DEV Community: Ahnhyeongkyu

3 Comments Your Thesis Committee Will Make About Your Stats Section (Fix Them Before They Do)

Ahnhyeongkyu — Tue, 01 Sep 2026 07:56:46 +0000

If your advisor or a reviewer has ever handed back your results section with red ink, it's usually one of three things — and none of them are about whether your test was 'right.'

1. You reported a p-value with no effect size.
A significant t-test or ANOVA tells a committee something happened, not how much. APA 7th edition expects Cohen's d, eta-squared, or an equivalent alongside every test statistic. If your write-up reads "t(48) = 2.31, p = .025" and stops there, that's the first thing a reviewer circles. Add the effect size and a one-line plain-language read of its magnitude (small/medium/large per convention for that test).

2. You ran multiple comparisons without correcting for them.
Running five t-tests instead of one ANOVA, or testing every pair after a significant omnibus test, inflates your false-positive rate. If you didn't apply a Bonferroni, Tukey HSD, or Holm correction (and say so explicitly in the write-up), expect a comment asking why not. This is one of the most common flags on multi-group comparisons and post-hoc analyses.

3. You didn't check — or didn't report checking — your test's assumptions.
Normality for parametric tests, homogeneity of variance for ANOVA/t-tests, linearity and independence for regression. Committees don't expect perfection; they expect a sentence: "Levene's test indicated equal variances (p = .41), so a standard independent t-test was used." Silence on assumptions reads as "didn't check," even if you did.

The fix for all three is the same: build the effect size, correction, and assumption check into your workflow before you write the results section, not after committee feedback forces a rewrite. Free calculators (t-test, ANOVA, regression, factor analysis, etc.) that surface effect sizes and assumption checks alongside the test statistic, and format the whole thing as an APA-ready sentence, save the second pass entirely — https://statmate.org

5 Reasons Reviewers Send Back Your Stats Section (and How to Fix Each One)

Ahnhyeongkyu — Sat, 29 Aug 2026 07:36:29 +0000

Reviewer comment #2 always looks the same: "Please justify your choice of statistical test and report effect sizes / assumption checks." Here are the 5 most common reasons a stats section gets bounced back, and the one-line fix for each:

No effect size reported — a p-value alone doesn't tell a reader how big the effect is. Add Cohen's d (t-tests), eta-squared (ANOVA), or r (correlation) next to every significance test.
No assumption check — reviewers now expect a one-sentence note on normality/homogeneity even when the test is robust to violations. State it, don't skip it.
Multiple comparisons left uncorrected — running 5 pairwise t-tests without Bonferroni/Tukey inflates your Type I error and reviewers know it.
Wrong test for the data structure — using independent t-test on paired/repeated data (or vice versa) is the single most common silent error in submitted manuscripts.
Results reported inconsistently with APA 7 — df, exact p, and test statistic formatting mismatches across the same paper signal a rushed writeup.

Each of these takes under a minute to check once you have your test statistic and sample size — I built free calculators for all 20 common tests (with the effect size and assumption check included in the output) at statmate.org if you want to double check before you resubmit.

How to Report a Mann-Whitney U Test in APA 7th Edition (With Effect Size)

Ahnhyeongkyu — Sat, 22 Aug 2026 07:04:34 +0000

When your data fails the normality assumption, Mann-Whitney U replaces the independent t-test — but its APA write-up trips people up because you report U and z, not t.

The template:
A Mann-Whitney U test indicated that [DV] was [higher/lower] for Group A than Group B, U = [value], z = [value], p = [value], r = [value].

What each piece means:

Report medians, not means — the test ranks data, so means misrepresent what was actually compared.
U is the test statistic itself; most software also gives you a normal approximation z once n is large enough, which is what you report alongside U.
Effect size (r): divide z by the square root of total N (r = z / √N). APA 7 requires an effect size for every inferential test — this is the one reviewers ask for when it's missing.
Common mistake: reporting only p without U/z/r. A p-value alone tells a reader nothing about magnitude.

Worked example:
A Mann-Whitney U test indicated that anxiety scores were higher for the treatment group (Mdn = 14) than the control group (Mdn = 9), U = 210, z = 2.87, p = .004, r = .41.

If you're running this test on your own data, statmate.org's Mann-Whitney calculator outputs U, z, p, and r together with this exact sentence pre-filled from your numbers, so you can skip the manual formula work.

Choosing the Right Statistical Test: A Practical Decision Guide for Grad Students

Ahnhyeongkyu — Fri, 21 Aug 2026 07:05:58 +0000

Picking the wrong statistical test is one of the most common reasons a thesis committee sends an analysis chapter back for revisions. Here's a quick decision framework.

Step 1 — What's your outcome variable?

Continuous (scores, measurements) → Step 2a
Categorical (yes/no, groups) → Step 2b

Step 2a — Comparing continuous outcomes

2 independent groups, normal distribution → independent-samples t-test
2 independent groups, non-normal/ordinal → Mann-Whitney U
3+ independent groups, normal → one-way ANOVA (+ Tukey HSD post-hoc)
3+ independent groups, non-normal → Kruskal-Wallis (+ Dunn's post-hoc)
Repeated measures, 2 timepoints → paired t-test
Repeated measures, 3+ timepoints → repeated-measures ANOVA or Friedman test
Relationship between two continuous variables → Pearson (linear, normal) or Spearman (monotonic, non-normal)

Step 2b — Categorical outcomes

2 categorical variables, independence → chi-square test of independence (check expected cell counts ≥5, otherwise Fisher's exact)
Predicting a binary outcome from multiple predictors → logistic regression

Step 3 — Check assumptions before trusting the result
Normality (Shapiro-Wilk), homogeneity of variance (Levene's), and for repeated designs, sphericity (Mauchly's) should be checked, not assumed. If normality fails on a 2-group comparison, that's the cue to switch from t-test to Mann-Whitney.

Step 4 — Report in APA format
APA 7th edition wants the test statistic, degrees of freedom, exact p-value (or p < .001), and an effect size. Example: t(48) = 2.14, p = .038, d = 0.61.

Getting the test right is half the work; writing it up the way your committee expects is the other half. If you'd rather skip the manual assumption-checking and APA formatting, statmate.org runs the test selection and formats the write-up for you — but the framework above works fine by hand, in R, or in SPSS too.

I ran 5 of StatMate's stats calculators against Python's scipy.stats — raw data, code, and real output, not a marketing table

Ahnhyeongkyu — Fri, 21 Aug 2026 07:05:57 +0000

I run StatMate, a set of 20 free browser-based statistics calculators (t-tests, ANOVA, chi-square, correlation, non-parametric tests). Instead of asking you to trust a marketing page, here are 5 of the calculators run directly from the actual source code, cross-checked against Python's scipy.stats (v1.13.1) with the same raw data — copy the code blocks below and run them yourself.

StatMate's calculators use jStat v1.9.6 + simple-statistics v7.8.8 (see package.json). I ran the real TypeScript functions from lib/statistics/ with npx tsx, then ran the identical arrays through scipy.stats.

1. Independent t-test (Welch, n1=15, n2=15)

g1 = [23,25,21,28,30,22,26,24,29,27,20,31,25,23,28]
g2 = [19,22,18,24,20,21,17,23,25,19,22,20,18,24,21]

StatMate (lib/statistics/t-test.ts): t(25.68) = 4.31, p < .001
scipy stats.ttest_ind(g1, g2, equal_var=False): t=4.3077, df=25.68, p=.000213

2. One-way ANOVA (3 groups, n=10 each)
StatMate (lib/statistics/anova.ts): F(2, 27) = 24.19, p < .001
scipy stats.f_oneway(*groups): F=24.1870, p=9.57e-7

3. Pearson correlation (n=30)
StatMate (lib/statistics/correlation.ts): r = .996, p < .001
scipy stats.pearsonr(x, y): r=.9964, p=1.57e-31

4. Chi-square test of independence (2x3 table)

table = [[15,10,5],[8,12,20]]

StatMate (lib/statistics/chi-square.ts): χ²(2) = 10.09, p = .0064
scipy stats.chi2_contingency(table): χ²=10.0896, p=.00644

5. Mann-Whitney U
StatMate (lib/statistics/mann-whitney.ts): U = 0, p < .001
scipy stats.mannwhitneyu(mw1, mw2): U=0.0, p=3.66e-5

All 5 matched to floating-point precision between two independent implementations (StatMate's own jStat/simple-statistics code and scipy). No R involved — I don't have R in this environment, so I'm not claiming an R comparison. If anyone wants to add an R cross-check, the raw arrays above are enough to reproduce it in t.test()/aov()/etc. — I'd genuinely like to see that comparison and will update if someone runs it.

Calculators used: https://statmate.org (free, no signup for calculations).

undefined

Ahnhyeongkyu — Thu, 20 Aug 2026 07:43:58 +0000

Reporting a Kruskal-Wallis H Test in APA 7 Style

The Kruskal-Wallis H test is the non-parametric alternative to a one-way ANOVA, used when your dependent variable doesn't meet normality assumptions or is ordinal. Here's how to report it correctly in APA 7 format.

The basic format

H(df) = [statistic], p = [value], followed by an effect size (epsilon-squared or eta-squared-H), and post-hoc detail if the result is significant.

Example:

A Kruskal-Wallis H test showed a statistically significant difference in [DV] across the three groups, H(2) = 8.42, p = .015, ε² = .11. Dunn's post-hoc tests with Holm-Bonferroni correction indicated group A scored significantly higher than group C (p = .012), but no other pairwise differences were significant.

Three things people consistently get wrong

Naming the post-hoc correction. Don't just write "p < .05" — state the correction method explicitly. Bonferroni is conservative; Holm-Bonferroni is generally preferred and worth naming.
Skipping the effect size. H and p alone aren't sufficient for APA 7 — report epsilon-squared (ε²) or eta-squared-H alongside them.
Miscounting df. df = k − 1 (number of groups minus 1), not sample size minus 1.

Why this matters beyond style compliance

Reviewers and thesis committees increasingly check for correction-method transparency and effect sizes as a proxy for statistical rigor — omitting them is a common reason for "minor revisions" requests.

Disclosure: I built statmate.org, a tool that runs the test and generates this exact APA-formatted writeup automatically. Sharing the manual reasoning here regardless, since knowing the rule matters more than the tool.

How to choose the right statistical test for your thesis (a decision-tree walkthrough)

Ahnhyeongkyu — Wed, 19 Aug 2026 07:36:01 +0000

Picking a test is where most thesis stats panic starts — not APA formatting. This is a plain decision path for the most common designs in social science/health/education research:

Comparing two groups? → independent samples → independent t-test (normal) or Mann-Whitney U (non-normal/ordinal). Same subjects measured twice → paired t-test or Wilcoxon signed-rank.
Comparing 3+ groups? → one-way ANOVA (normal, equal variance) or Kruskal-Wallis H (non-normal). Two categorical factors at once → two-way ANOVA.
Relationship between two continuous variables? → correlation (Pearson if linear+normal, Spearman if not) or simple regression if you need prediction, not just association.
Categorical vs categorical? → chi-square test of independence (or Fisher's exact if any expected cell count < 5).
Scale reliability before you even run the above? → Cronbach's alpha first — a test result built on an unreliable scale isn't worth reporting.

The underlying question is always: what's the outcome variable's type (continuous/ordinal/categorical), how many groups, and are your normality/variance assumptions met? Get those three answers and the test picks itself.

If you want to walk through this interactively with your own data, StatMate has a free decision-tree wizard (statmate.org/wizard) that asks these same questions and routes you to the right calculator — no signup, and it runs 20 tests including the ones above.

The APA 7 Statistics Reporting Cheat Sheet: Exact Syntax for 9 Common Tests

Ahnhyeongkyu — Wed, 12 Aug 2026 07:41:34 +0000

Grad students lose more time figuring out how to report a result than running it. APA 7 has exact formatting rules for every test, scattered across a 400-page manual. Here's the syntax I wish I'd had, for the tests most theses actually use.

Independent-samples t-test
t(df) = value, p = .xxx, d = value
Example: t(58) = 2.45, p = .018, d = 0.63.

One-way ANOVA
F(df_between, df_within) = value, p = .xxx, η² = value
Example: F(2, 87) = 4.12, p = .020, η² = 0.09.

Chi-square test of independence
χ²(df, N = n) = value, p = .xxx
Example: χ²(1, N = 200) = 6.73, p = .009.

Pearson correlation
r(df) = value, p = .xxx
Example: r(48) = .42, p = .002.

Simple linear regression
Coefficient: b = value, SE = value, t(df) = value, p = .xxx. Model fit: R² = value, F(df1, df2) = value, p = .xxx.

Mann-Whitney U (non-parametric t-test alternative)
U = value, z = value, p = .xxx — report medians (Mdn), not means.

Wilcoxon signed-rank test
z = value, p = .xxx, r = value — effect size is r, not d.

Cronbach's alpha (reliability)
α = value, reported per scale, not per item. α ≥ .70 is the conventional (debated) cutoff.

Logistic regression
b = value, SE = value, Wald χ²(df) = value, p = .xxx, OR = value — reviewers will ask for the odds ratio if it's missing.

Three things that trip people up every time: p-values drop the leading zero (p = .018, never p = 0.018); report exact p over p < .05 unless p is very small (p < .001); and effect sizes are not optional in APA 7 — a missing effect size is one of the most common methods-section desk-reject notes.

If you'd rather not hand-format these, I built statmate.org — it runs the test and outputs the line above automatically for 20 test types. But the syntax above works on its own, no tool required.

The 5-Minute Pre-Flight Checklist Before You Run Any Statistical Test

Ahnhyeongkyu — Mon, 10 Aug 2026 07:33:51 +0000

Every semester I watch grad students run a t-test or ANOVA the moment their data is entered — then get flagged in peer review for skipping assumption checks. Here's the checklist I actually use before touching any parametric test, condensed from what committees ask about:

Sample size vs. group count. Below roughly 15-20 per cell, parametric tests get shaky — consider the nonparametric equivalent (Mann-Whitney instead of independent t, Kruskal-Wallis instead of one-way ANOVA) before you finish collecting the rest of your sample.
Normality. Run Shapiro-Wilk on residuals, not raw scores, for regression/ANOVA. For n > 50, eyeball a Q-Q plot instead — Shapiro-Wilk gets oversensitive to trivial deviations at large n.
Homogeneity of variance. Levene's test for t-tests/ANOVA. If it fails, don't switch tests — use Welch's correction instead.
Independence. The one assumption no software checks for you. Repeated measurements on the same subject, or nested data (students within classrooms), violate this even when everything else passes — you need repeated-measures or mixed models instead.
Effect size, decided before you see the p-value. Committees ask for Cohen's d / eta-squared / r regardless of significance. Compute it as part of your plan, not as an afterthought when p = .06.

None of this needs paid software — R's shapiro.test() and leveneTest() (car package) cover 1-4 for free. If you'd rather have the checks run automatically alongside the test and the write-up formatted in APA style, StatMate's calculators (statmate.org) do that step for you — but the checklist above holds regardless of what tool you use.

How to Report a Two-Way ANOVA in APA 7th Edition (With Templates)

Ahnhyeongkyu — Thu, 06 Aug 2026 07:42:54 +0000

A two-way ANOVA gives you three separate results to report, not one: the main effect of Factor A, the main effect of Factor B, and their interaction. Most write-ups miss the interaction term or report it after — instead of before — checking whether it changes how the main effects should be interpreted.

1. Check the interaction first.
If the interaction is significant, the main effects can be misleading on their own (the effect of Factor A depends on the level of Factor B), so report the interaction before either main effect and consider simple-effects follow-up tests rather than trusting the marginal means.

2. APA 7 template for each term:
"A two-way ANOVA was conducted to examine the effects of [Factor A] and [Factor B] on [DV]. There was a statistically significant interaction between [Factor A] and [Factor B], F(df1, df2) = X.XX, p = .0XX, η²p = .XX."

If the interaction is not significant, drop it and report main effects instead:
"There was a statistically significant main effect of [Factor A], F(df1, df2) = X.XX, p = .0XX, η²p = .XX. There was no significant main effect of [Factor B], F(df1, df2) = X.XX, p = .0XX, η²p = .XX."

3. Degrees of freedom. df1 = (levels of factor − 1) for that term (or the product of both, for the interaction); df2 = total N − total number of cells. Get both from your ANOVA table, not just the F-value.

4. Effect size. Partial eta-squared (η²p) is standard for factorial ANOVA. Rough conventions: .01 small, .06 medium, .14 large (Cohen, 1988) — report it, don't just call the result 'significant.'

5. If you need this fast: I built a free two-way ANOVA calculator that runs the test and formats the APA sentence + table for you — statmate.org/calculators/two-way-anova. No signup needed to use it.

How to Report a One-Way ANOVA in APA 7th Edition (with Worked Example)

Ahnhyeongkyu — Sat, 01 Aug 2026 07:55:14 +0000

A one-way ANOVA answers whether means differ across 3+ independent groups. Here's the APA 7th edition write-up format, copy-paste ready.

Overall model:

A one-way ANOVA was conducted to compare [DV] across [k] groups. There was a statistically significant effect, F(df_between, df_within) = X.XX, p = .XXX, η² = .XX.

Post-hoc (if significant):

Tukey's HSD post-hoc comparisons indicated that Group A scored significantly higher than Group B, p = .XXX.

Non-significant:

There was no statistically significant effect of [IV] on [DV], F(df1, df2) = X.XX, p = .XXX.

Rules worth remembering: report eta-squared (η²) or partial eta-squared as the effect size, not just p; always report both between- and within-group df; run post-hoc tests only when the omnibus F is significant, and correct for multiple comparisons (Tukey, Bonferroni, or Games-Howell if variances are unequal).

If you'd rather generate these numbers from your own data than compute them by hand, the free StatMate ANOVA calculator runs the omnibus test, post-hoc comparisons, and effect size, and formats the output in this exact style.

Choosing the Right Statistical Test: A Practical Decision Guide (t-test, ANOVA, Chi-square, Mann-Whitney, Kruskal-Wallis)

Ahnhyeongkyu — Wed, 29 Jul 2026 07:20:32 +0000

Picking the wrong statistical test is one of the most common mistakes in thesis and research writing. Here's a quick decision framework:

1) What's your outcome variable? Continuous → parametric family (t-test/ANOVA) if roughly normal, otherwise rank-based (Mann-Whitney/Kruskal-Wallis). Categorical → Chi-square or Fisher's exact (use Fisher's when any expected cell count < 5).

2) How many groups? 2 groups → t-test (independent or paired) or Mann-Whitney. 3+ groups → ANOVA or Kruskal-Wallis, followed by a post-hoc test if the omnibus result is significant.

3) Repeated measures on the same subjects? Use paired/repeated-measures variants (paired t-test, repeated-measures ANOVA, Friedman) instead of independent-sample versions — treating repeated measures as independent samples is a common validity error reviewers flag.

4) Report effect size, not just p-values. APA style expects Cohen's d, eta-squared, or r alongside the test statistic — a p-value alone doesn't tell a reader how large the effect is.

If you want to run these without setting up R/SPSS, statmate.org has free versions of all of the above (calculator-only, no signup for the calculation itself).