MODULE 10
Principal Component Analysis
From five survey questions to one principal component: learn what SPSS preserves, what it loses, and what the new score means.
Examples on this page use constructed or simulated teaching data, not empirical findings. About the data
- Research question
- Variables (trained here)
- Variable roles
- Measurement levels (trained here)
- Method choice (trained here)
- Statistical logic (trained here)
- SPSS workflow (trained here)
- Output interpretation (trained here)
- Defensible conclusion (trained here)
Understand
Why create a composite variable?
Index
Align items and average them. Often equal weights; examine internal consistency with Cronbach's alpha.
Sum of Z-scores
Standardize before summing. For binary possessions, owning a rare item earns a large positive Z; lacking a common item gives a large negative Z.
Principal components
Use linear combinations of standardized items to retain as much total variance as possible in PC1; later components are uncorrelated with PC1.
PCA compresses observed variance; it is distinct from factor analysis, which models shared variation through latent factors and unique/error components (handout p. 17). Neither automatically establishes validity or causality.
Try
Which items belong in PC1?
Communalities (extraction)
V43 is below the course's .25 communality rule. Remove it and rerun PCA.
27.8% of this item's variance captured
28.4% of this item's variance captured
45.6% of this item's variance captured
23.2% of this item's variance captured · below course threshold
25.9% of this item's variance captured
Dashed line: .25 course rule of thumb, not a universal cutoff. Source: course PDF pp. 10–11.
SPSS
Read the SPSS output before saving scores
SPSS menu and decisions
- Check missing values and measurement levels. Choose Analyze → Dimension Reduction → Factor.
- Select V40–V44. Under Extraction choose Principal Components and a fixed number of 1 component.
- Read Communalities. The handout removes V43 first, reruns, then removes V40.
- For V41, V42, V44, read Total Variance Explained (47.694%) and Component Matrix (.645, .790, .626).
- Once satisfied, open Scores → Save as variables; inspect and label the new standardized component score.
SPSS syntax: reproduce the workflow
FACTOR /VARIABLES V41 V42 V44 /MISSING LISTWISE /ANALYSIS V41 V42 V44 /PRINT INITIAL EXTRACTION /CRITERIA FACTORS(1) /EXTRACTION PC /ROTATION NOROTATE /SAVE REG(ALL).
SPSS typically saves FAC1_1. Rename and label it after checking which direction higher scores represent; the sign of a principal component is arbitrary. The original handout labels the resulting score econpolicy.
Report your conclusion
The three-item PC1 retains 47.694% of their standardized total variance (eigenvalue 1.431); communalities are .416, .623 and .392, respectively. Report the observed loadings, the meaning of high scores, missing-case handling and the retained variance. The handout reports N = 926 for saved scores.
Test yourself
Exit ticket: interpret, don't just click
1. The five-item SPSS output gives V43 communality .232. What does the course workflow suggest?
- Remove V43 and rerun PCA
- Keep it because it is positive
- Multiply every response by .232
Answer: A. The handout's classroom cutoff is .25, so .232 is below it. The next run can change the other communalities.
2. The final solution has eigenvalues 1.431, .902, .667. How much standardized variance does PC1 retain?
- 1.431%
- 47.694%
- 100%
Answer: B. Three standardized variables have total variance 3. Divide 1.431 by 3, then multiply by 100.
3. V42 loading = .790. What is the corresponding communality?
- About .624
- Exactly .790
- About .210
Answer: A. .790 squared is approximately .624 (the published communality is .623, due to rounding).
4. Does a positive component loading prove that the item causes the component?
- Yes
- No
Answer: B. A loading describes correlation with the component, not causation.