Description
Determines how many factors or components to retain by comparing observed eigenvalues with those from random data of the same size, keeping only those that beat chance.
The number-of-factors decision is the most consequential step in factor analysis, and the popular Kaiser rule (eigenvalues above 1) consistently retains too many. Parallel analysis generates many random datasets with the same number of respondents and items, computes their eigenvalues, and retains factors whose observed eigenvalues exceed the 95th percentile of the random ones. Simulation studies show it is among the most accurate retention methods. It can be run on principal components or on the reduced correlation matrix for common factors, and with polychoric correlations for ordinal items. It is a few lines of code in R (psych::fa.parallel) or Python.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Latent
- Stage
- Evaluation
- Data
- Quantitative
- Effort
- Hours
Steps
- Prepare the item response matrix and decide whether you are extracting components or common factors.
- Generate at least 500-1,000 random or permuted datasets with identical dimensions.
- Compute eigenvalues for each random dataset and take the 95th percentile for each position.
- Retain factors whose observed eigenvalues exceed the corresponding 95th-percentile random eigenvalue.
- Cross-check with the scree plot, the MAP test and interpretability, and report all in the analysis.
Sources
- Discovering Statistics Using R/SPSS (Andy Field)
- Quantitative User Experience Research (Chris Chapman & Kerry Rodden)
- Movie Booking Playbook (Vivek Gopinath)
- Horn (1965) A rationale and test for the number of factors in factor analysis
- Discovering Statistics (Field)
- Quantitative UX Research (Chapman)
- Movie Booking Playbook (Gopinath)
- The Inquiry Series, Ch 50 — Factor Analysis — Discovering the Latent Structure of Experience
- Factor and Discriminant Analysis (Sauro & Lewis, MeasuringU training)
Also known as
- Horn's parallel analysis
Categories
- Analytics & Behaviour
- Survey & Psychometrics