Laboratory
Statistical analysis with publication-ready results
Upload your data, pick a test — tables and figures are prepared in one action, either for an Uzbek dissertation (OAK) or for an international journal (APA/Scopus). Computation runs in your browser.
No sign-up — one click to a real first result.
28
analyses
20
certified NIST datasets
7.67
correct digits — on the Filip dataset
What makes it different
The analysis is the means; the export is the product
A real table, not a picture of one
Tables arrive in Word as native cells, never as an image. Figures are embedded in the document at 600 dpi. No manual reformatting.
Two standards, one result
Table conventions for an Uzbek dissertation and for an international journal are opposite. You choose which one applies, right before export.
A p-value never travels alone
Every result carries an effect size and its confidence interval. That omission is the single most frequent reason reviewers send a paper back.
Data never leaves the browser
Computation happens entirely on your machine. Clinical or dissertation data is not uploaded to a server.
Two standards
A dissertation table and a journal table are not the same object
Put real OAK dissertation practice next to APA 7 / Scopus requirements and the rules come out opposite. There is no single correct table — the standard is chosen at export time.
Table 1
Group means
| Group | n | M | SD |
|---|---|---|---|
| Experimental | 24 | 4,82 | 0,61 |
| Control | 24 | 4,15 | 0,73 |
Source: compiled from the author's own research.
Table 1
Group means
| Group | n | M | SD |
|---|---|---|---|
| Experimental | 24 | 4.82 | 0.61 |
| Control | 24 | 4.15 | 0.73 |
| Element | OAK (dissertation) | APA 7 / Scopus |
|---|---|---|
| Table rules | Full grid — every cell | No vertical rules, 3 horizontal |
| Table caption | “Table 1” above, right-aligned | Number + italic title above |
| Figure caption | “Fig. 1.” below, bold | Below (above in APA) |
| Decimal mark | Comma — 0,05 | Period — 0.05 |
Available analyses
Every test is checked against reference values
Computation is deterministic. The AI never does the maths — it may only write prose around numbers that are already computed and verified.
Descriptive
Parametric
t-test (one sample)
t, df, p; the CI of the mean difference and Hedges' g.
t-test (independent samples)
Welch by default, with fractional degrees of freedom; Hedges' g and its 95% CI.
t-test (paired samples)
t, df, p; the CI of the mean difference and Hedges' g.
One-way analysis of variance
F, df, p; partial η² with a 90% CI; Tukey, Games–Howell, Scheffé or Dunnett comparisons.
Factorial analysis of variance
Main effects and their interaction; F, df, p; partial η² with a 90% CI.
Analysis of covariance (ANCOVA)
Covariate-adjusted means; F, df, p; partial η² with a 90% CI.
Repeated-measures analysis of variance
Mauchly sphericity test, ε̂ and a Greenhouse-Geisser or Huynh-Feldt correction; fractional df.
Stationarity test (ADF and KPSS)
Checks whether a time series has a unit root or is stationary around a level/trend, using the Augmented Dickey–Fuller (ADF) and KPSS tests; the two tests’ null hypotheses are opposites.
Nonparametric
Mann-Whitney U test
U, Z, p; rank-biserial correlation; median with both IQR bounds.
Wilcoxon signed-rank test
The rank-based counterpart for paired samples: T, Z, p and an effect size.
Kruskal-Wallis test
H, df, p; the ε² effect size; Dunn comparisons.
Friedman test
A nonparametric test for repeated measures; χ², df, p and Kendall W.
Survival analysis (Kaplan–Meier)
Estimates censored time-to-event data with the Kaplan–Meier curve; when groups are assigned, compares them with the log-rank test.
Correlation
Correlation analysis
Pearson, Spearman or Kendall's τ_b; a confidence interval always, significant or not.
Correlation matrix
Pairwise correlations across several variables; n and a CI for every cell.
Principal component analysis
Eigenvalue, share of variance explained and loadings for each component; KMO and Bartlett test.
Categorical
Pearson's χ² test
2×2 and r×c tables; φ or Cramér's V; expected counts shown.
McNemar’s test
Tests marginal homogeneity for paired binary measurements: the exact conditional test, two asymptotic forms, the matched-pairs odds ratio and its exact interval.
ROC analysis (AUC)
Reports the area under the curve with DeLong’s standard error and confidence interval; the operating point at Youden’s J, and every coordinate of the curve.
Regression
Linear regression
Per predictor B, SE, β, t, p and its CI; R², F, VIF and the condition number.
Logistic regression
Per predictor B, SE, Wald, p; OR with a 95% CI; model χ² and pseudo-R².
Poisson regression
B, standard error, Wald z and p for each predictor; RR with a 95% CI; model χ², McFadden R² and the dispersion statistic.
Reliability
Reliability analysis
Cronbach’s α with confidence intervals; no “acceptable” threshold is imposed.
Intraclass correlation (ICC)
Assesses measurement reliability when several raters measure the same objects; all six forms are reported.
Agreement coefficient (Cohen’s and Fleiss’ κ)
Measures chance-corrected agreement when several raters assign the same objects to categories: Cohen’s κ for two raters, Fleiss’ κ for three or more.
Find the test you need
Answer a few questions about your study design and see which analyses fit — no data needed.
Which analysis do you need?
What is your goal?
Accuracy
Grounds for trusting the number that goes into your dissertation
The Filip dataset is a 10th-degree polynomial over 82 observations at a condition number of 1.78·10¹⁵. Several commercial packages have historically returned zero correct digits on it. We use a QR decomposition rather than the normal equations, and we publish the achieved figure for every dataset.
- Dataset
- Filip
- κ
- 1.77·10¹⁵
- Coef.
- 7.67
- Floor
- 7
Questions
Frequently asked questions
Does this replace SPSS or JASP?
For these tasks, yes. The analyses are computed with effect sizes and confidence intervals, and the result comes out directly as a journal-ready table and figure. Factor analysis, survival analysis and panel econometrics are not included yet.
Does OAK require statistical analysis?
No statute says so. But since the impact-factor and Hirsch-index requirements took effect on 1 January 2020, OAK-listed and international journals expect statistical grounding in practice — the results chapter is 50-60% of a dissertation.
Where is my data stored?
Computation runs in your browser and the file is never uploaded. When you close the page the data stays on your device.
How can I be sure the result is accurate?
Our achieved figures on the certified NIST StRD datasets and the reference-library versions are published openly. Every analysis is checked against SciPy and mpmath values.
Which files can I upload?
CSV, TSV and Excel (.xlsx). SPSS (.sav) and Stata (.dta) are not supported yet — export CSV from those programs instead.
Does an AI write the results?
No. Every number is computed by deterministic algorithms. Prose is only ever formed around numbers that are already computed and verified, and it never makes a causal claim.
Start with the sample data
No file to prepare — open the sample dataset and see a real first result, and how it looks on export, straight away.
Start analysis