Course contentsShow
← Data Science
Lesson 217 of 2,1456. The Normal Distribution in DepthFree lesson

Evaluating Transformation Effectiveness

Assess whether transformations successfully improved normality using statistical tests and visual diagnostics.

Evaluating Transformation Effectiveness

You'll learn: How to check whether your data transformation actually made your data more normal.

Why Evaluation Matters

Transforming data (using log, square root, Box-Cox, etc.) is only useful if it actually improves normality. Think of it like adjusting a recipe—you need to taste the dish afterward to know if your changes worked! Simply applying a transformation without checking the results is like seasoning blindly.

How to Evaluate Effectiveness

Compare Before and After

Use the same tools you learned for testing normality, but apply them both before and after transformation:

Visual checks:

  • Create Q-Q plots for both original and transformed data
  • Compare histograms side-by-side
  • The transformed version should show straighter Q-Q plot lines and more bell-shaped histograms

Statistical tests:

  • Run Shapiro-Wilk, Kolmogorov-Smirnov, Anderson-Darling, or Jarque-Bera tests on both versions
  • Higher p-values after transformation indicate improvement
  • Look for p-values crossing the typical 0.05 threshold from below to above

What Success Looks Like

A successful transformation moves your data closer to the normal distribution pattern. Your Q-Q plot points hug the diagonal line more tightly, your histogram looks more symmetric and bell-shaped, and your statistical test p-values increase (suggesting the data is less different from normal).

If transformation doesn't help—or makes things worse—try a different transformation method or accept that your data may not need to be perfectly normal for your analysis goals.

Key Takeaway: Always validate transformation effectiveness by comparing normality diagnostics (visual and statistical) before and after—improvement means better alignment with the normal distribution.