0.70 is the figure most Nigerian departments accept as the minimum for a self-developed questionnaire, with 0.80 and above regarded as good. Below 0.70 the usual cause is not a bad topic — it is unreversed negative items, a section measuring two different things at once, or too few items. All three are fixable before you collect your main data.
What is Cronbach’s alpha actually measuring?
Internal consistency: the extent to which the items in one section of your instrument behave as though they are measuring the same underlying thing.
If Section B has ten items about study habits, and a respondent who agrees strongly with item 1 also tends to agree with items 2 through 10, the section is internally consistent and alpha will be high. If the answers scatter with no pattern, the items are not measuring one construct, and alpha will be low.
Note what it is not. Alpha says nothing about whether your instrument measures the right thing — that is validity, a separate matter your supervisor and validators handle. A questionnaire can be perfectly consistent and still measure the wrong construct. Panels occasionally ask this, and “reliability is consistency, validity is correctness” is the answer.
What counts as a good alpha?
The conventional bands, which most Nigerian departments teach:
| Alpha | Usual verdict |
|---|---|
| 0.90 and above | Excellent — but check for redundancy |
| 0.80 to 0.89 | Good |
| 0.70 to 0.79 | Acceptable — the common minimum |
| 0.60 to 0.69 | Questionable; some departments accept it for a new instrument |
| Below 0.60 | Poor — repair the instrument |
Two honest caveats. These bands are a widely taught convention rather than a statistical law, and different authors set the cut-offs slightly differently — so if your project handout names a threshold, that number outranks this table. And a very high alpha is not automatically better: above roughly 0.95 you may simply have asked the same question in several different ways, which wastes your respondents’ patience and adds nothing.
Report it with its context, not on its own: “A pilot test involving 30 respondents produced a Cronbach’s alpha of 0.84 for Section B (10 items) and 0.78 for Section C (8 items), indicating that the instrument was reliable.” The number of items and the pilot sample belong in that sentence, in Chapter Three.
Why is my alpha low?
Four causes, in the order you should check them.
1. You did not reverse-score the negative items
This is the most common cause by a wide margin, and it is the easiest to miss. If Section B contains “I struggle to concentrate when studying at night” among nine positively worded items, a respondent who agrees with the other nine will disagree with that one — and to the calculation, that looks like inconsistency.
Recode before you compute anything: Transform > Recode into Different Variables, mapping 1→4, 2→3, 3→2, 4→1 on a four-point scale, into a new variable rather than over the original. The full sequence is in the guide to analysing project data in SPSS step by step. Fixing this alone routinely moves an alpha from 0.4 to 0.8.
2. You put the whole questionnaire in at once
Alpha is computed per scale, not per instrument. If you feed Section B (study habits) and Section C (academic performance) into one Reliability Analysis, you are asking whether two different constructs behave as one, and they should not. Run each section separately and report an alpha for each.
3. Your section has too few items
Alpha rises with the number of items, all else being equal. A three-item section can be perfectly sound and still return a modest alpha simply because it is short. If a section is genuinely important to your study, four to six well-written items is a more comfortable place to be than two.
4. An item is genuinely bad
Double-barrelled items (“I find the library quiet and well stocked” — which one are you agreeing with?), ambiguous wording, or an item nobody understood in the same way. These show up clearly in the diagnostic below.
How do you find the item that is dragging it down?
SPSS will tell you directly if you ask it to.
- Go to Analyze > Scale > Reliability Analysis.
- Move in the items of one section, leaving the Model on Alpha.
- Click Statistics and tick Scale if item deleted.
- Run it, and read the column headed “Cronbach’s Alpha if Item Deleted”.
That column shows what alpha would become if each item were removed. Nearly every value will sit slightly below your current alpha — that is normal, and it means the item is contributing. The one to look at is any item whose deletion would raise alpha noticeably. Check the “Corrected Item-Total Correlation” column alongside it; an item near zero or negative there is not measuring what the rest of the section measures.
Can you just delete the bad item?
Usually yes, with three conditions.
- Do it at the pilot stage, not after main data collection. Repairing an instrument during a pilot is normal research practice. Dropping items from your final data until alpha crosses 0.70 is a different activity, and a panel that notices will treat it as one.
- Check the item is not essential to your research question. If it is the only item covering a concept your Chapter One promised to measure, replace it with a better-worded item rather than deleting it outright.
- Report what you did. One sentence in Chapter Three: “Following the pilot test, item 7 was removed from Section B as its deletion improved the reliability coefficient from 0.64 to 0.79, and the final instrument comprised nine items in that section.”
That sentence is a strength, not an admission. It shows you ran a pilot, read the output and acted on it, which is exactly what a pilot is for.
How many respondents does the pilot need?
Most Nigerian departments ask for between 20 and 30, drawn from a population similar to your sample but not part of it — a neighbouring department, another campus, a different section. If those respondents later appear in your main data, your pilot was not independent and your reliability figure is not what you claimed.
Check your project handout for the number your department expects, because this is one of the details supervisors specify and panels ask about.
What if you are using someone else’s instrument?
Then the original author has probably already published a reliability coefficient, and you cite it — but you still compute your own on your own pilot, because reliability is a property of a scale in a particular population, not a permanent property of the questionnaire. An instrument validated on postgraduates in another country may behave differently among 300-level undergraduates in your department.
Cite the source of the instrument in your reference list along with the author’s reported alpha, and report yours beside it. Where those citations sit is covered in the guide to writing Chapter Two.
What will the panel ask about this?
Three questions, reliably.
“What was your reliability coefficient?” Know the number for each section without looking it up.
“How did you compute it, and on what sample?” Name the software and the pilot size. Never report an alpha you did not compute yourself — you will be asked, and there is no recovery from not knowing where your own number came from.
“Is your instrument valid?” This is the different question. Validity comes from your supervisor and validators reviewing the instrument against your objectives, and it is reported separately from reliability. Both belong in Chapter Three, and the wider set of questions is inventoried in the guide to what is asked during a project defence. Where the resulting figures appear in your results chapter is covered in writing Chapter Four.
Keep the numbers and the chapter in step
Tesify holds your chapters in the structure your department expects, so a reliability figure you report in Chapter Three and the instrument you describe stay attached to each other while you work. Over 9,000 students have used it to write more than 15,000 chapters, and every word is still written by you.
Frequently asked questions
What is a good Cronbach’s alpha?
0.70 and above is the conventional minimum, 0.80 to 0.89 is good and 0.90 and above is excellent, though very high values can indicate redundant items. If your department’s handout states a threshold, follow that instead.
Is 0.6 an acceptable Cronbach’s alpha?
It is borderline. Some departments accept it for a newly developed instrument in an under-researched area, but many will ask you to improve it. Try reverse-scoring and the “alpha if item deleted” diagnostic before accepting it.
Why is my Cronbach’s alpha so low?
Most often because negatively worded items were not reverse-scored, or because two different sections were entered into one reliability analysis. Check both before concluding the instrument is bad.
Can Cronbach’s alpha be negative?
Yes, and it almost always means an item is scored in the opposite direction from the others. Reverse-score the negative items and run it again.
Do I compute one alpha for the whole questionnaire?
No. Compute one per section or construct. A single alpha across sections measuring different things is not meaningful, however good the number looks.
How many respondents should the pilot test have?
Usually 20 to 30, from a similar population but not from your main sample. Confirm the figure your department expects, since handouts often specify it.
Can I remove an item to improve my alpha?
At the pilot stage, yes, and you should report that you did and why. Removing items from your main data until the number improves is a different matter and is visible to anyone reading the chapter carefully.
Do I need a pilot test if I am using a published instrument?
Yes. Reliability is a property of a scale in a particular population, so compute your own coefficient on your own respondents and report it alongside the author’s.
Is reliability the same as validity?
No. Reliability is consistency — whether the items behave alike. Validity is correctness — whether the instrument measures what you claim. Both belong in Chapter Three and they are established differently.
What if my department does not ask for Cronbach’s alpha?
Some departments accept a test-retest procedure instead. Follow your handout, name whichever method you used, and report the resulting coefficient with the sample it came from.
Where does the alpha go in my project?
In Chapter Three, in the reliability of the instrument section, with the number of items and the pilot sample. It is not a finding, so it does not belong in Chapter Four.