What Is a Factor Loading?
A factor loading is a coefficient that represents the strength and direction of the relationship between an observed variable (like a survey question) and an underlying latent factor. Factor loadings range from -1.0 to +1.0, similar to correlation coefficients. A loading of 0.80 means the observed variable is strongly related to the factor, while a loading of 0.20 means the relationship is weak. When squared, a factor loading tells you the proportion of variance in the observed variable that's explained by the factor, so a loading of 0.80 means the factor accounts for 64% of that variable's variance. Factor loadings are the primary output you interpret when running factor analysis.
Why Factor Loadings Matter
Factor loadings are how you determine what each factor represents and whether your survey items are measuring what you think they're measuring. Without examining loadings, factor analysis is just math without meaning. They're also the basis for decisions about scale refinement, items with low loadings get dropped, items with cross-loadings get rewritten, and items with strong, clean loadings confirm your construct is solid.
How Factor Loadings Work
Interpreting Loading Values
Factor loadings are read like correlation coefficients between the item and the factor:
| Loading Range | Interpretation | Action |
|---|---|---|
| 0.70 - 1.00 | Strong | Excellent indicator of the factor, keep |
| 0.50 - 0.69 | Moderate | Good indicator, keep |
| 0.40 - 0.49 | Weak but acceptable | Borderline, keep if theoretically justified |
| 0.30 - 0.39 | Minimal | Consider removing unless sample is very large |
| Below 0.30 | Negligible | Remove, item doesn't belong to this factor |
The squared loading gives you the communality contribution, how much variance the factor explains in that item. An item with a loading of 0.70 has 49% of its variance explained by the factor (0.70² = 0.49). An item at 0.40 has only 16% explained, most of its variance is noise or something else entirely.
Cross-Loadings
A cross-loading occurs when a single item loads meaningfully on two or more factors. This is a problem because it means the item isn't cleanly measuring one construct, it's straddling two.
Example of a cross-loading problem:
The item "This brand offers good quality at a fair price" might load at 0.55 on a "Quality" factor and 0.48 on a "Value" factor. It's pulling toward both because it literally references both concepts. The fix is to split it into two separate items: one about quality, one about price fairness.
Rules of thumb for cross-loadings:
- An item should load above 0.40 on its primary factor and below 0.30 on all other factors
- The difference between the primary loading and the highest cross-loading should be at least 0.20
- Items that violate these guidelines should be revised or removed
Worked Example
You developed a 10-item scale to measure two dimensions of service quality: "Reliability" and "Responsiveness." After surveying 300 customers and running EFA with oblimin rotation, here's the loading matrix:
| Item | Reliability | Responsiveness |
|---|---|---|
| Delivers on promises | 0.82 | 0.08 |
| Consistent experience | 0.77 | 0.12 |
| Error-free service | 0.71 | 0.15 |
| Dependable outcomes | 0.68 | 0.22 |
| Quick response time | 0.11 | 0.84 |
| Willing to help | 0.09 | 0.79 |
| Prompt attention | 0.18 | 0.75 |
| Handles issues fast | 0.14 | 0.72 |
| Timely and reliable | 0.45 | 0.41 |
| Good overall service | 0.38 | 0.35 |
Items 1-8 have clean loadings, they load strongly on one factor and weakly on the other. Item 9 ("Timely and reliable") cross-loads because it blends both constructs. Item 10 loads poorly on both factors, it's too general to measure either dimension well. You'd drop items 9 and 10 and have a clean 8-item scale.
Pattern vs. Structure Matrix
When using oblique rotation (which allows factors to correlate), your software produces two matrices:
- Pattern matrix: Shows the unique contribution of each factor to each item, controlling for other factors. This is what you interpret for factor loadings.
- Structure matrix: Shows the total correlation between each item and each factor, including shared variance. These values are always higher.
Always interpret the pattern matrix for item-factor assignments. The structure matrix is useful for understanding total relationships but inflates the apparent strength of loadings.
Sample Size and Loading Stability
Loadings from small samples are unstable. The threshold for "significant" loadings depends partly on sample size:
- n = 100: Loadings should exceed 0.55 to be considered stable
- n = 200: Loadings above 0.40 are generally stable
- n = 300+: Loadings above 0.30 can be meaningful
- n = 500+: Even loadings of 0.25 may be statistically significant (though practically they're still weak)
These aren't hard rules, but they highlight why larger samples produce more trustworthy factor solutions.
When to Use Factor Loadings
- Scale development and validation to confirm that items measure their intended constructs
- Questionnaire refinement to identify items that should be dropped, revised, or reassigned
- Construct validity assessment to demonstrate that your measures capture distinct dimensions
- Computing factor scores: items can be weighted by their loadings to create composite scores
Common Mistakes to Avoid
- Using a universal cutoff of 0.30 regardless of sample size: with small samples, higher cutoffs (0.50+) are needed for stable interpretation
- Ignoring cross-loadings and keeping problematic items: cross-loading items muddy your factors and reduce discriminant validity
- Interpreting the structure matrix instead of the pattern matrix when using oblique rotation, the structure matrix inflates loadings because it includes shared factor variance
How Quali-Fi Supports Factor Loading Analysis
Quali-Fi's Intelligence tier generates rotated factor loading tables with cross-loadings highlighted automatically. The platform flags items that fall below your specified threshold and identifies cross-loading problems, giving you a clear action list for scale refinement without manual matrix inspection.
Analyze your scale with Quali-Fi
Frequently Asked Questions
Can factor loadings be negative?
Yes. A negative loading means the item is inversely related to the factor. For example, the item "This brand is overpriced" might load at -0.65 on a "Value" factor. Reverse-coded items often produce negative loadings. This is normal and interpretable, the item is still a strong indicator of the factor, just in the opposite direction.
What's the minimum number of items per factor?
A factor defined by only one or two items is considered weak and unreliable. The standard minimum is three items per factor, with four or more preferred. Single-item factors are uninterpretable in EFA and should prompt you to either add more items measuring that construct or merge it with a related factor.
Should I remove an item with a loading of 0.39?
It depends on context. If your sample is large (300+) and the item is theoretically important to the construct, keeping it at 0.39 is defensible. If the sample is small, or other items measure the same construct more cleanly, removing it will sharpen the factor. The decision should balance statistical criteria with theoretical considerations.