Alpha comes back at .41. The scale is a 12-item burnout measure with 214 nurses behind it, published, used for twenty years, and it has never once reported a coefficient that low. First reaction is that something is wrong with the sample. Second reaction, usually around midnight, is that something is wrong with the study.
Neither is likely. A coefficient that far below the published value almost never means the construct fell apart in your hospital. It means the numbers going into the formula are not the numbers you think they are.
What .41 Is Telling You
Cronbach's alpha measures internal consistency: how much the items move together. High alpha says people who agreed with one item tended to agree with the rest. When items pull in opposite directions, the covariances that drive the coefficient shrink, and alpha falls.
Now think about what a reverse-worded item does to that math. "I feel energized at the end of my shift" belongs on a burnout scale, but a burned-out nurse answers it with a 1 while answering every distress-worded item with a 5. Rescoring is what puts the item back on the same direction as the rest. Skip that step and the item does not simply add noise, it actively cancels the signal from the items around it.
The Item Pointing Backwards
Alpha alone cannot tell you which item is the problem, which is why the number by itself is a dead end at midnight. Item-total correlations can. Each one asks how strongly a single item tracks the total of the others, and on a scale that holds together they land somewhere around .30 to .70, all positive.
A negative item-total correlation is a different kind of finding. It says the item runs against its own scale. Item 9 comes back at -.34, and once you look, items 3, 9, 11, and 12 are all negative, all four of them the reverse-worded block. The recode step ran on the wrong column names, or ran before the merge, or never ran at all. Four items out of twelve pointing backwards is exactly the kind of thing that drags .88 down to .41.
One lone miscoded item costs less, which is why it hides longer. A single unrescored item on a 12-item scale might move alpha from .88 to about .70, low enough to worry about and high enough that some people write it up anyway. Read the item-total column and the cause is unmistakable either way.
Getting There Without the Detour
On a Mac the whole check is one sitting. The dataset is on the disk in front of you, the codebook is open in another window, and nothing has to be uploaded or configured before the first number appears.
Quanta is our statistics app for the Mac, and it prints the item-level detail with the coefficient by default. Run the reliability analysis and you get alpha, item-total correlations, and alpha-if-item-deleted in one output, so the diagnosis arrives with the symptom instead of requiring a second request. Flag the four reverse-worded items, rerun, and alpha comes back at .88. The whole correction takes about a minute, and the analysis runs locally on Apple Silicon, offline, which matters when the responses came from staff at a named hospital.
Quanta does not think for you. It will not tell you which items were meant to be reverse-worded, because that answer lives in the instrument and its manual. What it removes is the translation between your question and your evidence, and the question here was never hard: which item is fighting the scale?
What You Write Down
Report the corrected coefficient and the correction. Rescoring an item the manual says is reverse-worded is a fix, not a choice, and it belongs in the methods section in one sentence. Dropping items until alpha clears .80 is a different act entirely, and reviewers can tell the difference.
Two honest caveats. Alpha rises as items are added, so a long scale can post a comfortable coefficient while measuring two things at once. Internal consistency is also not validity: a scale can hold together beautifully and still measure something other than burnout.
What you get back is the hour after midnight you would have spent doubting the study. The data was fine. One column was upside down.
Quanta is available for the Mac at relicheck.com and on the Mac App Store, with a free 30-day trial.