Reliability analysis (alpha, split-half, Guttman, ICC)
Checks the internal consistency of a scale (Cronbach's alpha, split-half, Guttman's lambdas), item-total statistics, and rater agreement (intraclass correlation).
Menu: Analyze → Scale → Reliability Analysis…
When should I use it?
- Building a questionnaire scale.
- Checking agreement between raters (ICC).
What are the assumptions?
- Items measure one construct and are scored in the same direction (reverse-score first).
How do I run it in StatLab?
- Move the items, choose the model, tick 'Scale if item deleted'.
- For ICC choose the model (one-way, two-way random, two-way mixed) and type (consistency or absolute agreement).
- Change 'ICC test value' to test whether the ICC exceeds a benchmark such as .70.
How do I read the output?
| Output | Meaning |
|---|---|
| Cronbach's Alpha | Around .70 acceptable, .80 good. |
| Corrected Item-Total Correlation | Below .30 = item fits poorly; negative = probably needs reversing. |
| Alpha if Item Deleted | Alpha without that item. |
| ICC Single / Average Measures | Reliability of one rater / of the mean of all raters. |
| F Test with True Value | Tests H0: ICC = test value; single and average measures get their own F. |
Worked example
On the practice survey dataset (File → Sample dataset), this command — which the dialog's Paste button also produces — runs the procedure:
PROC relReliability /items=(q1 q2 q3) /model=alpha /item=YES /deleted=YES /inter=NO /summary=NO /icc=NO /iccModel=twoway-mixed /iccType=consistency /testValue=0.
Output tables: Case Processing Summary · Reliability Statistics · Item Statistics · Item-Total Statistics · Scale Statistics.
How would I report it in APA style?
Cronbach's α for the 3-item scale was .78 (N = 150).
Computed by StatLab's validated procedure code when this page was built. The example only illustrates the steps; choose variables that fit your own research question.
What are common mistakes?
- Forgetting to reverse negatively worded items.
- Reporting the average-measures ICC when single ratings will be used.