Wilcoxon Signed-Rank Test in SPSS: Menus and Reporting

SPSS runs the Wilcoxon signed-rank test from two different menu locations, and they do not produce identical output. The Legacy Dialogs route gives you a compact Ranks table and a Test Statistics table with a Z and an asymptotic significance value. The newer Related Samples dialog runs through the Model Viewer, adds a one-line hypothesis-test summary, and exposes an exact test and a Hodges–Lehmann confidence interval without needing the separately licensed Exact Tests module for small samples. Both compute the same statistic on the same data; this guide covers both paths, what each output table actually contains, and how to get and report an effect size, since SPSS does not print one by default. For the statistical background — what the test’s null hypothesis actually is, how zero differences and ties are handled, and when to trust the exact test over the normal approximation — see CASRAI’s guide to Wilcoxon signed-rank test assumptions and reporting; this page assumes that background and focuses only on the SPSS mechanics.

nn

Two menu paths to the same test

n

Analyze > Nonparametric Tests > Legacy Dialogs > 2 Related Samples opens the classic dialog that has been stable across SPSS versions for over a decade. Analyze > Nonparametric Tests > Related Samples opens the newer, wizard-style dialog (available since SPSS Statistics 18, still current in the subscription version), which drives the Model Viewer output format used by several of SPSS’s newer procedures. Neither path is deprecated; IBM documents both. The legacy dialog is faster for a single, well-understood pairwise comparison. The newer dialog is worth the extra clicks when you want the exact test or the Hodges–Lehmann estimate without a separate module, or when the analysis is part of a larger workflow that already uses the newer nonparametric-tests family (e.g. alongside a Kruskal-Wallis or Friedman run in the same session).

nn

Running it via Legacy Dialogs

n

With your paired variables in two columns (one row per case, e.g. pretest and posttest):

n

n
1. Analyze > Nonparametric Tests > Legacy Dialogs > 2 Related Samples.

n

2. Move the first variable into the Variable1 cell of the Test Pairs box, and the second into Variable2. Each row in this box is one paired comparison; you can stack several pairs to run multiple Wilcoxon tests in one dialog.

n

3. Under Test Type, confirm Wilcoxon is checked (it’s the default; the Sign test checkbox next to it runs a different, less powerful test on the same pairs and is not required).

n

4. Optional: click Exact… to request the exact test instead of the normal approximation for small or heavily tied samples — this is part of the separately licensed IBM SPSS Statistics Exact Tests add-on; without it, only the asymptotic (normal-approximation) result is available from this dialog.

n

5. Optional: click Options for descriptive statistics (mean, SD, min, max, quartiles) on each variable, run alongside the test itself.

n

6. Click OK.

n

nn

Running it via the Related Samples dialog

n

Analyze > Nonparametric Tests > Related Samples opens a three-tab dialog:

n

n
• Objective tab — leave “Automatically compare observed data to hypothesized” selected for a quick default run, or choose Customize tests and check Wilcoxon matched-pair signed-rank test (2 samples) explicitly if you want to be certain which test runs (the automatic mode picks a test based on the data’s measurement level and distribution, which is not always the one you intend).

n

• Fields tab — move the paired variables into the Test Fields list.

n

• Settings tab — select Customize tests here too if you need the exact test or a specific confidence-interval level; the exact option is available directly in this dialog without the Exact Tests add-on for reasonably small samples, though very large exact computations still fall back to the asymptotic method.

n

n

Output lands in the Model Viewer as a one-line hypothesis-test summary (“Retain/Reject the null hypothesis” with a significance value). Double-click that summary row to open the detail view, which shows the same ranks-and-statistic information as the legacy output, plus, when requested, the Hodges–Lehmann estimate of the median difference and its confidence interval — a genuinely useful addition the legacy dialog does not offer.

nn

Reading the Ranks table

n

The Legacy Dialogs output (and the detail view under the newer dialog) starts with a Ranks table, split into three rows — Negative Ranks, Positive Ranks, and Ties — each showing N, Mean Rank, and Sum of Ranks, plus a Total row. “Negative Ranks” is the count of pairs where Variable2 < Variable1 (a decrease); “Positive Ranks” is Variable2 > Variable1 (an increase); “Ties” is the count of pairs with no difference at all. Two things catch people out here: which direction counts as positive depends entirely on the order you entered the two variables into the Test Pairs box, not on any substantive meaning SPSS infers — check the variable order before interpreting the sign of an effect. And the “Ties” row in this table is not a warning message, it’s the actual count of zero differences SPSS excluded before ranking, using the Wilcoxon convention (see the zero-handling discussion in the assumptions guide linked above for why other software can report a different N from the same data using a different convention).

nn

Reading the Test Statistics table

n

The Test Statistics table reports Z and Asymp. Sig. (2-tailed) — the standard normal-approximation p-value SPSS produces by default. If you requested the exact test, an additional Exact Sig. (2-tailed) column appears; for small or tied samples this can diverge meaningfully from the asymptotic value, and the exact figure is the one to trust and report in that situation. SPSS’s default output does not include an effect size, a confidence interval on the median difference (outside the newer dialog’s Hodges–Lehmann option), or the underlying rank sums as a separate labeled statistic — those come from the Ranks table above, or have to be computed by hand.

nn

Getting an effect size SPSS doesn’t print

n

SPSS reports Z but not a standardized effect size for this test. The matched-pairs rank-biserial correlation, r = (W⁺ − W⁻)/(W⁺ + W⁻), can be computed directly from the Sum of Ranks values already in the Ranks table — W⁺ is the Positive Ranks sum, W⁻ is the Negative Ranks sum. The alternative shortcut, r = Z/√N, is common in published papers but ambiguous unless you state explicitly whether N is the number of pairs or the number of observations (2 × pairs) — the two choices produce different numbers from the same output. See the effect-size section of the assumptions and reporting guide for the full derivation and citation. Report which formula and which N you used; a bare “r = .42” without that context is not reproducible from your SPSS output alone.

nn

Worked example

n

Ten participants rate task difficulty before and after a training intervention on a 1–10 scale. Running Analyze > Nonparametric Tests > Legacy Dialogs > 2 Related Samples with difficulty_pre and difficulty_post as the pair produces a Ranks table showing, say, 8 Negative Ranks (scores decreased — task felt easier afterward), 1 Positive Rank, and 1 Tie, with a Sum of Ranks of 42 for the negative ranks and 3 for the positive rank. The Test Statistics table, computed from those same rank sums, reports Z = −2.31 with Asymp. Sig. (2-tailed) = .021. The rank-biserial effect size from the formula above is r = (3 − 42)/(3 + 42) = −.87. Reported in text: “A Wilcoxon signed-rank test indicated that perceived task difficulty was significantly lower after training (Mdn change = −2) than before (Z = −2.31, p = .021, r = .87, N = 9 ranked pairs, 1 tie excluded).” This example is illustrative to show which numbers map to which output cells, not a real published dataset — run the actual test on your own data before citing a result.

nn

Which nonparametric test is this, and when does it apply

n

The Wilcoxon signed-rank test is for two related (paired, repeated-measures) samples — the same cases measured twice, or matched pairs. It is not interchangeable with the Mann-Whitney U test, which is the nonparametric option for two independent groups; SPSS keeps these as separate procedures for exactly that reason, and picking the wrong one because “they’re both nonparametric” is a common, avoidable error. Use CASRAI’s statistical-test decision tree if you’re unsure which design you have. Against the parametric alternative — the paired-samples t-test, covered for SPSS in CASRAI’s t-test in SPSS guide — the Wilcoxon signed-rank test is the standard fallback when the distribution of paired differences is not reasonably symmetric/normal, or when the outcome is ordinal rather than interval. Check that assumption with SPSS’s Explore procedure before choosing; see CASRAI’s guide to normality of distribution and to levels of measurement for how to make that call.

nn

Frequently asked questions

n

Do I need the Exact Tests add-on to get an exact p-value?

n

Only from the Legacy Dialogs route. The newer Analyze > Nonparametric Tests > Related Samples dialog can compute the exact test itself for reasonably sized samples without that separately licensed module — check the Settings tab’s Customize Tests option.

n

Why do the Ranks table’s Negative and Positive labels seem reversed from what I expected?

n

SPSS defines the sign relative to Variable2 minus Variable1, in the order you placed them in the Test Pairs box (Legacy Dialogs) or Test Fields list (newer dialog) — not relative to any assumption about which variable is “before” or “after.” Check the variable order, not just the labels, before interpreting direction.

n

My Ties row isn’t zero — does that invalidate the test?

n

No. Ties (pairs with zero difference) are a normal, expected part of real data and SPSS excludes them from the ranking by default, which is the standard Wilcoxon convention. It only becomes a real concern when ties make up a large share of the sample, which shrinks the effective sample size the test is actually run on — see the zero-handling section of the linked assumptions guide.

n

Can I run more than one pair at a time?

n

Yes, in the Legacy Dialogs box — add multiple variable pairs as separate rows in the Test Pairs box and SPSS runs a separate Wilcoxon test per row, printing one Ranks/Test Statistics pair per comparison. It does not apply any multiple-comparisons correction across those rows automatically; add one yourself (e.g. Bonferroni) if you’re running several related comparisons and treating them as a family.

n

What effect size should I report, and does SPSS calculate it for me?

n

SPSS does not print a standardized effect size for this test. Compute the matched-pairs rank-biserial correlation from the Sum of Ranks values in the Ranks table, and state your N convention explicitly if you instead use the r = Z/√N shortcut — see “Getting an effect size SPSS doesn’t print” above.

nn

Related: CASRAI’s research-tools hub for the full statistical-software collection, and the Kruskal-Wallis test for the nonparametric option when comparing three or more related or independent groups.

n