How to Run a T-Test in SPSS and Read the Output

SPSS runs three different procedures under the name “t-test,” and each one lives at a different menu path with its own output table. Picking the wrong one, or reading the wrong row of the output it gives you, is the most common way this analysis gets reported incorrectly. This guide covers all three menu paths, how to read Levene’s Test for Equality of Variances to pick the correct row of the independent-samples output, and how to turn the numbers SPSS prints into a correctly formatted result line.

The three t-tests SPSS runs, and which one you need

SPSS groups all three under Analyze > Compare Means (relabeled Compare Means and Proportions in SPSS Statistics 29 and later; the three procedures below sit in the same place under either name):

• One-Sample T Test — compares a single sample’s mean against a fixed, known value (a population mean, a test standard, a benchmark score).
• Independent-Samples T Test — compares the means of two separate, unrelated groups (e.g., treatment vs. control, two different cohorts) recorded in one grouping variable.
• Paired-Samples T Test — compares two measurements taken from the same cases (before/after, two conditions on the same participants, two raters on the same items).

The choice is about your study design, not your data values — if the same participant contributes both scores, it is paired, not independent, regardless of how the columns are arranged.

Independent-samples t-test: menu path and setup

Go to Analyze > Compare Means > Independent-Samples T Test. Move your continuous outcome into Test Variable(s) and your two-group categorical variable into Grouping Variable. SPSS will show the grouping variable as var(? ?) until you click Define Groups and type the two group codes exactly as they’re coded in your data (e.g., 1 and 2, not the text labels) — this step is the most common place people get stuck, and SPSS will not run the test until it’s filled in.

Paired-samples t-test: menu path and setup

Go to Analyze > Compare Means > Paired-Samples T Test. Unlike the independent-samples test, there is no grouping variable here — your data needs to already be in wide format, with each row representing one case (one subject, one pair) and the two measurements for that case sitting in two separate columns (e.g., pretest and posttest, or rater1 and rater2). If your paired data is currently stacked in one long column with a time or condition flag, restructure it to wide format first (Data > Restructure) — the Paired-Samples T Test dialog cannot use long-format data directly.

In the dialog, click one variable, then Ctrl-click (Cmd-click on Mac) the second, and move both into Paired Variables together — they populate as a single numbered row, Pair 1, with your first click becoming Variable 1 and your second becoming Variable 2. That click order sets the direction SPSS subtracts in (Variable 1 minus Variable 2), so keep it deliberate if the sign of the difference matters for your write-up. When a dataset has several pre/post outcome columns — say, an anxiety score and a depression score, each measured twice — it’s easy to pair the wrong two columns together by clicking in the wrong order. Check the Paired Variables list in the dialog before running, not just the output afterward: SPSS will run whatever pairing you gave it without warning you the two variables measure different things.

One-sample t-test: menu path and setup

Go to Analyze > Compare Means > One-Sample T Test. Move your variable into Test Variable(s) and enter the fixed comparison value in Test Value. This is the only one of the three tests with no Levene’s test in its output, since there is only one sample and no second variance to compare it against.

Reading Levene’s Test to pick the right row

The independent-samples output table has two rows — “Equal variances assumed” and “Equal variances not assumed” — and which one to report is decided by Levene’s Test for Equality of Variances, printed in the same table:

• If Levene’s Sig. value is greater than .05, the equal-variances assumption holds. Report the “Equal variances assumed” row (the standard Student’s t-test, with df = n1 + n2 − 2).
• If Levene’s Sig. value is .05 or less, variances differ significantly between groups. Report the “Equal variances not assumed” row — this is Welch’s t-test, and its degrees of freedom will be a non-integer (e.g., df = 27.43) because SPSS applies the Welch–Satterthwaite correction. A non-integer df in your output is a direct sign you’re looking at (or should be looking at) this row.

A frequent reporting error is quoting Levene’s own F and Sig. values as if they were the t-test result. They aren’t — Levene’s test only decides which row of t/df/Sig.(2-tailed) to report; it is not itself the finding to report unless variance equality is the actual research question.

Reading the rest of the output table

Once you’ve picked the correct row (or, for one-sample and paired tests, the only row), the columns you need are:

• t — the test statistic.
• df — degrees of freedom for that row.
• Sig. (2-tailed) — the p-value for a two-tailed test. If your hypothesis is directional, halve this value to get the one-tailed p and confirm the sign of t matches the predicted direction before halving.
• Mean Difference — the raw difference between the two means (or between the sample mean and the test value, for a one-sample test).
• 95% Confidence Interval of the Difference — report alongside the p-value; APA’s Publication Manual (7th edition) calls for an effect size and interval estimate to accompany significance testing for each primary outcome, not p alone.

The Paired Samples Correlations table: the output most guides skip

Running the Paired-Samples T Test procedure prints three tables, not one: Paired Samples Statistics (mean, N, SD, and standard error for each of the two variables), Paired Samples Correlations, and the Paired Samples Test table covered above. The correlations table sits between the other two and is easy to skip on the way to the p-value, but it’s diagnostic, not decorative.

It reports the Pearson correlation, with its significance, between your two paired variables — for example, how strongly pretest and posttest scores track each other across participants. A paired-samples design is more statistically powerful than an independent-samples design specifically because it removes between-subject variability from the error term, and that removal only works to the extent the two measurements are genuinely correlated within cases. A strong, positive, significant correlation here confirms the pairing is doing real work. A low, near-zero, or negative correlation is worth stopping on: it can mean the two variables aren’t really tracking the same underlying trait within subjects, that the pairing itself is misaligned (see the wrong-column-order caution above), or that a paired design wasn’t well suited to this outcome in the first place — in which case the paired t-test’s power advantage over an independent-samples test may not actually be present in your data, even though SPSS will still compute a result.

Convention does not require reporting this correlation in your APA write-up alongside the t-test result itself, but check it before you trust the result, the same way you’d check Levene’s test on the independent-samples output.

Effect size: SPSS doesn’t print Cohen’s d by default

Classic SPSS output for all three t-test procedures does not include a standardized effect size. Compute Cohen’s d from what SPSS does print:

• One-sample / paired-samples: d = Mean Difference ÷ SD of the difference (SPSS prints “Std. Deviation” in the same output block).
• Independent-samples: d = Mean Difference ÷ pooled SD, where the pooled SD is derived from each group’s own descriptive-statistics output (SPSS prints group means and SDs in the “Group Statistics” table above the test table). Use the pooled-SD version only when you’re reporting the “equal variances assumed” row; for Welch’s test, report Cohen’s d computed from the pooled SD but flag that the test itself did not assume equal variances, or report Hedges’ g for the small-sample correction instead.

Newer SPSS Statistics versions (28 and later) can print an effect-size column directly if you check the box in the T-Test dialog’s Options panel — if it’s available in your version, use it and skip the manual calculation, but still confirm which row (equal/unequal variances) it corresponds to.

Reporting the result in APA style

A complete APA-style t-test report states the test type implicitly through the numbers, the direction of the effect in the text, and all four quantities together. Examples:

• Independent-samples: “An independent-samples t-test found that scores were significantly higher in Group A (M = 24.6, SD = 3.1) than Group B (M = 21.2, SD = 3.4), t(58) = 4.02, p < .001, d = 1.04.”
• Paired-samples: “Post-test scores were significantly higher than pre-test scores, t(29) = 3.15, p = .004, d = 0.58.”
• One-sample: “The sample mean (M = 52.3, SD = 8.7) differed significantly from the population benchmark of 50, t(41) = 1.71, p = .095, d = 0.26.”

When you reported the “Equal variances not assumed” row, the non-integer df carries through into the write-up exactly as SPSS printed it (e.g., t(27.43) = 2.11) — do not round it to a whole number.

For the paired-samples test specifically, df = n − 1, where n is the number of pairs, not the total number of observations across both columns — a common transcription error is counting rows in the raw wide-format dataset when a case has missing data on one of the two paired variables, which SPSS excludes listwise before computing df, so the reported df can come in one or more lower than the row count of your data view.

Before you trust the result: check the assumptions

The t-test assumes the outcome variable is roughly normally distributed within each group (or the difference scores are, for a paired test) and, for the independent-samples version, that observations are independent across groups. See CASRAI’s guide to checking the normal-distribution assumption for how to test this in SPSS before running the t-test itself; if normality is badly violated and your sample is small, the nonparametric Wilcoxon signed-rank test is the standard paired-samples alternative.

Frequently asked questions

What does it mean if Levene’s test is significant?

A significant Levene’s test (Sig. ≤ .05) means the two groups’ variances differ by more than chance would predict, so the standard pooled-variance t-test’s assumption is violated. Report the “Equal variances not assumed” (Welch) row instead — SPSS calculates it in the same table regardless of the Levene’s result, so no separate procedure is needed.

Why does SPSS give two different p-values in the same row?

It doesn’t — each row has one t, one df, and one Sig. (2-tailed) value. The two rows correspond to two different versions of the test (equal vs. unequal variances assumed); Levene’s test tells you which single row to report, not which single value within a row.

Can I run a t-test in SPSS if I have more than two groups?

No — the Independent-Samples T Test procedure is built for exactly two groups. Comparing three or more group means requires one-way ANOVA (Analyze > Compare Means > One-Way ANOVA), which extends the same logic to more than two groups without inflating the Type I error rate the way running multiple pairwise t-tests would.

Do I need to report Levene’s test itself in my results?

Convention varies, but reporting it briefly (e.g., “Levene’s test indicated unequal variances, F = 5.12, p = .028, so a Welch-corrected t-test was used”) is good practice when it changes which row you report, since it explains the non-integer degrees of freedom to a reader.

What does a low or negative correlation in the Paired Samples Correlations table mean?

It means the two paired measurements aren’t tracking each other closely within cases, which weakens the statistical-power advantage a paired design is meant to provide over an independent-samples design. It’s a diagnostic check on the pairing, not a reason on its own to discard the t-test result — but it’s worth confirming the variables were paired correctly and that a within-subjects design genuinely fits the outcome being measured.

Related CASRAI resources

For the statistical logic behind the test itself — hypotheses, assumptions, and when to use a t-test versus other comparisons — see CASRAI’s t-test guide. For software background, see what SPSS is and who uses it, or compare it directly against alternatives in SPSS vs. R, SPSS vs. Stata, and jamovi vs. JASP (both free, SPSS-menu-style alternatives). When a z-test is the more appropriate choice than a t-test, see Z-test vs. t-test. To confirm which measurement scale your variable actually is before choosing a test, see levels of measurement. For the paired-samples design’s nonparametric counterpart, see the Wilcoxon signed-rank test guide.