A t-test is a statistical test that tells you whether the average of one group is meaningfully different from another average, or from a known value — used when you're comparing one or two groups on a single numeric outcome. If you've collected scores from two conditions and you're staring at a spreadsheet wondering whether the difference you see is real or just random noise, the t-test is almost certainly the tool you need.
Key Takeaways
- A t-test compares means (averages) to decide whether a difference between groups is statistically significant or likely due to chance.
- There are three types: one-sample (one group vs. a fixed value), independent-samples (two separate groups), and paired-samples (the same group measured twice).
- Use a t-test when your outcome is numeric (continuous) and you're comparing one or two groups — never three or more (that's ANOVA).
- A p-value below .05 conventionally means the difference is statistically significant, but you should always report an effect size (Cohen's d) alongside it.
- If your data are badly skewed or ordinal, a Mann-Whitney U or Wilcoxon test may fit better than a t-test.
What Does a T-Test Actually Do?
A t-test asks a simple question: are these averages far enough apart that I can rule out random luck as the explanation? Imagine two groups scored differently on an anxiety scale. The difference might reflect a genuine effect — or it might just be that the people in each group happened to vary. The t-test weighs the size of the gap between the means against how much the scores bounce around within each group. A big gap with tight, consistent scores is convincing; a small gap with wildly scattered scores is not.
The output is a t statistic and a p-value. The t statistic is essentially a signal-to-noise ratio — the bigger it is, the stronger the evidence of a real difference. The p-value translates that into a probability: how likely you'd see a difference this large if there were truly no difference at all.
When Should I Use a T-Test?
Use a t-test when all of these are true:
- Your outcome variable is continuous (test scores, reaction times, blood pressure, income).
- You're comparing one or two groups — not three or more.
- Your data are roughly normally distributed (a bell-ish shape, no extreme skew).
If you have three or more groups, you need a one-way ANOVA instead. If your outcome is a category (passed/failed, yes/no), you need a chi-square test. And if your data are heavily skewed or on a ranked/ordinal scale, a non-parametric alternative is the safer choice.
The Three Types of T-Test (and How to Tell Them Apart)
Choosing the wrong type is the single most common t-test mistake we see from thesis students. Here's the plain breakdown.
One-Sample T-Test
Compares your group's average against a known or hypothesised value. Example: testing whether the mean IQ in your sample differs from the population value of 100.
Independent-Samples T-Test
Compares the averages of two separate, unrelated groups. Example: comparing exam scores between students who used Method A vs. students who used Method B. Each person appears in only one group.
Paired-Samples T-Test
Compares two measurements from the same people (or matched pairs). Example: measuring blood pressure before and after a treatment in the same patients. The data are linked, which is why it's called "paired."
| T-Test Type | What You're Comparing | Example |
|---|---|---|
| One-sample | One group vs. a fixed number | Sample mean vs. population norm of 100 |
| Independent-samples | Two unrelated groups | Treatment group vs. control group |
| Paired-samples | Same group, two time points | Pre-test vs. post-test scores |
A Worked Example With Real Numbers
Say you ran a study with 30 participants split into two groups to test whether a mindfulness app reduces stress. The app group (n = 15) scored a mean stress of M = 42.3 (SD = 8.1), and the control group (n = 15) scored M = 49.7 (SD = 9.4). Because these are two separate groups, you'd run an independent-samples t-test.
The analysis returns:
t(28) = 2.31, p = .028, d = 0.84
Here's what each piece means:
- t(28) — the t statistic is 2.31, and 28 is the degrees of freedom (roughly your total sample size minus the number of groups: 30 − 2 = 28).
- p = .028 — there's about a 2.8% probability of seeing a difference this large if the app had no real effect. Because .028 is below the .05 threshold, the result is statistically significant.
- d = 0.84 — Cohen's d, the effect size. A value of 0.84 is a large effect (0.2 = small, 0.5 = medium, 0.8 = large), so the difference isn't just real, it's practically meaningful.
The takeaway: the app group reported meaningfully lower stress than controls, and we can be reasonably confident it wasn't a fluke.
How Do I Report a T-Test in APA 7 Format?
APA 7 wants the test statistic, degrees of freedom, exact p-value, and an effect size. For the example above:
An independent-samples t-test showed that the app group reported significantly lower stress (M = 42.3, SD = 8.1) than the control group (M = 49.7, SD = 9.4), t(28) = 2.31, p = .028, d = 0.84, 95% CI [0.86, 13.94].
Note the APA details: t, M, SD, p, and d are all italicised, and the p-value drops the leading zero (.028, not 0.028). Always include the means and standard deviations for each group, plus a confidence interval for the mean difference where possible.
What Counts as "Significant"? And What to Watch For
A p-value below .05 is the conventional cutoff for significance, meaning you'd expect a result this extreme less than 5% of the time by chance alone. But "significant" is not the same as "important" — with a very large sample, a trivial difference can hit p < .05. That's why reporting Cohen's d matters: it tells you the size of the effect, independent of your sample size.
Two assumptions to check before trusting your t-test:
- Normality: the outcome should be roughly bell-shaped in each group. With small samples and heavy skew, consider a Mann-Whitney U test instead — if you're weighing that decision, see our guide on Mann-Whitney vs. the t-test.
- Equal variances: the independent-samples t-test assumes both groups have similar spread. When they don't, use Welch's t-test (most software, including StatRyx, applies the correction automatically).
Can I Run a T-Test Without Learning Statistics Software?
Yes — and this is where most researchers lose hours. SPSS can run every t-test, but a single-user licence costs roughly $99+ per month and the menu-driven workflow is unforgiving if you pick the wrong dialog box. R and jamovi are free and powerful, but R demands code and jamovi is a desktop install you have to maintain.
StatRyx is an AI-powered statistical analysis tool that replaces those manual SPSS workflows. You upload your data, describe your question in plain English, and StatRyx picks the correct t-test variant, checks the normality and variance assumptions for you, runs the analysis, and hands back a fully APA 7-formatted write-up — the sentence you can paste straight into your thesis. For non-statisticians who just need the right answer reported correctly, it removes the two biggest failure points: choosing the wrong test and mis-formatting the output.
**Stop calculating this by h