Data Lab brings AI data analysis for research into one browser tab. Load a CSV, describe and chart it, run a statistical test and get each result as a table with an APA sentence. The numbers are calculated in your browser; the AI only writes up what you computed. It is for thesis, survey and lab data.
What you can do with it
Data Lab has six tabs: Data, Describe and chart, Tests, Which test?, Sample size and Report.
- Data: upload a CSV, TSV or Excel file, paste rows from Excel, Google Sheets or SPSS, or press "Use sample data". Each column is read as numeric or categorical, and you can switch it.
- Describe and chart: for each numeric variable, n, missing values, mean, SD, median and range; for each categorical variable, counts and percentages. Charts: histogram, bar chart, box plot by group, and scatter plot with a fitted line. Export the descriptives as CSV.
- Tests: pick a test and two variables, press Run. Results stack up, can be removed, saved to History or exported as CSV.
- Which test?: answer up to five questions and get a recommended test, why, the assumptions to check first, an APA reporting template and the alternative if assumptions fail.
- Sample size: for a t-test (two independent groups, paired, or one sample), two proportions, a prevalence survey (with finite population correction and design effect), a correlation, or diagnostic accuracy. Or enter the sample you have and see its power.
- Report: an AI-written results section from your results (2 credits), and Download PDF report (no credits) with the dataset summary, each analysis's table and chart, and the results section if you wrote one. The PDF is saved from your browser's print dialog.
Exactly which tests it runs
| Group | Tests |
|---|---|
| Compare groups | Independent t-test (Welch), Mann-Whitney U, paired t-test, Wilcoxon signed-rank, one-way ANOVA, Kruskal-Wallis, McNemar (paired yes/no) |
| Relationships | Pearson and Spearman correlation, chi-square, Fisher's exact test (2 × 2, with odds ratio) |
| Prediction | Linear regression (one predictor), logistic regression (up to 10 numeric or categorical predictors, odds ratios with CIs) |
| Clinical and diagnostic | Kaplan-Meier with the log-rank test, diagnostic accuracy (sensitivity, specificity, PPV, NPV, likelihood ratios, exact CIs), ROC curve with AUC and the Youden cut-off |
| Bayesian | Bayesian independent, paired and one-sample t-tests (JZS Bayes factor BF10 and BF01 with a choice of prior scale, an evidence label, and the effect size with its 95% credible interval), Bayesian proportion (beta-binomial, posterior mean and 95% credible interval) |
| Scales and surveys | Cronbach's alpha, with item-total correlations and alpha if each item is deleted |
Every result has an APA 7 sentence you can copy, and the warnings that apply to it (small groups, sparse tables, separation in logistic regression). The numbers are calculated, not generated, and were checked against scipy, R and published examples. Each saved analysis can be exported as R, Python or SPSS code, so an examiner can rerun it. The Bayesian tests export to R (BayesFactor) and Python (pingouin); SPSS gets a note pointing to R or JASP, because SPSS uses different priors.
Data Lab does not run formal normality tests, Levene's test, Welch's ANOVA, post-hoc comparisons, Friedman, repeated-measures ANOVA, multiple linear regression, mixed models or Fleiss' kappa. When the chooser recommends one of these, it says so and points you to SPSS, R, jamovi or JASP.
How it works, step by step
- Open Data Lab and upload your file or paste your data with a header row.
- Check the column types in the preview and switch any that are wrong, such as a group coded 1 and 2 that should be categorical.
- On Describe and chart, read the summaries and look at a histogram or box plot of your outcome. Skew and outliers show here.
- If you are unsure which test fits, open Which test? and answer the questions.
- On Tests, choose the test and the two variables and press Run. Repeat for each analysis.
- Press Save to History to keep the results, or Results as CSV to export them.
- On Report, add your research question if you like and press Write the results section (2 credits). Open it in the editor or copy it.
A worked example
Example: the built-in sample data has 12 students, six with tutoring and six without, and their exam scores. Run "Independent t-test (Welch)" with score as the outcome and group as the group. Data Lab returns:
- tutoring: M = 78.67, SD = 8.71, n = 6; control: M = 66.17, SD = 6.31, n = 6
- t(9.11) = 2.85, p = .019, Cohen's d = 1.64
- A warning: "Small groups: check that the values are roughly normal."
The APA sentence reads: "score was significantly different between tutoring (M = 78.67, SD = 8.71) and control (M = 66.17, SD = 6.31), t(9.11) = 2.85, p = .019, d = 1.64." With six per group, you would look at the box plot before trusting it, and report the small sample as a limitation. These numbers come from running the sample data through Data Lab's own code.
How the AI write-up stays tied to your numbers
The results section is written from the result tables, their APA sentences, their warnings and the variable summaries. The raw rows are never sent. The AI is instructed to use only the numbers given, exactly as written, and to add no test or variable. It writes four parts: Sample, Results (one paragraph per result), Interpretation (without claiming causation from correlations) and Limitations (from the warnings, sample size and tests used). The numbers are not re-checked automatically after writing, so compare each one with your result cards before you use the text.
What it does not do
- It runs the tests above and no others, and it does not test assumptions formally. It warns on small samples and sparse tables and tells you what to check.
- Linear regression takes one predictor; logistic regression takes up to 10.
- Excel files are read for their values only; formulas are not run. Files up to 20 MB, first 50,000 rows.
- Saved analyses keep results, not data. Upload the data again to run more tests.
- It does not decide what your results mean for your field. Clinical or policy conclusions need a qualified professional's review.
Who uses it
Undergraduates and master's students analyzing a dissertation dataset (see AI for college students and AI for master's students), PhD candidates checking results before a chapter (see AI for PhD students), and clinicians running a first look at audit data (see AI for medical research). For a quick check without signing in, the free data analysis tool runs descriptives, correlations, t-tests and chi-square. To plan the analysis before you collect data, use the Methodology Builder.
Learn more
Last updated 2026-11-12