Cronbach’s Alpha in SPSS: Reliability Analysis and Item-Total Statistics

Computing Cronbach’s alpha in SPSS takes one menu path and produces a table most users skim past — the item-total statistics block that tells you whether any single item is dragging the scale down. This guide covers the Reliability Analysis menu path, the checkboxes that get you item-level output (not just the overall coefficient), how to read the “Cronbach’s Alpha if Item Deleted” column to spot a problem item, and where the field-specific threshold conventions come from.

Before you run it: reverse-code negatively worded items

SPSS computes alpha from the raw values you give it. If your scale mixes positively and negatively worded items (e.g., “I feel confident” alongside “I feel anxious”) and you don’t reverse-score the negative items first, alpha will come out artificially low — sometimes negative — because the items are correlating in opposite directions by design, not because the scale is unreliable. Reverse-code first: Transform > Recode into Different Variables, mapping a k-point scale so that old value 1 becomes k, 2 becomes k−1, and so on, and give the recoded item a distinct variable name (e.g., item4_r) so you don’t overwrite your raw data. Run Reliability Analysis on the reverse-coded versions, not the originals.

Running Reliability Analysis: the menu path

Go to Analyze > Scale > Reliability Analysis. Move every item that belongs to the scale into the Items box — include only items intended to measure the same construct; mixing items from unrelated subscales into one run will suppress alpha and make the item-total statistics uninterpretable. Leave Model set to Alpha (the default) unless you specifically need split-half or another model.

Click Statistics and check these boxes before running:

• Item — descriptives for each item individually.
• Scale — descriptives for the total scale score.
• Scale if item deleted — this is the one that matters most here. Without it checked, SPSS prints only the overall alpha and none of the item-total diagnostics covered below.

Click Continue, then OK. SPSS returns two tables: Reliability Statistics (the overall Cronbach’s Alpha and the item count) and Item-Total Statistics (one row per item) — the second table is where item-level diagnosis happens.

Reading the Item-Total Statistics table

Each row corresponds to one item in the scale. The columns that matter:

• Scale Mean if Item Deleted / Scale Variance if Item Deleted — what the scale’s total score and variance would look like without that item. Rarely the deciding factor on their own; used mainly to sanity-check the other two columns.
• Corrected Item-Total Correlation — the correlation between that item and the sum of all other items in the scale (the item’s own contribution is excluded, which is what makes it “corrected” rather than a plain item-to-total correlation). A low value here means the item isn’t moving together with the rest of the scale.
• Squared Multiple Correlation — how well the other items, taken together, predict this one. Useful as a secondary check when the corrected item-total correlation is borderline.
• Cronbach’s Alpha if Item Deleted — the coefficient the scale would have if that specific item were removed and alpha recalculated on the remaining items. This is the direct answer to “is this item helping or hurting reliability.”

Using “Alpha if Item Deleted” to spot a problem item

Compare each row’s “Cronbach’s Alpha if Item Deleted” value against the overall alpha printed in the Reliability Statistics table:

• If a row’s value is higher than the overall alpha, removing that item would improve the scale’s internal consistency — that item is the weak link. Pair this with a low Corrected Item-Total Correlation (conventionally below about 0.30, though some methodologists use 0.20 as the floor) before concluding the item is genuinely problematic rather than just contributing less than the others.
• If every row’s value is lower than the overall alpha, every item is contributing positively and there’s no single item to flag on this criterion alone.

A flagged item doesn’t automatically get deleted. Check first whether it’s measuring a genuinely different sub-dimension of the construct (in which case the fix may be splitting the scale, not dropping the item), whether it was worded ambiguously or double-barreled, or whether it simply needs reverse-coding you missed in the step above. Dropping items purely to chase a higher alpha number, especially in a scale that’s already published or validated, trades content coverage for a marginally higher coefficient — a tradeoff worth documenting, not doing silently.

What counts as an acceptable alpha, and why it depends on the field

SPSS prints the overall Cronbach’s Alpha as a single number with no built-in verdict on whether it’s “good enough” — that judgment depends on what the score is used for and which field’s convention applies. As a starting reference: 0.70–0.79 is broadly treated as acceptable for exploratory research and newly developed instruments; 0.80–0.89 as good, the usual target for published survey research; and 0.90 and above as excellent for reporting purposes but worth checking for item redundancy, since it can mean several items are asking near-identical questions rather than each contributing distinct information. Fields that use a score to make decisions about individual people — clinical diagnosis, patient-reported outcome measures used for individual-level decisions, high-stakes admissions or licensure testing — typically expect a materially higher bar than a group-comparison study in exploratory social-science research, where 0.70 is often treated as sufficient. See CASRAI’s full Cronbach’s alpha interpretation guide for the complete threshold table, the case for reporting the confidence interval around alpha rather than the point estimate alone, and when McDonald’s omega is the more defensible statistic than alpha in the first place.

Reporting the result

APA style (7th edition) calls for the number of items (k), the sample size (N), and the alpha coefficient itself, ideally alongside a confidence interval rather than the point estimate alone. A minimal reporting line: “The scale demonstrated acceptable internal consistency (k = 8 items, N = 214, α = .81).” If you flagged and removed an item using the process above, report both the original and the revised alpha, and say explicitly which item was dropped and why — a silent item count change between a scale’s original publication and your use of it is the kind of thing a reviewer or replicator will ask about.

Frequently asked questions

Why is my Cronbach’s alpha negative?

A negative alpha almost always means at least one item correlates negatively with the rest of the scale — most commonly because a reverse-worded item wasn’t reverse-coded before running the analysis. Check the Corrected Item-Total Correlation column: a negative value there points directly at the item causing it.

Do I need “Scale if item deleted” checked to get the overall alpha?

No — the Reliability Statistics table with the overall Cronbach’s Alpha prints regardless. “Scale if item deleted” is what additionally produces the Item-Total Statistics table, which is where the item-level diagnostics live. If you only see the overall number, that box wasn’t checked.

Should I drop every item with a low corrected item-total correlation?

Not automatically. A low correlation flags an item worth reviewing for content and wording, not an automatic removal candidate — especially in a validated or previously published scale, where changing item composition changes what the instrument has actually been shown to measure. Weigh the reliability gain against the content-validity cost.

Can I run Reliability Analysis on a scale with only two items?

Yes, though alpha is mathematically more sensitive to item count with very short scales — a two-item scale needs a comparatively high inter-item correlation to reach conventional thresholds, and the Item-Total Statistics table is less informative with only two rows to compare. For two-item scales, some methodologists prefer reporting the raw inter-item correlation directly instead of alpha.

Related CASRAI resources

For the full conceptual treatment — what alpha actually measures, the complete field-specific threshold table, and when to use McDonald’s omega instead — see CASRAI’s Cronbach’s alpha guide and reliability and APA-format reporting guide. For the software itself, see what SPSS is and who uses it, or compare it against alternatives in SPSS vs. R and SPSS vs. Stata. To run the equivalent analysis in a different package, see CASRAI’s t-test in SPSS and t-test in Stata guides for the same menu-path-and-output-table approach applied to a different procedure. To confirm your items are genuinely measuring one construct before trusting any alpha figure, see levels of measurement.