SPSS · Scale reliability

Cronbach’s alpha in SPSS: how to run it, interpret it and fix a low value

Your alpha is 0.58 and your supervisor says the scale fails. Which item do you drop, and how many can you drop? A step-by-step guide, including how to report it in your thesis.

⏱ ~6 min readSources: Nunnally & Bernstein (1994) · Hair et al. (2019)Updated:

You run the reliability analysis and alpha comes back at 0.58. Your supervisor says the scale fails. You open the Item-Total Statistics table and have no idea which column to look at, which item to drop or how many you are allowed to drop. This guide covers Cronbach’s alpha in SPSS from the clicks to reading each column, the rules for removing items, six reasons alpha comes out low, and how to report it in your thesis.

Cronbach’s alpha thresholds from below 0.6 (unacceptable) to above 0.95 (possibly redundant)
The commonly cited Cronbach’s alpha thresholds.

1What is Cronbach’s alpha?

Cronbach’s alpha (Cronbach, 1951) measures the internal consistency of a scale: whether the items meant to measure one concept move together. It usually runs from 0 to 1 (it can go negative when something is wrong with the data), and the closer to 1, the more consistent the items are.

✦ Three things to remember
  • Run it separately for each construct, never on the whole questionnaire.
  • Alpha rises with the number of items, so long scales get a head start.
  • A high alpha does not prove a scale measures the right thing. That is the job of EFA and CFA.

2What is a good Cronbach’s alpha?

  • ≥ 0.8: Good
  • 0.7 to < 0.8: Acceptable, the usual benchmark (Nunnally & Bernstein, 1994)
  • 0.6 to < 0.7: Acceptable for exploratory research or new concepts (Hair et al., 2019)
  • < 0.6: Unacceptable
  • > 0.95: A warning sign that items may repeat each other

There is one more condition: each item’s corrected item-total correlation should be at least 0.3.

3How to run Cronbach’s alpha in SPSS

⚠️ Before you run it

Check the data is numeric, 1–5 in the right direction, and that reversed items have been reverse-coded. This is the top cause of a low or negative alpha.

🖱️ Clicks
  1. Analyze→Scale→Reliability Analysis
  2. Move the items of one construct (say, SAT1–SAT5) to the Items box. Model: Alpha.
  3. Click Statistics and tick Item, Scale and Scale if item deleted, then Continue and OK.
  4. Repeat for each remaining construct.

4Reading the Cronbach’s alpha output: which column matters?

The Reliability Statistics table gives you the overall alpha. The Item-Total Statistics table shows how each item contributes:

ColumnWhat it tells you
Scale Mean if Item DeletedThe scale’s mean without this item; rarely used to decide
Scale Variance if Item DeletedThe scale’s variance without this item; rarely used
Corrected Item-Total CorrelationHow this item correlates with the rest: should be ≥ 0.3
Cronbach’s Alpha if Item DeletedThe new alpha without this item: if it is clearly higher, this item is dragging the scale down
Example: a Satisfaction scale (SAT1–SAT5)

Cronbach’s alpha = 0.742.

ItemCorrected Item-TotalAlpha if Item Deleted
SAT10.6120.668
SAT20.5870.677
SAT30.6340.661
SAT40.5560.689
SAT50.1820.821

SAT5 has an item-total correlation of 0.182 (below 0.3), and removing it lifts alpha to 0.821. Consider dropping SAT5 and re-running with the other four.

Item-Total Statistics table where SAT5 has a correlation of 0.182 and alpha rises to 0.821 without it
Example satisfaction scale: SAT5 is dragging alpha down.

5How to remove items properly

  1. Remove one item at a time, starting with the lowest item-total correlation, then re-run, because the remaining results will change.
  2. Keep at least 3 items per construct.
  3. Read the question before you drop it: if it covers a core part of the concept, explain why you keep it or look for another cause.
  4. Do not remove items just to gain a few points of alpha on a scale that already passes.
  5. Record which items you removed and why. Your committee will ask.

6Low Cronbach’s alpha: 6 causes and fixes

CauseFix
Reversed items not reverse-codedRecode 1↔5, 2↔4 and re-run; alpha often jumps
Wrong coding (such as Automatic Recode on Likert)Check Frequencies for each item and recode in the right direction
One item off-topic or confusingCheck the corrected item-total and consider removing it as in section 5
The scale actually measures several aspectsRun EFA to see whether it splits into 2 factors
Too few items (only 2)Report the correlation or Spearman–Brown; design at least 3 next time
Translation shifted the meaningCompare with the original scale; note it as a limitation
⚠️ Data integrity

Never edit, delete or add responses to make alpha look better. If a scale still fails after proper handling, report that honestly and discuss it under limitations. That is a legitimate result too.

7How to report Cronbach’s alpha in your thesis

Present one summary table for all constructs:

ConstructItemsCronbach’s alphaLowest item-total correlation
Service quality (SQ)50.8640.592
Perceived value (PV)40.8120.548
Satisfaction (SAT)4 (SAT5 removed)0.8210.601
✍️ A paragraph you can adapt

“The results show that all scales had a Cronbach’s alpha above 0.7 and that all item-total correlations exceeded 0.3 (Nunnally & Bernstein, 1994). Item SAT5 had an item-total correlation of 0.182 and was removed; alpha for the Satisfaction scale then rose from 0.742 to 0.821. The remaining items were carried forward to exploratory factor analysis.”

Practice reading the tables before your data is in

Have not finished collecting, but want to see what the alpha, rotated matrix and regression tables will look like for your model? On The Light Data’s model canvas, draw your model, pick SPSS and read the full set of result tables on simulated data. Simulated data is for learning only and never replaces real survey data.

🎯 In short
  • Alpha ≥ 0.7 (≥ 0.6 for exploratory work) and item-total correlation ≥ 0.3.
  • Run each construct separately; reverse-code reversed items first.
  • Remove one item at a time, keep at least 3, weigh the content and record your reasons.
  • Items that pass move on to EFA.
→
Next step
Exploratory factor analysis in SPSS
KMO, Bartlett’s test, variance explained and the rotated matrix.
→

Frequently asked questions

What is a good Cronbach’s alpha?

The usual benchmark is 0.7 or higher (Nunnally & Bernstein, 1994); 0.6 to 0.7 can be acceptable in exploratory research (Hair et al., 2019). Above 0.95 may mean the items repeat each other.

What item-total correlation means an item should go?

Items with a corrected item-total correlation below 0.3 are usually candidates for removal. Remove one at a time, re-run the analysis and consider what the item measures before removing it.

What if my scale has only 2 items?

Alpha is unstable with 2 items; report the inter-item correlation or the Spearman–Brown coefficient instead. Ideally each scale has at least 3 items.

Should I run Cronbach’s alpha on the whole questionnaire?

No. Run it separately for each construct, because alpha measures consistency among items that measure the same thing.

Why is my Cronbach’s alpha negative?

Usually a reversed item has not been recoded, or the data was coded wrongly. Reverse-code the item (1↔5, 2↔4) and run it again.