_2_Macroglobulin 
   
 6.10e-02 
   
 2.11e-02 
   
 FALSE 
  
 sex 
   
 Apolipoprotein_A1 
   
 1.44e-04 
   
 2.31e-04 
   
 TRUE 
  
 sex 
   
 age 
   
 7.40e-02 
   
 7.30e-02 
   
 TRUE 
  

# Covariate adjustment (parametric only)
PlotAnovaRelationshipsMatrix(
  Labelled,
  CatVars = c("Diagnosis", "sex"),
  ContVars = c("AXL", "Adiponectin", "Alpha_1_Antitrypsin"),
  covariates = "age"
)$FDRCorrected$plot
#> Warning: Using size for a discrete variable is not advised.