The table that tells you which items are doing the work and which are pulling against them. How to read a corrected item-total correlation, what alpha-if-deleted is really for, and when removing an item is defensible.
Quanta's item-total table carries four columns for every item: scale mean if deleted, scale variance if deleted, corrected item-total correlation, and alpha if deleted.
The corrected item-total correlation is the one to read first. It is the correlation between that item and the total of all the other items. The correction matters: the item is excluded from the total it is compared against, because leaving it in guarantees a correlation with itself and inflates every value in the column.
Read it as how well each item agrees with the rest of the scale. A rough guide: above .30 is a contributing item, .20 to .30 is weak, below .20 is contributing little, and negative is a red flag.
These are guides, not thresholds. A conceptually essential item at .25 may belong in the scale, and dropping it for a statistical reason can damage content validity in exchange for a small gain in a coefficient.
When an item correlates negatively with the rest of the scale, the overwhelmingly likely explanation is a reverse-worded item that was never rescored.
The pattern is unmistakable once you know it. On a 24-item emotional intelligence scale, nineteen items sit between .58 and .70 while five sit between −.65 and −.70. Every one of those five carries an _R suffix in its name. That is not five bad items; that is one missing scoring step.
The second tell is the alpha-if-deleted column. Those same five items each show alpha rising from .81 to roughly .85 if removed, while every other item would lower it. When several items cluster like that, look at their wording before you look at their quality.
Where the item is genuinely not reverse-worded, a negative correlation means it is measuring something else, or measuring nothing. That is a content problem rather than a scoring one, and it is worth understanding before the next round of data collection.
This column shows what alpha would become if each item were removed. Its intended use is diagnostic: an item whose deletion would substantially raise alpha is behaving differently from the rest, and that is worth investigating.
Its common use is worse. Researchers scan the column, delete whichever item raises alpha most, and repeat until a threshold is cleared. That is optimizing to a sample. The improvement is partly real and partly noise fitted to these particular respondents, and it will not replicate.
Three questions before removing an item. Does its content belong in the construct as you defined it? Is there a substantive reason it behaves differently, such as double-barrelled wording or an ambiguous stem? Is the gain meaningful, or is it .82 to .84?
A trivial gain is never worth breaking comparability with a published version of the instrument. If you do remove items, report the original and revised values, name the items, and give the reason. The reason should be about the item, not about the coefficient.
Lay the data out with one row per respondent and one column per item.
Rescore reverse-worded items before you read the table, or the diagnostics will point you at the wrong problem.
Configure the item set on the right. Item-total statistics appear inside the Reliability statistics card, beneath the scale statistics and the summary item statistics, so the diagnosis sits next to the number it explains.
Check the summary item statistics first. The mean, minimum, and maximum inter-item correlation compress the whole item-total table into three numbers. A negative minimum tells you there is a problem before you read a single item row.
Watch the cases used. Listwise deletion drops any respondent missing one item, so a long scale can lose more of the sample than you expect.
Item-level statistics usually belong in a table rather than in prose, particularly for a scale being developed or adapted.
Corrected item-total correlations ranged from .58 to .70, and no item's removal would have increased alpha by more than .01. All items were retained.
Reporting the range and stating that no deletion would help is a compact way to show the scale is sound and that you checked.
Where items were removed, give the before and after values, name the items, and state the substantive reason. An unexplained deletion reads as fitting the data.
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 procedures, including alpha and item-level statistics, are validated against R and Python and against closed-form values, with alpha agreeing to ten digits. 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.
Cronbach's alpha the scale-level estimate · Reverse-coded items the first thing a negative correlation should make you check · McDonald's omega reliability without the equal-loadings assumption · the complete list on the analyses page.