Two groups, one numeric outcome, and the question of whether the gap between them is real. What the test answers, when it fits, how it runs in Quanta on the Mac, what each result tab gives you, and how to report it in APA 7.
An independent-samples t test compares the mean of one outcome across two groups. Treatment and control. Two age bands. Respondents who completed a program and respondents who did not. Each person appears in exactly one group, which is what makes the samples independent.
The test asks one question: is the distance between these two averages larger than the everyday scatter within each group would lead you to expect. The statistic it returns, t, is that distance divided by the uncertainty around it. The p value converts it into the probability of seeing a gap at least this large if the two populations really had the same mean.
If your two measurements come from the same people, this is the wrong test. Use the paired-samples t test. For three or more groups, use the one-way ANOVA, which asks the question once rather than running several t tests and inflating your false-alarm rate. If you need to adjust for a covariate, use ANCOVA. All four sit together in Compare Means.
The classic Student t test assumes the two groups have similar spread. When they do not, the Welch version is the correct choice. Most packages leave that to a checkbox you have to know to tick.
Quanta checks the variance ratio and selects the appropriate version, then puts the decision on the result where a reader can see it. The card carrying your result is badged with the test that actually ran, so Welch t appears on the output rather than living in your memory of what you clicked. MacStats tested this in July 2026 and described exactly that behavior: the software worked out that the standard deviations differed between the two groups and used Welch instead of Student.
This is a recommendation carried out, not a restriction. Every test stays available if your design calls for a different one.
Bring in the data. Import from CSV, Excel, SPSS .sav, Stata .dta, or a ReliCheck survey package. Quanta reports the row count, the column count, and the missing-cell count as the file lands, so a bad import shows itself before it becomes a bad result.
Check the shape. The test wants long format: one row per person, one column holding the group label, one column holding the numeric outcome. If your two groups sit in two separate columns, reshape before you run.
Open Compare Means and choose t-Test. Compare Means holds four analyses on one segmented control: t-Test, Paired t-Test, One-way ANOVA, and ANCOVA. Picking one changes what the Setup panel asks for.
Configure the variables in the Setup panel on the right. Quanta is organized design-first: you set the outcome and the grouping variable, and the analysis runs from there. Results appear in the center.
Look before you test. Descriptives and Explore gives you the mean, median, skewness, histogram, and box plot for the outcome. Two minutes here catches the coding errors and stray values that no assumption check will flag for you.
The result arrives on four tabs: Test result, Effect size, Diagnostics, and Reporting. Every tab carries the same bottom bar, so you can add to report, copy the table, or copy the APA text from wherever you are standing.
An APA-shaped table headed with your variables, badged with the test that ran. Group, n, M, SD, t, df, and p, with a note stating that M and SD are the group mean and standard deviation and that p is two-tailed.
Underneath it, the part most packages do not give you by default: a Bayes factor beside the frequentist result, with the prior stated. A JZS Cauchy prior on the standardized effect, r scale 0.71, and a plain sentence naming what the number means.
This matters most when the result is null. A p of .705 tells you the difference was not statistically significant. It cannot tell you whether the data support the absence of a difference. BF10 = 0.18 can, and Quanta labels it: moderate evidence for no difference. One of those is an absence of a finding. The other is a finding.
Cohen's d as a headline number with a magnitude word attached, so negligible, small, or large sits next to the figure rather than being left for you to look up. A small p tells you an effect is detectable. Only the effect size tells you whether it is worth caring about, and a d of −0.06 labelled negligible closes the question faster than a p value ever will.
The assumptions, checked and stated in sentences rather than left as bare test statistics. The variance ratio with a verdict on whether the spreads are reasonably similar, and the group sizes with a judgment on whether they are adequate for the central limit theorem to do its work.
That second one is worth noticing. At n = 91 and n = 71, the honest answer about normality is that the sample sizes carry you, and Quanta says so rather than running a normality test whose p value would be misleading at that size.
The finding written out in APA style, ready to lift, alongside an explicit guardrail. A group difference is association, not causation. Report the effect size and CI alongside the p value. The caution is written into the summary you are about to paste, which is the opposite of how most statistical software behaves.
Every result carries a ReliCheck Intelligence card with one action: explain this result in plain language. It runs entirely on your Mac with Apple Intelligence. No upload, no API key, nothing to configure. The model only explains the result Quanta has already computed and validated. It never computes a statistic, and it never applies a decision on your behalf.
The boundary is deliberate. The number comes from a validated engine. The explanation comes from a model that is not allowed near the arithmetic.
Copy APA text puts the result line on your clipboard. Add to report sends the table into a Quanta report, which exports with the APA table, its note, and a per-analysis caution beneath it.
A real result line from Quanta, unedited:
An independent-samples Welch t-test compared Job Performance Score (0-100) between 25-34 (M = 70.36, SD = 9.75) and 35-44 (M = 70.93, SD = 9.20). The difference was not statistically significant, t(154.3) = −0.38, p = .705, d = −0.06.
Name the Welch correction when you use it. Reviewers read the fractional degrees of freedom as the signal that unequal variances were handled, and stating it plainly removes the question.
When the result is null, report the Bayes factor with it. A sentence that says the difference was not significant and the Bayes factor indicated moderate evidence for no difference is a conclusion. A sentence that stops at not significant is a shrug.
Quanta's t tests, ANOVA, and ANCOVA are validated against R's t.test, aov, and car, matched exactly. The engine is pure Swift with no R or Python at runtime. Full record on the validation page.
Compare Means is part of every Quanta subscription, $9.99 a month or $79.99 a year after the 30-day trial. One price for everyone. See pricing.
The analysis, the report builder, and the on-device explanations all work with the network off. Your dataset is never uploaded.
Paired-samples t test for two measurements on the same people · One-way ANOVA for three or more groups · ANCOVA to adjust for a covariate · Mann-Whitney U when the outcome is ordinal or skewed · the complete list on the analyses page.