Teaching Principal Components Using Correlations | JoVE Visualize

Area of Science:

Social and Behavioral Sciences
Data Analysis
Statistical Education

Background:

Introducing principal components (PCs) is challenging for students, particularly in social and behavioral sciences, due to complex matrix algebra and maximization proofs.
Standard PCA motivations, like variance maximization with unit length constraints, lack direct connection to 'variance explained' and restrict practical applications such as re-scaling or oblique rotations.
Existing teaching methods for PCA often fail to bridge the gap between theoretical complexity and practical interpretation for students.

Purpose of the Study:

To propose a more intuitive and accessible method for teaching principal components (PCs) to students in social and behavioral sciences.
To reframe PCA motivation around optimizing average proportions of variance explained, directly linking to the familiar R-squared statistic.
To eliminate the need for complex matrix algebra and optimization proofs in introductory PCA education.

Main Methods:

Proposed motivation focuses on optimizing (weighted) average proportions of variance explained in original variables.
This approach bypasses the need for unit length and uncorrelatedness constraints, allowing for direct interpretation of 'variance explained'.
The method is presented without requiring matrix algebra or optimization proofs, making it suitable for broader student accessibility.

Main Results:

The proposed method offers a more intuitive understanding of PCA by directly relating it to the R-squared statistic.
It provides a clear interpretation of 'variance explained' and resolves the covariance-based vs. correlation-based PCA decision.
The approach facilitates the use of re-scaling and oblique rotations, common in practical data analysis.

Conclusions:

The novel approach simplifies the teaching and learning of principal components analysis (PCA) in social and behavioral sciences.
By focusing on variance explained and R-squared, the method enhances student comprehension and practical application of PCA.
Integration of modern data science tools like heat maps and text mining further aids in the interpretation and application of PCs.