Your reliability output shows α = .58 for one subscale. Do you delete items, redistribute the questionnaire, or report the number as it is? Cronbach’s alpha is the first statistic an examiner checks in the methodology or results chapter, because every later analysis rests on the assumption that the scale scores are consistent. It is also one of the most misused statistics in master’s theses: computed before negatively worded items were reversed, judged against a threshold nobody can source, or pushed upward by deleting items until the number looks right.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.
This guide to Cronbach’s alpha in SPSS explains what alpha measures and what it does not, the SPSS steps, how to read the Item-Total Statistics table, the thresholds and where they come from, six fixes for a low value, the alternatives (McDonald’s omega and split-half), and how to write the result in APA style.
What is Cronbach’s alpha? Cronbach’s alpha (Cronbach, 1951) is a coefficient of internal consistency: it estimates how closely a set of items intended to measure the same construct covary with one another. It ranges in theory from 0 to 1 and depends on two things only, the number of items and the average correlation among them. Alpha is not a measure of validity, and a high value does not show that the items measure a single dimension.
What does alpha measure, and what does it not?
Alpha compares the variance shared by the items with the total variance of the scale score. The standardized form makes the logic visible: α = k × r̄ ÷ [1 + (k − 1) × r̄], where k is the number of items and r̄ the mean inter-item correlation. Eight items with a mean correlation of .40 give α = 3.2 ÷ 3.8 = .84; the same correlation with four items gives .73, and with twelve items .89. Scale length alone moves the coefficient, which is why a short subscale with a modest alpha can be better built than a long one with an impressive alpha.
Three things alpha does not do:
- It does not establish validity. A scale can be consistently measuring the wrong construct.
- It does not prove unidimensionality. Two correlated factors inside one subscale can still produce an acceptable alpha; dimensionality is a question for factor analysis.
- It does not measure stability over time. That is test-retest reliability, estimated from two administrations.
Alpha also assumes that all items load equally on the construct (tau-equivalence). When loadings differ, which is the usual case, alpha underestimates reliability, and Dunn, Baguley and Brunsden (2014) recommend reporting McDonald’s omega alongside it. Values above .95 deserve a second look: Tavakol and Dennick (2011) note that very high values often signal redundant items that ask the same question in different words.
How do you calculate Cronbach’s alpha in SPSS?
- Reverse negatively worded items first (see below). Computing Cronbach’s alpha in SPSS before this step is the most common cause of a low value.
- Analyze, then Scale, then Reliability Analysis.
- Move the items of one subscale only into the Items box. Do not mix subscales; alpha for the whole instrument is a separate run.
- Keep Model set to Alpha.
- Click Statistics and tick Item, Scale and Scale if item deleted under Descriptives for, and Correlations under Inter-Item.
- Click Paste instead of OK to keep the syntax, then run it. If a supervisor asks you to drop an item or exclude cases, you rerun one line rather than rebuilding the dialog.
- Repeat for each subscale, then once for the total scale if your instrument is designed to yield a total score.
📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.
How do you read the output?
Case Processing Summary shows how many cases were used. Reliability Analysis excludes any case with a missing value on any item, so 312 respondents can become 287 without warning; check this before interpreting.
Reliability Statistics gives Cronbach’s Alpha, Cronbach’s Alpha Based on Standardized Items and N of Items. Report the first value unless your items have different response formats.
Inter-Item Correlation Matrix should contain only positive correlations. A column of negative values is almost always an item that was not reverse-scored, or one that respondents understood in the opposite direction.
Item-Total Statistics is the table that tells you what to do:
| Column | Meaning | Decision rule |
|---|---|---|
| Scale Mean if Item Deleted | Mean of the scale without this item | Informational only |
| Corrected Item-Total Correlation | Correlation of the item with the sum of the other items | Values below .30 flag a weak item (Field, 2018); negative values flag a reversal problem |
| Squared Multiple Correlation | Variance of the item predicted by the others | Very low values support the weak-item diagnosis |
| Cronbach’s Alpha if Item Deleted | Alpha of the scale without this item | If it is clearly higher than the current alpha, the item lowers consistency |
What are the thresholds, and who set them?
No standard prescribes a mandatory cut-off, so cite the source of the threshold you use rather than quoting a bare number.
| Alpha | Usual label | Source |
|---|---|---|
| Below .60 | Unacceptable for a scale score used in inferential tests | Below every commonly cited bound |
| .60 to .69 | Questionable; sometimes tolerated for short or exploratory scales if justified | Nunnally and Bernstein (1994), with caution |
| .70 to .79 | Acceptable | Nunnally (1978): .70 for early-stage research |
| .80 to .89 | Good | Nunnally (1978): .80 for basic research |
| .90 to .95 | Excellent; required when decisions about individuals rest on the score | Nunnally (1978); Tavakol and Dennick (2011) |
| Above .95 | Check for redundant items | Tavakol and Dennick (2011) |
The .70 figure most committees expect originates in Nunnally’s (1978) recommendation for preliminary research. The same author recommended .80 for basic research and .90 or more for applied settings where important decisions depend on a score. A thesis that cites the rule it applies, and states which of these contexts applies, is on firmer ground than one that writes “.70 is the acceptable value” with no reference.
How do you handle reverse-scored items?
Negatively worded items (“I often think about leaving this job” in a satisfaction scale) are reversed so that a high score always means more of the construct. For a 5-point scale the new value is 6 minus the old one: 1 becomes 5, 2 becomes 4, 3 stays 3. In SPSS use Transform, then Recode into Different Variables, enter the five old-to-new pairs, and give the new variable a suffix such as _R; or use Transform, then Compute Variable, with the expression 6 − Q5. Never overwrite the original variable.
The symptoms of a forgotten reversal are unmistakable: a negative Corrected Item-Total Correlation, a markedly higher Alpha if Item Deleted for that item, and sometimes a negative alpha with the SPSS warning that the value is negative due to a negative average covariance among items. Reverse, rerun, and the problem usually disappears. Check scale scores afterwards: the mean of the reversed item should now sit on the same side of the midpoint as its neighbors.
Worked example: from the SPSS output to the APA sentence
A thesis measures job satisfaction with eight items on a 5-point scale among 312 teachers. The first run of the reliability analysis returns α = .61. In Item-Total Statistics, item 5 shows a Corrected Item-Total Correlation of −.28 and an Alpha if Item Deleted of .71, while the other seven items range from .38 to .64. Item 5 is the negatively worded item, not yet reversed.
After reversing item 5 and rerunning, α = .84 with all 312 cases retained. Corrected item-total correlations now range from .41 (item 2) to .67 (item 6). Alpha if Item Deleted ranges from .80 to .85; deleting item 2 would raise alpha by .01, which does not justify losing the only item covering pay. The mean inter-item correlation is .40, inside the .15 to .50 range that Clark and Watson (1995) describe as appropriate for a coherent but not redundant scale, and the standardized formula confirms the result: 8 × .40 ÷ (1 + 7 × .40) = .84.
The methodology sentence: “The eight-item job satisfaction scale showed good internal consistency, Cronbach’s α = .84 (N = 312), with corrected item-total correlations from .41 to .67. Item 5 was reverse-scored before analysis; no items were removed.” The accompanying table lists every subscale:
| Scale | Items | α (pilot, n = 42) | α (final, N = 312) | Mean inter-item r |
|---|---|---|---|---|
| Job satisfaction | 8 | .81 | .84 | .40 |
| Organizational commitment | 6 | .76 | .79 | .39 |
| Transformational leadership | 10 | .89 | .91 | .50 |
| Organizational trust | 5 | .70 | .72 | .34 |
Examiners read the weakest row first. Organizational trust at .72 with five items is acceptable under Nunnally’s (1978) rule, and saying so with the number of items prevents the question.
What do you do when alpha is below .70?
When Cronbach’s alpha in SPSS comes out below .70, work through the causes in this order; most low values are fixed by the first two.
- An item was not reversed. Look for negative inter-item correlations, reverse the item, rerun.
- A weak or ambiguous item. Corrected item-total correlation below .30 and a clearly higher alpha if deleted. Removal is defensible when the wording is also doubtful and the construct remains covered; report the deletion and keep both versions of the instrument in the appendix.
- Too few items. Two or three items rarely reach .70 even when they belong together. Report the mean inter-item correlation beside alpha, or add items at the pilot stage.
- Two dimensions in one subscale. Run an exploratory factor analysis on the subscale; if two factors appear, split it and compute alpha for each.
- A small or homogeneous pilot sample. Alpha from 15 respondents is unstable, and a very homogeneous group restricts variance. Pilot with at least 30 people from the target population.
- Data entry errors. Values such as 55 or items placed in the wrong subscale. Run Frequencies before anything else.
If alpha remains between .60 and .69 after these checks, you can report it with a justification (short scale, exploratory construct), support it with omega or split-half reliability, and name it as a limitation. Values below .60 call for revising the instrument before the main data collection.
Which alternatives should you know?
| Coefficient | What it assumes | When to prefer it |
|---|---|---|
| McDonald’s omega | Items load on one factor, loadings may differ | Whenever loadings are unequal; available in Reliability Analysis (Model: Omega) in SPSS 27 and later |
| Split-half (Spearman-Brown, Guttman) | The two halves are parallel | Speed or ability tests; when a quick second estimate is wanted (Model: Split-half) |
| Test-retest (Pearson r or ICC) | The construct is stable between administrations | Stability claims; requires two measurements |
| Composite reliability (CR) | A fitted measurement model | Theses using CFA or PLS-SEM; reported with AVE |
Omega is the most useful companion to alpha in a survey thesis: when the two agree, the tau-equivalence assumption did little harm; when omega is clearly higher, alpha was a lower bound and you can say so.
How do you report reliability in the thesis?
Report alpha for each subscale and, if the instrument is scored as a whole, for the total; give the number of items and the sample size for each coefficient; present pilot and final-sample values when both exist, because the final sample is the one your results depend on. Use the Greek letter in italics for the symbol only when your department’s style asks for it; APA 7 sets Greek letters in roman type, so “Cronbach’s α = .84” is the standard form. Never describe alpha as evidence of validity, and never delete items solely to improve the coefficient without reporting the change.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
Can I compute Cronbach’s alpha for a two-item scale?
SPSS will compute it, but alpha penalizes short scales, so a two-item measure rarely passes .70 even when the items are strongly related. Report the inter-item correlation and the Spearman-Brown coefficient instead, and say why the scale is short.
Should alpha be computed on the pilot sample or the final sample?
Both. The pilot value (ideally from 30 or more respondents) guides revisions before the main study; the final-sample value is the one that supports your results and belongs in the results or methodology chapter.
Is an alpha of .96 a problem?
Not automatically, but inspect the inter-item correlations. If several exceed .85, the items are near duplicates and the scale could be shortened without losing information. Report the value and comment on the redundancy.
Deleting one item raises alpha from .78 to .82. Should I remove it?
Only if the item is also weak (corrected item-total correlation below .30) and its wording is doubtful. An item that carries unique content is worth more than four hundredths of alpha, and reviewers notice scales trimmed to improve a statistic.
What is the difference between Cronbach’s alpha and composite reliability?
Composite reliability is computed from the standardized loadings of a confirmatory factor analysis or PLS-SEM measurement model and does not assume equal loadings. Both are read against the .70 convention; SEM-based theses usually report CR with AVE and alpha side by side.
Sources
- Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16(3), 297–334. https://doi.org/10.1007/BF02310555
- Nunnally, J. C. (1978). Psychometric Theory (2nd ed.). McGraw-Hill; Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric Theory (3rd ed.). McGraw-Hill.
- Tavakol, M., & Dennick, R. (2011). Making sense of Cronbach’s alpha. International Journal of Medical Education, 2, 53–55. https://doi.org/10.5116/ijme.4dfb.8dfd
- Dunn, T. J., Baguley, T., & Brunsden, V. (2014). From alpha to omega: A practical solution to the pervasive problem of internal consistency estimation. British Journal of Psychology, 105(3), 399–412. https://doi.org/10.1111/bjop.12046
- Clark, L. A., & Watson, D. (1995). Constructing validity: Basic issues in objective scale development. Psychological Assessment, 7(3), 309–319.
- Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics (5th ed.). SAGE.
- IBM. IBM SPSS Statistics documentation: Reliability Analysis. https://www.ibm.com/docs/en/spss-statistics
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.