Skip to content

EDA for ANOVA etc.

🎯 Purpose

This guide outlines how to perform exploratory data analysis (EDA) when preparing for statistical models that test group mean differences, including:

  • ANOVA (single DV, multiple groups)
  • ANCOVA (DV with covariate adjustment)
  • MANOVA (multiple DVs across groups)
  • MANCOVA (multiple DVs with covariates)

It focuses on preparing and understanding the data before modeling begins.

🟦 1. EDA for ANOVA

📌 Goal

Explore how a single dependent variable differs across categories.

🔍 Key EDA Steps

  • Identify the categorical group variable (independent variable)
  • Explore distribution of the DV by group
  • Assess group balance (sample sizes)

✅ EDA Tools

  • Descriptive statistics by group (mean, std, count)
  • Boxplots or violin plots of DV
  • Levene's test for equal variance

📊 Python Example

df.groupby('group')['outcome'].describe()
sns.boxplot(x='group', y='outcome', data=df)

🟨 2. EDA for ANCOVA

📌 Goal

Explore group differences in DV while adjusting for a continuous covariate.

🔍 Key EDA Steps

  • Confirm the covariate is linearly related to the DV
  • Assess whether the covariate varies across groups
  • Test for group × covariate interaction visually and numerically

✅ EDA Tools

  • Scatterplot of DV vs covariate, color-coded by group
  • Correlation matrix or regression line per group
  • Group means of the covariate

📊 Python Example

sns.lmplot(x='covariate', y='outcome', hue='group', data=df)
df.groupby('group')['covariate'].mean()

🟩 3. EDA for MANOVA

📌 Goal

Understand how multiple dependent variables vary across groups.

🔍 Key EDA Steps

  • Check correlation among DVs (they should not be independent)
  • Explore DV patterns by group using pairwise plots
  • Standardize DVs if they differ in scale

✅ EDA Tools

  • Pairplot of DVs with group hue
  • Correlation matrix of DVs
  • Z-scoring if necessary

📊 Python Example

sns.pairplot(df, vars=['dv1', 'dv2', 'dv3'], hue='group')
df[['dv1', 'dv2', 'dv3']].corr()

🟥 4. EDA for MANCOVA

📌 Goal

Explore multiple DVs across groups, adjusting for one or more covariates.

🔍 Key EDA Steps

  • Follow MANOVA EDA (correlation, scaling)
  • Assess how covariates relate to each DV
  • Examine whether covariate distributions are similar across groups

✅ EDA Tools

  • Scatterplots of each DV vs each covariate by group
  • Covariate summary stats by group
  • PCA to visually reduce complexity (optional)

📊 Python Sketch

for dv in ['dv1', 'dv2']:
    sns.lmplot(x='covariate', y=dv, hue='group', data=df)

📘 Summary Table

Method EDA Focus Tools Used
ANOVA DV by group (1 DV) Boxplots, descriptive stats, Levene's
ANCOVA DV + covariate by group lmplot, covariate means by group
MANOVA 2+ DVs by group Pairplot, correlation matrix, scaling
MANCOVA 2+ DVs + covariates by group Scatter matrix, group summaries, PCA

Let me know if you want to add dataset-ready templates or checklists for running these steps efficiently!

  • [[Links]]