SPSS
Cronbach’s Alpha
Reliability Analysis
Quantitative Research
Thesis Data Analysis
Kenya Research
Tobit Research Consulting | Statistical Analysis & Thesis Support Series | Reading time: ~14 minutes
What you will learn: Why reliability testing matters before you ever run your main analysis; the exact menu path to run Cronbach’s Alpha in SPSS (Analyze → Scale → Reliability Analysis); how to read every number in the output — Cronbach’s Alpha, Corrected Item-Total Correlation, and “Alpha if Item Deleted”; the acceptable threshold researchers actually cite (0.70, per Nunnally & Bernstein, 1994); when — and when not — to delete a poorly performing item; how to save scale scores as a new variable for your next analysis; and how to report reliability results correctly in your Chapter Three or Chapter Four.
1. Why Reliability Testing Comes Before Everything Else
Before you run a single regression, correlation, or t-test, your questionnaire needs to pass one basic test: are the items measuring what you think they’re measuring, consistently? This is what reliability analysis checks. If a set of items intended to measure, say, “customer satisfaction” or “job stress” doesn’t hang together statistically, then any conclusion you draw using that scale — no matter how sophisticated the downstream analysis — is built on a shaky foundation.
This is why reliability testing sits at the start of almost every Chapter Four in quantitative research, and why supervisors ask for it before looking at anything else. It answers one specific question: if you administered this same set of items again, under the same conditions, would you get a consistent pattern of responses? That consistency — not correctness, not validity, just internal consistency — is what Cronbach’s Alpha measures.
Reliability is not the same as validity. A scale can be highly reliable (consistent) while still not measuring the concept it claims to measure (not valid). Reliability analysis, including Cronbach’s Alpha, only tells you whether your items are internally consistent — it says nothing about whether they measure the right construct. Validity is a separate, additional step.
2. What Cronbach’s Alpha Actually Measures
Cronbach’s Alpha is a single number, typically between 0 and 1, that summarises how closely related a set of items are as a group. It works on the logic that if several items are genuinely measuring the same underlying construct, respondents who score high on one item should tend to score high on the others too. Alpha essentially averages all the possible ways of splitting your items into two halves and correlating those halves — giving you one overall figure for internal consistency.
A few practical things worth knowing before you run it:
Factor
Alpha Is Sensitive to the Number of Items
All else equal, a scale with more items tends to produce a higher Alpha, simply because there are more items to average across. This means a 4-item scale and a 15-item scale shouldn’t be judged by exactly the same intuitive standard — a shorter scale with a slightly lower Alpha is not necessarily worse.
Factor
Alpha Assumes Uni-Dimensionality
Cronbach’s Alpha assumes all items in the set are measuring one single underlying construct. If your “scale” is actually mixing two different constructs together (for example, mixing “job satisfaction” items with “job stress” items in one reliability run), Alpha can be misleading. Run reliability analysis separately for each conceptually distinct sub-scale.
3. How to Run Reliability Analysis in SPSS: Step-by-Step
- Prepare your data. Make sure all items belonging to the scale are coded in the same direction. Reverse-code any negatively worded items first — running reliability analysis with un-reversed items is one of the most common causes of a surprisingly low, even negative, Alpha.
- Open the Reliability Analysis dialog. Go to Analyze → Scale → Reliability Analysis.
- Move your items into the Items box. Select all the variables that belong to the scale you’re testing (for example, all items measuring “customer satisfaction”) and move them into the “Items” field on the right.
- Confirm the model is set to Alpha. In the “Model” dropdown, ensure “Alpha” is selected — this is the default and the model almost every thesis and research report should use.
- Open the Statistics sub-dialog. Click the “Statistics” button. Under “Descriptives for,” tick Item, Scale, and Scale if item deleted. Under “Inter-Item,” you can optionally tick Correlations. These selections are what produce the Corrected Item-Total Correlation and Alpha If Item Deleted columns you’ll need in the next step.
- Click Continue, then OK. SPSS will generate the reliability output in your Viewer window.
Before you run it: check that all your scale items are on the same measurement scale (e.g., all on a 5-point Likert scale) and that missing values have been handled — SPSS will exclude a case listwise from the entire reliability calculation if it has a single missing value on any item in the scale, which can silently shrink your sample size without you noticing.
4. Reading the Output: Alpha, Item-Total Correlation, and Alpha If Item Deleted
SPSS reliability output produces several tables. Two matter most for a thesis or research report: the overall Reliability Statistics table, and the Item-Total Statistics table.
| Output Element |
What It Tells You |
Where to Look |
| Cronbach’s Alpha |
The overall internal consistency of the full set of items — the single number most commonly reported in Chapter Four/Three |
Reliability Statistics table, top of output |
| Corrected Item-Total Correlation |
How strongly each individual item correlates with the sum of all the other items in the scale — a low value flags a weak or poorly fitting item |
Item-Total Statistics table, one row per item |
| Cronbach’s Alpha if Item Deleted |
What the overall Alpha would become if that specific item were removed — used to decide whether dropping an item would improve the scale |
Item-Total Statistics table, rightmost column |
Reading the two tables together: the overall Alpha tells you whether the scale as a whole is reliable. The Item-Total Statistics table tells you why — and specifically, which item, if any, is dragging the overall figure down. A thesis that reports only the overall Alpha, without checking the item-level detail, misses the diagnostic value of the analysis.
5. What Counts as “Acceptable”? The 0.70 Threshold, Explained Properly
≥ 0.70
the internal consistency threshold most widely cited in applied social science research
Nunnally & Bernstein, 1994
The 0.70 benchmark, commonly attributed to Nunnally and Bernstein’s 1994 work on psychometric theory, has become the default cut-off cited across business, education, and social science theses. It is a widely used convention, not an absolute law — and it’s worth understanding the broader scale so you can interpret a result that falls just under or comfortably over that line correctly.
| Alpha Value |
Common Interpretation |
| α ≥ 0.90 | Excellent (though very high values can sometimes indicate redundant items) |
| 0.80 ≤ α < 0.90 | Good |
| 0.70 ≤ α < 0.80 | Acceptable — the conventional minimum for most thesis-level research |
| 0.60 ≤ α < 0.70 | Questionable — often tolerated in early-stage or exploratory scales, but should be flagged and discussed |
| α < 0.60 | Poor — generally requires revising the scale or removing weak items before proceeding |
A value below 0.70 is not automatically fatal to your study — but it does require you to say something about it in your report rather than ignore it. Discuss which items were weak, whether you removed any, and what the implication is for interpreting later results based on that scale.
6. Should You Delete an Item? A Decision Framework
The “Alpha if Item Deleted” column is often misread as a simple instruction: “delete whichever item shows the highest value here.” That is not quite right, and treating it that way can weaken your instrument rather than strengthen it.
✅ Reasonable Grounds to Consider Deletion
- The item’s Corrected Item-Total Correlation is very low (commonly below 0.30) — it is not correlating meaningfully with the rest of the scale
- Removing it would raise the overall Alpha by a meaningful margin, not a trivial 0.01–0.02
- You can justify conceptually why that item may not belong with the others (e.g., it was ambiguously worded or measures a slightly different sub-concept)
- Deleting it still leaves you with enough items to adequately represent the construct
❌ Poor Practice
- Deleting items purely to inflate Alpha as high as possible, without conceptual justification
- Removing items one at a time in successive reliability runs until Alpha looks “impressive,” without reporting what was removed and why
- Deleting an item without checking whether it was simply mis-coded (e.g., a reverse-worded item that wasn’t reverse-scored)
- Reducing a scale down to just 2–3 items purely to chase a higher Alpha, undermining the scale’s content coverage
The right first move when an item looks weak: check whether it needed reverse-coding before assuming the item itself is the problem. A large share of “problem items” in student data turn out to be coding errors, not genuinely weak items.
7. Saving Scale Scores as a New Variable
Once you’ve confirmed your scale is reliable, you’ll typically want a single composite score for that construct — for example, an overall “Customer Satisfaction” score — to use in later analyses like correlation or regression, rather than working with each item separately.
- Open Reliability Analysis again. Analyze → Scale → Reliability Analysis, with your finalised set of items in the Items box.
- Click Statistics, then look for the Anova Table / Scale options section depending on your SPSS version — in most modern versions, scale score saving is handled by computing a new variable directly (see next step), since reliability analysis itself does not always expose a direct “save” checkbox in every version.
- Compute the composite score manually via Transform → Compute Variable. This is the most version-independent, reliable method: create a new variable (e.g., CustSat_Total) defined as the mean of the relevant items, e.g. MEAN(Q1,Q2,Q3,Q4).
- Verify the new variable. Run basic descriptives (Analyze → Descriptive Statistics → Descriptives) on your new composite variable to confirm it has a sensible range and no unexpected missing values.
Why “mean of items” rather than “sum of items” is usually preferred: a mean score stays on the same scale as the original items (e.g., 1–5), which makes it far easier to interpret and report than a raw sum, and it handles occasional missing item responses more gracefully.
8. How to Report Reliability Results in Your Thesis
Examiners look for a specific, minimal set of information when reliability results are reported in a methodology or findings chapter. A single sentence with the Alpha value is not enough on its own.
📋 Example of Adequate Reporting
“Reliability of the customer satisfaction scale (5 items) was assessed using Cronbach’s Alpha, which yielded a coefficient of 0.84, indicating good internal consistency (Nunnally & Bernstein, 1994). All items recorded corrected item-total correlations above 0.30, and no item’s removal would have materially improved the overall Alpha; all five items were therefore retained for subsequent analysis.”
Minimum Reporting Checklist
What Your Reliability Section Should State
The name of the scale/construct and the number of items tested; the Cronbach’s Alpha value obtained; a citation for the threshold used to judge that value (Nunnally & Bernstein, 1994 is the most commonly cited); whether any items were removed, and the justification if so; and the final number of items retained for the composite score used in later analysis.
9. Common Mistakes That Get Flagged at Defense
Running reliability analysis on the full questionnaire at once, rather than separately per construct. If your instrument measures three different variables (e.g., leadership style, job satisfaction, and employee performance), each needs its own separate reliability run — a single Alpha across all items mixed together tells you almost nothing useful.
Forgetting to reverse-code negatively worded items before running the analysis, producing an artificially low or even negative Alpha that misrepresents an otherwise sound scale.
Reporting Alpha without the item-total detail, missing the chance to show the examiner that weak items were identified and handled deliberately, rather than the reliability check being a box-ticking afterthought.
10. How Tobit Research Consulting Can Help
Tobit Research Consulting works with Masters and PhD students across Kenyan universities — including the University of Nairobi, Kenyatta University, JKUAT, Mount Kenya University, CUEA, and Strathmore — on the statistical analysis chapters that consistently cause the most delay in thesis completion. Reliability analysis is usually the first hurdle, and getting it right the first time saves weeks of back-and-forth with a supervisor.
SPSS Reliability & Statistical Analysis Support — Nairobi, Kenya
Whether you need to learn to run this yourself or want your dataset analysed by our team, we offer:
- One-on-one SPSS training covering reliability, validity, and your full analysis chapter
- Professional reliability, validity, and inferential analysis on your dataset
- Clear, APA-formatted output tables ready to insert into Chapter Three or Four
- Interpretation guidance so you can defend your results with confidence
We work with students across all Kenyan universities and can typically turn around a standard dataset within 3–5 days.
Request SPSS Support →
📍 Bruce House, 4th Floor, Nairobi CBD, Kenya | Tel: +254 728 430 728 | tobitresearchconsulting.com
Reference: Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric Theory (3rd ed.). McGraw-Hill.
