Teaching undergraduate research methods: a full-semester toolkit

Why the tooling matters more than instructors admit

When a student runs a t-test and the software prints a table, the format of that table becomes the student's mental model of what a t-test is. That's why we still see thesis defences that report a p-value without an effect size or a confidence interval — because SPSS's default output made it easy to. Choosing the right tool is a pedagogical decision, not just a budget one.

Coverage across a standard syllabus

Every test in a typical undergraduate research-methods or Stats 101 curriculum is a first-class test in thericerca. The full menu:

• Descriptives + frequencies — Table 1 in every dataset.
• One-sample tests — t-test and Wilcoxon signed-rank against a benchmark.
• Two-group comparisons — independent t-test, Welch's t-test, Mann-Whitney U, paired t-test, Wilcoxon signed-rank.
• Three-or-more group comparisons — one-way ANOVA, Welch's ANOVA, Kruskal-Wallis, with Tukey, Games-Howell, and Dunn's post-hocs — the last two only firing when the assumption-appropriate omnibus is significant.
• Factorial designs — two-way ANOVA and ANCOVA.
• Within-subjects designs — repeated-measures ANOVA, Friedman's test.
• Categorical association — chi-square with Cramér's V, Fisher's exact, McNemar for paired designs.
• Correlation — Pearson, Spearman, point-biserial, partial correlation.
• Regression — simple linear, multiple linear, logistic, ordinal, Poisson.
• Reliability — Cronbach's α, McDonald's ω, Cohen's κ, ICC.
• Factor analysis — EFA with KMO, Bartlett's test, and rotated loadings.
• Survival analysis — Kaplan-Meier, log-rank, Cox regression (for the health / social-work majors).
• Mixed-effects models — linear MixedLM with random intercepts.

Every result comes with an APA-formatted write-up, an effect size with a confidence interval, and the assumption checks visible in the report — so a student who sees "Pearson was swapped for Spearman because the normality assumption was not met" learns why the assumption mattered.

Pedagogical wins baked in

• Assumption-driven fallbacks are visible. When ANOVA gives way to Kruskal-Wallis on unequal variance, the report says so explicitly — a teachable moment on every dataset.
• Post-hoc gating is enforced. Tukey / Games-Howell / Dunn's fire only when the omnibus that earned them was significant, so students never accidentally learn the bad habit of over-testing.
• Reproducibility export doubles as an on-ramp to Python. Every completed study can be downloaded as a Jupyter notebook that reproduces the analysis in scipy / statsmodels / lifelines. Students see real code without having to write it — an easy bridge into a follow-on computing course.
• Numeric-claim validator catches hallucinations. The report footer reads "N numbers checked." A student can immediately see the difference between an analysis whose prose traces back to the data and one whose numbers were made up.

A concrete semester

• Week 1–3 — descriptive statistics, distributions, correlations. Free tool: test-picker for the first assignments.
• Week 4–6 — t-tests, chi-square. First guided study on real data.
• Week 7–9 — ANOVA, post-hocs, ANCOVA. Multi-group projects.
• Week 10–12 — regression (linear, logistic), effect sizes and CIs. Cover the reporting standards using the built-in APA output.
• Week 13–14 — scale reliability + EFA. Free reliability tool for take-home practice.
• Week 15 — final project: pick a dataset, run a full study, download the report + reproducibility notebook.

When students should graduate