Petra was doing exactly what I tell coaches to do before buying any instrument: reading the technical documentation, not the brochure. On the front page she found our headline figure, α = 0.94. Four pages in she found a subscale at 0.68, and she wrote to us with the question that deserves a public answer: if the instrument is so reliable, what is that number doing in your own table?
This article is the reply. The short version:
Cronbach’s alpha, formalised by the American psychologist Lee Cronbach in 1951, is a statistic between 0 and 1 measuring internal consistency: how strongly the items on a scale move together. By convention, values around 0.70 are acceptable, 0.80 good, and 0.90 excellent. Alpha also rises with the number of items, so a long scale scores higher than a short one of equal quality, which is why an instrument’s overall figure always exceeds its subscale figures.
Petra’s 0.68 was not a confession. It was arithmetic, and once you can see why, you will read every instrument’s reliability table more accurately than most of the people selling instruments would like.
Table Of Contents:
- What Cronbach’s alpha measures
- What counts as a good alpha
- Why more items raise alpha
- The PRI’s table, read correctly
- What alpha does not tell you
- FAQ
- References
What Cronbach’s alpha measures
Alpha asks one question of a scale: do the items built to measure one thing actually rise and fall together across respondents [1]? If a person who agrees with one optimism item tends to agree with the rest, the items are measuring something coherent, and alpha is high. If agreement on one item tells you nothing about the next, there is no single thing being measured, and alpha is low.
The everyday version: ask one friend whether a film is worth seeing and you get one person’s mood; ask nine and the stray answers cancel out. Each coherent item on a scale is another friend giving an independent reading of the same underlying thing, which is what makes multi-item measurement trustworthy, and what makes alpha sensitive to how many friends you asked.
What counts as a good alpha
The professional conventions: around 0.70 is acceptable, 0.80 is good, 0.90 and above is excellent [2][3]. The PRI’s overall figure of 0.94 sits in the top band [4].
And one caveat that honest instruments state rather than bury: alpha can be too high. A value near or above 0.95 can signal redundancy, near-duplicate items padding the scale instead of each adding information [3]. So does the PRI’s 0.94 sit too close to that ceiling? No, and the reason is the development history rather than a rhetorical move: redundancy is caught at item level, and the PRI’s validation did exactly that work, cutting 260 candidate items down to the 64 scored items that each earned their place [4]. A high alpha after that discipline reflects coherence. A high alpha instead of that discipline is the thing to be suspicious of, and the technical documentation is where you can tell the difference.
Why more items raise alpha
Now the part that answers Petra’s email. Alpha is not a pure quality score. It is mathematically sensitive to scale length: all else equal, adding coherent items pushes alpha up, because each one contributes signal while the noise in individual answers averages out [2][3]. This is built into how the statistic is defined, not a quirk of any instrument.
The consequence follows directly. An instrument’s overall score pools every item, so its alpha is high almost mechanically, provided the items cohere. Its subscales measure one narrow capacity each, with a handful of items, and post lower figures because they are short. Comparing a subscale’s alpha against the whole instrument’s is comparing a short scale with a long one, and the length moves the number before quality enters the picture.
The PRI’s table, read correctly
Apply that to the numbers Petra was reading. The full PRI, pooling all 64 scored items, reports α = 0.94. Its six domains are far shorter scales, and report 0.76 to 0.85, squarely in the acceptable-to-good range for their length [4]. Its twelve drivers are shorter still, several measured with only two items, deliberately, because a narrow, well-defined capacity does not need more.
Two items is where the length effect is starkest. Even a sound two-item scale struggles to post a high alpha. To see why without any real data getting in the way, picture two well-built two-item scales that measure their narrow capacities cleanly: one might land near 0.65, another near 0.75, and both are solid results at that length, not soft ones [4]. The same items, if you tripled their number, would push the figure higher on length alone. Reading a short scale’s alpha against the 0.94 is the misreading; the fair benchmark for any scale is what alpha can be expected to reach at its number of items.
This is also why we report reliability at more than one level instead of quoting the flattering overall figure and stopping, and the full evidence in context is on the Evidence page. One number flatters. A table informs.
What alpha does not tell you
Alpha is necessary and nowhere near sufficient, and any instrument leaning on it alone is telling you half a story.
It says nothing about validity: a set of items can cohere beautifully around the wrong construct [2]. It says nothing about behaviour over time: a scale can be internally consistent within one sitting and still measure a state that legitimately moves between sittings, which for a development instrument is the point, not a flaw. How those properties are established, and how they differ from consistency, is in how resilience is measured. If you are comparing several instruments’ figures side by side, the 2026 guide to resilience scales and assessment tools does that work, and every reliability figure you meet there deserves the same length-adjusted reading this article has taught you. Reliability is also only one of the properties field-wide reviews score instruments on, and the field’s larger verdict, that no instrument yet counts as a gold standard, is an argument of its own: is there a gold standard for measuring resilience?
Petra did not lower her guard after our reply. She upgraded it. She now asks every vendor for two things together, subscale alphas and the item counts behind them, and an instrument that will only show her the headline number does not make her shortlist. That is the reader this article hopes to make of you: not more trusting, better armed. The next time an instrument shows you a single headline figure, ask for the full table and the item counts underneath it. The question costs one email, and how a vendor answers it will tell you most of what you need to know.
FAQ
What is a good Cronbach’s alpha value?
Around 0.70 is acceptable, 0.80 is good, and 0.90 or above is excellent, by professional convention [2][3]. But the value is only half the question, because alpha rises with the number of items, so you should always read the figure against the scale’s length. And treat anything near or above 0.95 with suspicion rather than admiration, since at that level the items may simply be repeating each other [3]. High is good. Too high is a warning.
Is a Cronbach’s alpha of 0.6 acceptable?
It depends what you are asking 0.6 to do. On a long scale it is a weak result. On a two- or three-item subscale it can be a sound one, because alpha is mathematically constrained by item count [2]. So judge the figure against what a scale of that length can reach, not against a whole instrument’s pooled number. Length first, verdict second.
Why is Cronbach’s alpha lower for subscales than for the full scale?
Here is what the headline figure never tells you: alpha rises with the number of items, all else equal [2][3]. The full instrument pools every item and posts a high figure; each subscale measures one narrow capacity with far fewer items and posts a lower one, on length before quality. In the PRI, that is the whole explanation for 0.94 overall against 0.76 to 0.85 at domain level [4]. The gap is not a warning sign. It is arithmetic.
Is Cronbach’s alpha the same as validity?
No, and the difference matters more than most instrument pages admit. Alpha measures internal consistency, whether the items move together; validity is whether they measure the right thing, and a scale can be highly consistent about the wrong construct [2]. So when you evaluate an instrument, ask for both. Consistency without validity is a well-built ruler for the wrong dimension.
References
[1] Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16(3), 297-334.
[2] Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology, 78(1), 98-104.
[3] Tavakol, M., & Dennick, R. (2011). Making sense of Cronbach’s alpha. International Journal of Medical Education, 2, 53-55.
[4] Sinclair, N., Hafner, G., & Sinclair, P. D. (2022). Personal Resilience Indicator: Validation summary and psychometric properties (PRI Technical Report). Mind Matters.
