What Is a Good Cronbach’s Alpha for a Dissertation, and How Do You Report It? (South Africa, 2026)

For most dissertation questionnaires, a Cronbach’s alpha of about 0.70 or higher is the conventional minimum, and values up to roughly 0.90 are treated as strong. Calculate it for each subscale on your own sample, report the number of items and respondents, and never present alpha as proof that your instrument is valid.

What does Cronbach’s alpha actually measure?

Cronbach’s alpha is a measure of internal consistency: the extent to which the items that are supposed to measure one construct move together in your data. It was introduced by Lee Cronbach in a 1951 paper in Psychometrika (Cronbach, 1951), and it has become the default reliability coefficient for questionnaires scored on a Likert scale.

Three practical points follow from how it works:

• It is a property of scores from your sample, not a permanent property of the questionnaire. The same instrument can give different alphas in different samples.
• It depends on the number of items. Tavakol and Dennick (2011) note that alpha rises when more related items are added and falls when the test is very short.
• It is calculated for one scale or subscale at a time. Mixing items from different constructs into one alpha produces a number that means nothing.

What is a good Cronbach’s alpha for a dissertation?

The convention you will meet in South African methodology chapters is 0.70. The Utrecht Work Engagement Scale test manual, for instance, describes 0.70 as the critical value it draws from Nunnally and Bernstein (1994), and reports that alphas for its three scales usually range from 0.80 to 0.90. Tavakol and Dennick (2011), writing for medical education, report that acceptable values are described in the literature as ranging from 0.70 to 0.95, and add that a maximum of 0.90 has been recommended.

A workable reading for a student, in line with those sources, is:

These bands are conventions, not laws. Keith Taber’s review of how alpha is used and reported in research instruments (Taber, 2018) is a useful read if your supervisor questions a threshold: his concern is that the number has to be interpreted in the context of the scale, not applied mechanically. Your faculty guide or your supervisor’s preference overrides any general rule.

How do you calculate Cronbach’s alpha in SPSS?

The steps are the same for any scale with several Likert items. The site’s longer guide to analysing questionnaire data in SPSS covers the rest of the workflow.

1. Reverse-score first. If some items are negatively worded, recode them so that a high score means the same thing on every item. An unrecoded item can push alpha down, or even below zero.
2. Open the reliability dialog. In SPSS choose Analyze, then Scale, then Reliability Analysis.
3. Move the items of one subscale into the Items box and keep the model set to Alpha.
4. Open Statistics and tick “Scale if item deleted”. Add inter-item correlations if you want them.
5. Run it and read three things: the overall Cronbach’s Alpha, the “Corrected Item-Total Correlation” column, and the “Cronbach’s Alpha if Item Deleted” column.
6. Repeat for each subscale, and for the total scale only if you will analyse a total score.

The same coefficient is available in free tools. In jamovi it sits under the Factor menu as Reliability Analysis, and in R the alpha() function in the psych package returns it with item statistics. The site’s comparison of SPSS, R, jamovi and JASP helps you choose if you do not have a university licence.

What does a reliability result look like, and how do you write it up?

Illustrative example, not a real study. Imagine an honours psychology study of academic self-efficacy among 120 first-year students, using an eight-item subscale adapted from a published instrument. The output you would tabulate looks like this:

Overall alpha for the eight items is 0.82. Item 5 stands out: its corrected item-total correlation is 0.12, and the alpha would rise to 0.84 if it were removed. Before deleting anything, look at the item. If it was negatively worded and you forgot to reverse-score it, recode it and rerun. If it is simply ambiguous, you may decide to drop it, but you must say so in the methodology chapter and note that the instrument now differs from the published version.

A reporting sentence that examiners accept names the scale, the number of items, the sample size and the coefficient, in the same place every time:

“The eight-item academic self-efficacy subscale showed good internal consistency in this sample (Cronbach’s α = .82, n = 120).” Then add one clause comparing it with the value in the original validation study, so that the reader sees both.

When is Cronbach’s alpha the wrong tool?

Alpha fits a multi-item Likert scale. Other designs need a different coefficient:

• Items scored right or wrong (0 or 1). Use KR-20, the coefficient from Kuder and Richardson (1937), which is alpha for dichotomous items. The same SPSS dialog returns it when the items are coded 0 and 1.
• Stability over time. Use test-retest reliability: the same people complete the instrument twice and you correlate the two sets of scores. State the interval and why you chose it, because too short invites memory effects and too long invites genuine change.
• Coding or observation by two people. Use an agreement coefficient such as Cohen’s kappa for the codes, which is a common way to show that two coders applied a coding frame consistently in content analysis and observation schedules.
• A scale with more than one dimension. Report alpha for each subscale. Tavakol and Dennick (2011) warn that alpha does not simply measure unidimensionality, so a decent alpha on a mixed set of items does not show that one construct is being measured.
• A single-item measure. Alpha cannot be computed. Say that reliability could not be estimated and treat it as a limitation.

If your supervisor asks about alternatives to alpha, omega coefficients are the one most often mentioned, and the jamovi and R tools above can compute them. Check your faculty’s expectation before swapping coefficients, because many South African methodology templates still ask for alpha by name.

What should you do when alpha comes out low?

A low alpha is a finding, not a disaster, if you handle it honestly. Work through these checks in order:

1. Reverse-coding. This is the most common cause of a surprising result. Check every negatively worded item.
2. Scale length. A subscale of three or four items often gives a modest alpha simply because it is short.
3. Mixed constructs. If the items you grouped measure different things, split them according to the original instrument’s structure.
4. Language and translation. If you used an Afrikaans, isiZulu or other translated version, or changed wording for a South African sample, say so, because adaptation can change how items behave. The site’s guide to proving a questionnaire is valid covers the adaptation steps.
5. Sample size and range. A very small or very homogeneous sample can depress alpha.
6. Item deletion, last. Delete only when the item is also weak on content grounds and the corrected item-total correlation is low, and report both the original and the revised alpha.

If you cannot fix it, report the low value, explain the likely reason and state it as a limitation. Examiners penalise hidden problems far more than reported ones, and a transparent limitation is easier to defend at the examination stage than a number you quietly adjusted.

Where do reliability results go in the methodology chapter?

Put the instrument’s description, the original reliability evidence from the validation study and the reliability you found in your own pilot and main sample in the instrument section of the methodology chapter. Many students use a single table: scale, number of items, original alpha (with reference), pilot alpha and main-study alpha. The table lets the examiner see at a glance that you did not rely on a number borrowed from another country’s sample. If you are still deciding which instrument to use, the site’s guides to validated scales for an HRM dissertation and validated scales for a nursing research report show how published instruments are described.

Which mistakes make examiners doubt a reliability section?

• Reporting only the alpha from the original paper and none from your own data.
• Calculating one alpha across items from several subscales.
• Forgetting to reverse-score negatively worded items.
• Deleting items until the alpha reaches 0.70 without a content reason.
• Treating a high alpha as evidence of validity.
• Giving the coefficient without the number of items and respondents.
• Using alpha for dichotomous items instead of KR-20, or for a single item.

How can Tesify help with your reliability section?

Writing the instrument section, the reliability table and the limitation paragraph so that they match your data takes careful organising. 9,000+ students have written 15,000+ chapters with Tesify, which helps you structure and organise your dissertation, and the text stays 100% written by you. Organise your methodology chapter with Tesify.

Frequently asked questions

Is a Cronbach’s alpha of 0.60 acceptable for a dissertation?

It is below the 0.70 convention most examiners expect. Some supervisors accept a lower value for a short scale in an exploratory study, but you must report it honestly, explain why it is low and say what it means for your conclusions.

Can Cronbach’s alpha be too high?

Yes. Tavakol and Dennick (2011) note that a very high alpha may mean some items are redundant, and they report that a maximum of 0.90 has been recommended.

Do I report alpha for the whole questionnaire or for each subscale?

Report it for each subscale that you score and analyse on its own. Report a total-scale alpha only if you also use a total score.

Is the alpha from the original validation study enough?

No. Reliability belongs to scores from a particular sample, so you should report the original value for context and your own value from your data.

What do I use for yes/no or right/wrong items?

For items scored 0 or 1 the equivalent coefficient is KR-20, from Kuder and Richardson (1937). Most statistics packages return it when you run alpha on dichotomous items.

Does a high alpha prove my questionnaire is valid?

No. Alpha says only that the items agree with each other. Validity needs separate evidence, such as expert review and construct evidence.

Should I delete an item to raise my alpha?

Only if the item is also weak on content grounds. Deleting items to chase a number changes the construct you measure, and an examiner may ask you to justify it.

How many respondents do I need to calculate alpha?

There is no single rule. Alpha computed on a very small pilot is unstable, so state your sample size next to the value and treat a small-pilot alpha as provisional.

References cited

• Cronbach, L.J. (1951) Coefficient alpha and the internal structure of tests. Psychometrika, 16(3), pp. 297–334.
• Kuder, G.F. and Richardson, M.W. (1937) The theory of the estimation of test reliability. Psychometrika, 2(3), pp. 151–160.
• Taber, K.S. (2018) The use of Cronbach’s alpha when developing and reporting research instruments in science education. Research in Science Education, 48(6), pp. 1273–1296.
• Tavakol, M. and Dennick, R. (2011) Making sense of Cronbach’s alpha. International Journal of Medical Education, 2, pp. 53–55.