The most reported and most misread statistic in survey research. What alpha actually measures, the four things it is routinely mistaken for, and how Quanta writes the limits into the sentence you paste.
Cronbach's alpha estimates internal consistency: the extent to which the items on a scale move together. If six self-awareness items all tap the same underlying thing, someone who scores high on one should tend to score high on the others.
Mechanically it compares the variance shared among items against the total variance. An alpha of .85 means most of what the scale measures is common to its items rather than item-specific noise.
That is the whole of it, and it is narrower than the way alpha gets used. Alpha is a property of a set of responses from a particular sample, not a fixed property of an instrument. The same questionnaire returns different alphas in different populations, which is why quoting the alpha from the original validation study rather than from your own data is a mistake reviewers increasingly catch.
Quanta states this on the result rather than leaving it to you to remember. The card carrying the interpretation is badged Do not overclaim validity, and the report-ready sentence it produces ends with the limit attached: alpha is internal-consistency evidence only, it does not establish validity.
Alpha is not evidence of unidimensionality. This is the most consequential misunderstanding. A scale made of two distinct but correlated factors can produce a comfortable alpha while measuring two things. Alpha cannot detect that, and a high value is regularly cited as though it had. If unidimensionality matters to your argument, a factor analysis answers that question and alpha does not.
Alpha is not validity. It says the items agree with each other, not that they measure what you named them. Six items can consistently measure the wrong construct.
Alpha does not have a threshold at .70. The figure comes from a passing suggestion in Nunnally's textbook for early-stage research and hardened into a rule nobody can source. A .65 on a short scale in exploratory work may be fine. A .70 on a high-stakes placement instrument is not.
Alpha rises with item count regardless of quality. Add enough mediocre items and alpha climbs. A .94 on forty items is less impressive than a .82 on five, and very high alphas often signal redundant items rather than an excellent measure. Above roughly .95, suspect you are asking the same question repeatedly.
The Signal takes each of these apart at length in why most researchers misinterpret Cronbach's alpha, and follows one live case in the reverse-coded item that wasn't.
The run opens with three figures: Cronbach's alpha, the cases used with the missing-data rule named as listwise deletion, and the number of items in the scale. Read the case count against your sample size. Listwise deletion drops any respondent missing a single item, so a 24-item scale can lose a substantial share of a sample to scattered non-response.
The Reliability statistics card carries three tables. Scale statistics gives N, mean, variance, and SD for the total score. Summary item statistics gives the mean, minimum, maximum, range, max/min ratio, and variance for item means, item variances, and inter-item correlations. Item-total statistics gives scale mean if deleted, scale variance if deleted, corrected item-total correlation, and alpha if deleted, for every item.
The summary table is the fastest diagnostic on the screen and it is usually skipped. A mean inter-item correlation of .15 with a minimum of −.66 tells you immediately that something is wrong, because items on a coherent scale do not correlate negatively with each other. That single line will find a scoring error faster than reading twenty-four item rows.
The Plain-language interpretation card then gives a report-ready readout, an interpretation, a recommended action, and a caution. The recommendation is to report with item-total diagnostics and validity limitations, which is a prompt to include the evidence rather than only the coefficient.
Alpha assumes tau-equivalence: that every item contributes equally to the underlying construct, in other words that all the factor loadings are the same.
Real scales rarely satisfy this. Some items are stronger indicators than others, which is the ordinary case rather than a defect. When loadings differ, alpha underestimates reliability, sometimes substantially.
So a scale reported at .68 and dismissed as inadequate may have a true reliability well above .70. This is the practical reason methodologists recommend McDonald's omega, which does not make the equal-loadings assumption.
Lay the data out with one row per respondent and one column per item. Item-level responses, not a total score. Alpha cannot be computed from a summed scale.
Rescore reverse-worded items first. An unreversed item does not merely weaken alpha, it drags it down sharply. A 24-item scale with five unrescored reverse items can report .81 when its true internal consistency is well above .90. See reverse-coded items.
Quanta detects scale items automatically and offers them for a reliability analysis, which saves selecting two dozen columns by hand and reduces the risk of quietly omitting one.
Configure the item set on the right and the tiles, tables, and interpretation appear in the center panel.
Use the action bar to add the result to a report, copy the table, or copy the APA text. Every result carries the ReliCheck Intelligence card for a plain-language explanation computed on your Mac.
Report alpha computed on your own sample, with the number of items and the sample size. Two decimal places, no leading zero.
A reliability analysis was carried out on the scale (k = 24 items; n = 250). Internal consistency was good, Cronbach's α = .81. Alpha is internal-consistency evidence only; it does not establish validity.
That is the sentence Quanta produces, and the final clause is the part worth keeping. Most reliability sections stop at the coefficient and leave a reader to infer that a good alpha means a good measure.
Report the number of items and the number of cases. Alpha without k cannot be judged, because the coefficient rises with item count.
If you removed items to raise alpha, say which and why, and give both values. Silently deleting items until a threshold is met is a form of overfitting.
Every result carries a ReliCheck Intelligence card with one action: explain this result in plain language. It runs entirely on your Mac with Apple Intelligence. No upload, no API key, nothing to configure. The model only explains the result Quanta has already computed and validated. It never computes a statistic, and it never applies a decision on your behalf.
From the action bar you can add the result to a report, copy the table, or copy the APA text. Exported reports carry the APA table, its note, and a caution specific to that analysis beneath it.
Quanta's reliability estimates are validated against R and Python and against closed-form values, with alpha agreeing to ten digits. Cronbach's alpha matches R 4.6 and Python 3 to ten decimal places on the published head-to-head. The engine is pure Swift with no R or Python at runtime. Full record on the validation page.
Scale & Reliability is part of every Quanta subscription, $9.99 a month or $79.99 a year after the 30-day trial. One price for everyone. See pricing.
The analysis, the report builder, and the on-device explanations all work with the network off. Your dataset is never uploaded.
Item-total statistics which items are carrying the scale · Reverse-coded items the most common cause of a low alpha · McDonald's omega reliability without the equal-loadings assumption · the complete list on the analyses page.