Independent and Paired t Tests Without SPSS: Step by Step
You can run independent-samples, paired-samples and one-sample t tests without SPSS in a free browser tool that uses the same menu, Analyze ▸ Compare Means and Proportions, and prints the same kind of tables. This guide walks through all three in TRTT on a built-in sample survey, explains the two rows of the independent-samples table (the part that confuses most people), and moves on to one-way ANOVA for three or more groups.
All numbers below come from TRTT's built-in demo_omnibus.sav (400 respondents), unweighted. To follow along, open trtt.app in a desktop browser and click Sample Dataset.
Which t test?
| Your question | Test | Example in this guide |
|---|---|---|
| Do two separate groups differ on average? | Independent-samples t test | Men vs women on likelihood to recommend (0–10) |
| Do the same people score differently on two measures? | Paired-samples t test | The same respondents on two agreement items |
| Does one group's mean differ from a fixed value? | One-sample t test | Mean likelihood to recommend vs 7 |
| Do three or more groups differ? | One-way ANOVA | Likelihood to recommend across six age bands |
The outcome should be numeric (scale), and each test assumes roughly normal data within groups or a reasonably large sample. With 30 or more per group, moderate departures from normality rarely matter.
Independent samples, step by step
Question: do men and women differ in how likely they are to recommend the brand (nps, 0–10)?
- Choose Analyze ▸ Compare Means and Proportions ▸ Independent-Samples T Test….
- Select Q4. Likelihood to recommend (0-10) [nps] and click ▶ to move it to Test Variable(s):.
- Move S1. Gender of respondent [gender] to Grouping Variable:.
- Click Define Groups…, keep Use specified values, type 1 for Group 1: and 2 for Group 2: (the codes for Male and Female; check them in Variable View), and click Continue.
- Click OK.
Group Statistics:
| Gender | N | Mean | Std. Deviation | Std. Error Mean |
|---|---|---|---|---|
| Male | 197 | 6.95 | 2.09 | 0.149 |
| Female | 203 | 6.86 | 2.21 | 0.155 |
Independent Samples Test (selected columns):
| Levene's F | Levene's Sig. | t | df | Sig. (2-tailed) | Mean Difference | 95% CI | |
|---|---|---|---|---|---|---|---|
| Equal variances assumed | 2.009 | .157 | 0.452 | 398 | .652 | 0.097 | −0.326 to 0.520 |
| Equal variances not assumed | 0.452 | 397.78 | .651 | 0.097 | −0.325 to 0.520 |
Men score 0.10 points higher on average, but the difference is far from significant (p = .652) and the confidence interval spans zero comfortably.
Levene's test and the two rows of output
SPSS, and TRTT following it, prints two rows because there are two versions of the test:
- Equal variances assumed is the classic Student's t test, which pools the two groups' variances.
- Equal variances not assumed is Welch's t test, which doesn't. Its degrees of freedom are adjusted, which is why you see 397.78 rather than a whole number.
Levene's Test for Equality of Variances helps you choose. If its Sig. is above .05 (here .157), there's no evidence that the variances differ, and the traditional choice is the first row. If it's below .05, use the second row.
Many statisticians now recommend reading the Welch row every time, since it works well whether or not the variances are equal. When the groups are of similar size, as here, the two rows barely differ anyway. Whatever you choose, decide before looking at the p-values, and say in your report which version you used.
Effect size. Since version 27, SPSS can add an effect-size table (Cohen's d and variants); TRTT doesn't print one yet. Cohen's d is easy to compute from the output: the mean difference divided by the pooled standard deviation. Here, 0.097 / 2.15 ≈ 0.05, a negligible effect.
Paired samples
Question: do respondents agree more with "I check the price before I buy anything" (q1) or with "I am willing to pay more for better quality" (q7)? Same people, two items, so the paired test applies.
- Choose Analyze ▸ Compare Means and Proportions ▸ Paired-Samples T Test….
- Click Q1_1… [q1] in the variable list; it fills Variable1:. Click Q1_7… [q7]; it fills Variable2:.
- Click Add ▸. The pair appears in the list on the right (double-click a pair to remove it).
- Click OK.
Results: 370 respondents answered both items (the others said "Don't know" or "Refused", which are declared missing). Means 3.02 (q1) and 3.06 (q7). The Paired Samples Correlations table shows r = .49, and the Paired Samples Test table gives a mean difference of −0.04, t(369) = −0.64, p = .525, 95% CI −0.155 to 0.079. No meaningful difference.
The correlation table matters: a paired test is more powerful than an independent one exactly because the two measurements are correlated.
One sample
Question: is the average likelihood to recommend different from 7?
- Choose Analyze ▸ Compare Means and Proportions ▸ One-Sample T Test….
- Move nps to Test Variable(s):.
- Type 7 in Test Value:.
- Click OK.
Result: mean 6.91 (SD 2.15, N = 400), t(399) = −0.88, p = .377, mean difference −0.095, 95% CI −0.306 to 0.116. The sample mean is consistent with 7.
Going further: one-way ANOVA
With three or more groups, don't run a series of t tests; use one-way ANOVA. Question: does likelihood to recommend differ across the six age bands?
- Choose Analyze ▸ Compare Means and Proportions ▸ One-Way ANOVA….
- Move nps to Dependent List: and S2b. Age band (recoded from S2) [age_band] to Factor:.
- Click Options… and tick Descriptive, Homogeneity of variance test and Welch. Click Continue.
- Click Post Hoc…, tick Tukey under Equal Variances Assumed (or Games-Howell if variances differ), keep Significance level: at 0.05, and click Continue.
- Click OK.
Results: group means range from 6.52 (45–54) to 7.53 (35–44). Levene's test is not significant (based on mean: p = .699). The ANOVA table gives F(5, 394) = 1.98, p = .080, and the Welch test agrees (p = .084). Not significant at .05, and none of the Tukey pairwise comparisons is either: the largest gap, 35–44 versus 45–54 (1.01 points), has p = .095. Eta squared, computed from the sums of squares (45.2 / 1840.4), is about .02.
Syntax
The same four analyses as syntax, for Window ▸ Syntax:
T-TEST GROUPS=gender(1 2) /VARIABLES=nps.
T-TEST PAIRS=q1 WITH q7 (PAIRED).
T-TEST /TESTVAL=7 /VARIABLES=nps.
ONEWAY nps BY age_band
/STATISTICS DESCRIPTIVES HOMOGENEITY WELCH
/POSTHOC=TUKEY ALPHA(0.05).
Click Paste in any dialog to get the command for your own settings, plus equivalent R and Python code in the Paste panel. More in Run SPSS syntax in your browser.
Reporting
APA-style examples for the results above:
- Independent: "Men (M = 6.95, SD = 2.09) and women (M = 6.86, SD = 2.21) did not differ in likelihood to recommend, t(398) = 0.45, p = .652, d = 0.05."
- Paired: "Agreement with the price item (M = 3.02, SD = 1.14) and the quality item (M = 3.06, SD = 1.14) did not differ, t(369) = −0.64, p = .525."
- ANOVA: "Likelihood to recommend did not differ significantly across age bands, F(5, 394) = 1.98, p = .080, η² = .02."
If you used the Welch row, report its decimal df, for example t(397.78) = 0.45.
Open TRTT in your browser – free
Related: Chi-square test without SPSS · Cronbach's alpha without SPSS · Run SPSS syntax in your browser
FAQ
Can I run a t test online for free?
Yes. TRTT runs independent, paired and one-sample t tests in your browser, free, on .sav, CSV or Excel data, with the computation on your own computer. Online t test calculators work too if you only have summary numbers.
Which row do I read in the Independent Samples Test table?
"Equal variances assumed" when Levene's test is not significant, "Equal variances not assumed" (Welch) when it is. Many statisticians recommend the Welch row by default.
Does TRTT report Cohen's d?
Not yet. Compute it from the output: mean difference divided by the pooled standard deviation (independent samples), or mean difference divided by the SD of the differences (paired).
Are the results the same as SPSS?
The menus, options and tables follow SPSS. TRTT is tested against PSPP and published SPSS algorithms, so the t, df and p-values follow the standard formulas. TRTT is an independent program, not SPSS.
What about non-parametric alternatives?
Analyze ▸ Nonparametric Tests ▸ Legacy Dialogs has 2 Independent Samples… (Mann-Whitney U), 2 Related Samples… (Wilcoxon) and K Independent Samples… (Kruskal-Wallis H).
TRTT is an independent product. It is not affiliated with or endorsed by IBM. IBM and SPSS are trademarks of International Business Machines Corporation.