Thursday at ten, a doctoral student sits down across from their chair with chapter 4 printed and tabbed. The finding is on page three. Students in the intervention section scored higher on the end-of-year reading measure, t(94) = 2.41, p = .018. The student has a paragraph ready about what that means for the program. The chair reads the table, then asks something else entirely. "Why an independent samples t-test here? And did you check the variances?"
Nothing about those two questions is hostile. They are the job. A chair does not approve a results chapter, they approve the reasoning that produced it, because the same questions arrive again at the defense, and again from reviewers who will have less patience and no relationship with the student.
The Question Behind the Question
"Why this test?" is asking whether the design chose the test or whether familiarity did. Two independent groups, one continuous outcome, observations that do not depend on each other: the t-test follows from that, and saying so in one sentence closes the question. The trouble starts when the honest answer is that it was the procedure the student had run before.
"Did you check the variances?" works differently. That one asks for evidence that a specific thing was done. Answering "I think it was fine" tells the chair the assumption was never examined, and the conversation stops being about the study. Answering "Levene's test came back F = 0.62, p = .43, so I used the pooled estimate" takes about ten seconds and moves the meeting forward.
Worth being honest about what assumption checks are and are not. They are not a hurdle to clear before the real analysis. They change what gets reported. Had Levene's test been significant, the correct move is Welch's correction, the degrees of freedom stop being a whole number, and page three reads differently. Checking is how you find out which result you actually have.
Where the Answer Should Live
The gap here is almost never knowledge. Most doctoral students can define Levene's test by the time they draft chapter 4. Retrieval is the problem. That analysis ran three weeks ago in a session that produced forty lines of output, four of which went into the chapter. Everything else closed with the window, and reconstructing it on Wednesday night means rerunning the whole thing and hoping the file is the same file.
A Mac helps with this in an unglamorous way. Your work sits in files, on a machine you own, in one folder you can open in eleven months. Dataset, output, and manuscript stay one window apart on the same desk. No session to reconnect to, no account to still have access to, no export that expired. Continuity is a small feature until the week you need something you ran in April.
Assumption Checks That Arrive Without Being Asked
Quanta, our statistics app for the Mac, treats those checks as part of the result rather than as an option buried two menus deep. Run an independent samples t-test and the output arrives with group ns, means, and standard deviations, Levene's test for equality of variances, a normality check on the residuals, the pooled and Welch versions side by side, and Cohen's d printed next to p. Nothing to switch on. The output that answers Thursday's question is the same output that produced the number in the chapter, saved in the same project.
The engine underneath is native Swift on Apple Silicon, verified against NIST reference datasets, and it runs entirely on the machine with no connection required. That matters when the dataset is 96 third graders and the district agreement says where their scores may live.
Quanta does not think for you. It will not decide whether a t-test suits your design, whether a violated assumption calls for Welch's correction or for rethinking how you grouped people, or whether an effect of d = 0.49 is meaningful in a reading intervention. Those judgments are the student's, and a committee is right to test them. What the software removes is the translation between the question and the evidence, so the judgment is what gets discussed.
What Thursday Looks Like Instead
Same meeting, same chair, same chapter. "Why an independent samples t-test?" Because there are two independent groups and one continuous outcome. "Did you check the variances?" Yes, and here is the line. The student turns one page in an appendix that was generated rather than assembled, and the meeting moves to whether a half standard deviation of reading growth justifies keeping the program another year.
That second question is the one worth an hour. Getting to it should not cost the first forty minutes.
Approval comes faster when the record answers first. Defenses go the same way, and so do reviewers, who tend to ask the identical questions eight months later in writing.
Quanta is available for the Mac at relicheck.com and on the Mac App Store, with a free 30-day trial.