The right statistical test, run and written up.

Open your data and type your question in your own words. Kaarta asks what it needs to know about your study, picks and runs the test, gives you the tables, charts and assumption checks, and drafts a paragraph for your paper.

Did anxiety go down after the programme?

  1. Your question, read as anxiety_pre and anxiety_post interpreted “Anxiety” before and after the programme matches anxiety_pre and anxiety_post.
  2. What do you want to do? Compare samples interpreted The researcher asks whether anxiety “went down after the programme”, which is a comparison of samples.
  3. What type of data is your dependent variable? Interval or ratio inferred anxiety_pre, anxiety_post is scale data (confirmed).
  4. How many sets of data are you comparing? Two inferred 2 occasions of one measure are being compared (anxiety_pre and anxiety_post).
  5. Are the sets of data independent or related? Related declared
Result
Paired samples t-test
Assumptions met
t(47)
3.64
p
< .001
Cohen’s d
0.53
Met
Normality
Shapiro-Wilk on the difference scores: W = 0.958, p = 0.084. No evidence against normality.
Write-up
A paired samples t-test was conducted to compare the anxiety_pre and anxiety_post conditions. There was a statistically significant difference, with a conventionally medium standardized effect: scores were higher in anxiety_pre than in anxiety_post (anxiety_pre: M = 41.30, SD = 9.00; anxiety_post: M = 39.20, SD = 9.75; N = 48 pairs), t(47) = 3.64, p < .001, d = 0.53, 95% CI for the mean difference (anxiety_pre minus anxiety_post) [0.94, 3.26].
Box plot + points
10305070anxiety_preanxiety_postOccasion

The study design determines the statistical test.

Any statistics package will calculate a test statistic on two columns of numbers. The hard part is knowing which test your study actually calls for. That takes knowing how the study was run, not just the numbers in the file, and a statistician’s expertise to match the design to a test.

anxiety-trial.csv 48 participants, each measured before and after the programme
Pairing ignored
Treated as two separate groups
Welch’s t-test
t(93.40) = 1.10, p = .28
Wrong for this design
Pairing respected
Each person compared with themselves
Paired samples t-test
t(47) = 3.64, p < .001
Right for this design

Both tests see the same 2.10-point drop in average anxiety. Because the paired test looks at each person’s own change, which varies much less here than the raw scores do across people, its standard error is much smaller. The paired test isn’t right because its p-value is under .001; it’s right because the same 48 people were measured twice. Illustrative data.

Any of these study facts can change the test.

  1. Whether observations are related
    Related
    Paired samples t-test
    Independent
    Welch’s t-test
  2. How the outcome is measured
    Interval or ratio
    Paired samples t-test
    Ordinal
    Wilcoxon signed-rank
  3. What you need to control for
    Nothing
    One-way ANOVA
    A covariate
    One-way ANCOVA

Set up the analysis

Ask your research question in your own words

Write your question the way you would put it to a colleague. A language model identifies which variables you mean and the role each one plays, such as outcome or grouping.

From there, a hand-written statistical engine selects the test that fits your question, your data and your study design.

See it in the demo

Inspect your data and how the file was imported

See summaries, distributions and grouped views of each variable, and a preview of your rows.

Kaarta also reports what happened on import: which missing-value codes it recognised, which columns it read as numbers and how many empty rows it dropped.

Confirm each variable’s measurement level

Kaarta infers whether each variable is nominal, ordinal or scale, and lets you confirm or change it.

If an ordinal outcome is stored as text labels (such as Poor, Fair, Good, Very good), you set the category order from lowest to highest.

Answer the design questions, then press Run

A spreadsheet cannot show whether measurements came from the same people or different groups, whether a factorial design is between, within or mixed, or whether one participant spans multiple rows. Kaarta asks you directly, with a plain explanation under each choice if you want it.

Once those questions are settled, Kaarta shows the test, variables, settings and rows it will use, and runs the analysis only when you press Run.

See it in the demo

Check and interpret the result

See the assumption checks alongside the result

Every result includes the assumption checks that apply to that test, such as Shapiro-Wilk or Levene’s test, calculated on the exact rows used in the analysis.

When an assumption that the test relies on is violated, Kaarta stops and explains why instead of publishing a misleading result or quietly swapping tests. In ANCOVA, for example, that happens when the regression slopes differ across groups.

See it in the demo

Read the effect size in context

Standardised effect sizes such as Cohen’s d or partial eta squared appear right next to the test statistic and p-value, labelled small, medium or large wherever an established convention exists.

Trace how the test was chosen

Every result records the rows it used, its method details (such as two-tailed testing and 95% confidence intervals), the full output tables and the decision trail that led to the test.

Each step on the trail shows where the answer came from: interpreted from your question, inferred from the data file, or declared by you.

See it in the demo

Follow up when needed

Run post hoc tests and simple effects

When an omnibus test is significant, Kaarta offers pairwise follow-up comparisons on the same sample, proposing Tukey HSD after a one-way ANOVA, or Games–Howell when Levene’s test shows unequal variances.

If a between-groups factorial design has a significant interaction, Kaarta offers simple-effects comparisons rather than main-effect post hocs. You pick the method and press Run.

Compare the result without outliers

For supported tests, Kaarta screens the analysed sample once using a fixed rule and lists any flagged cases up front.

If you choose to compare, it runs the same analysis without them and shows both results. The original is unchanged.

Repeat the same analysis across multiple variables

Apply a confirmed analysis across a batch of variables, such as testing the same group comparison across two dozen outcomes.

Each analysis is reported on its own, with its raw p drawn against the significance level and the number expected by chance alone beside it.

Report and reproduce the work

Copy a draft results paragraph in APA style

The APA write-up is generated directly from the fitted analysis, so every statistic, estimate and confidence interval in the prose matches the result card.

Copy the paragraph, then review and adapt the wording for your manuscript.

See it in the demo

Pick a figure and download it

Choose from the chart types suited to your analysis, such as a box plot with individual points or group means ±1 SE, with significant findings highlighted in colour.

Download the one you want as a PNG.

Export the analysis to R or Python

Download an R or Python script containing the exact options used in your analysis so you or a reviewer can rerun it independently. Each export states whether it is a faithful reproduction script or a template to adapt.

Every statistical runner in Kaarta is also checked in Kaarta’s test suite against reference values computed in R, as shown in the comparison table.

The examples on this page use made-up data. The R comparison is from Kaarta’s test suite.

Every analysis Kaarta can run.

Find yours by what you want to do: describe a sample, compare samples, analyse relationships, or examine structure and reliability.

Describe a sample

  • Mean, median and mode
  • Standard deviation, range and interquartile range
  • Category counts
  • Bar charts, pie charts and histograms

Compare samples

Measured outcomes

  • One-sample, independent-samples and paired t-tests
  • One-way and factorial between-groups ANOVA
  • Repeated-measures ANOVA, one factor or two within-participant factors
  • Mixed-design ANOVA
  • One-way and factorial ANCOVA
  • Mixed-design ANCOVA, one between- and one within-subjects factor

Ranked outcomes

  • Wilcoxon signed-rank, one-sample and paired
  • Mann–Whitney U
  • Kruskal–Wallis
  • Friedman

Categorical outcomes

  • Chi-square goodness of fit and test of contingencies
  • McNemar
  • Cochran’s Q

Analyse relationships

Association

  • Pearson and Spearman correlation
  • Correlation matrices of 3 to 12 variables
  • Point-biserial, rank-biserial and eta
  • Chi-square with phi or Cramér’s V

Prediction

  • Bivariate, standard multiple and hierarchical regression
  • Binary logistic regression, standard and hierarchical
  • Ordinal and multinomial logistic regression

Examine structure

  • Principal components analysis
  • Exploratory factor analysis, maximum likelihood

Examine reliability

  • Cronbach’s alpha and KR-20
  • Absolute-agreement ICC(2,1)
  • Cohen’s kappa, unweighted and quadratic-weighted

Follow-ups and options

  • Pairwise comparisons after a significant omnibus test
  • Simple effects after a between-groups interaction
  • Contingency-cell residuals
  • Sensitivity comparisons
  • One analysis across many variables, each reported on its own
  • Significance level of .05, .01 or .001

Your analysis runs on your computer.

All statistical calculations, charts and exported R or Python scripts are produced locally on your machine. Kaarta only uses a language model over the network to interpret your questions.

Stays on your computer

  • Your data file. Kaarta discards the uploaded copy once it has been imported.
  • Your data rows, unless you type specific values into the chat
  • Any figure that plots individual observations

Sent to the language model

To interpret your question, Kaarta sends what you type in the chat together with a summary of your dataset: the file name, row count, and each column’s name, measurement level, missing-value count, number of unique values and skew. If a categorical column has only a few distinct values, those category labels are included so Kaarta can recognise group names.

Individual data rows are never sent automatically, and values from columns initially recognised as identifiers or free text are excluded from the summary.

Your next analysis, run and written up.

Open a CSV or Excel file and type your question in your own words. Kaarta is coming to macOS and Windows.

Kaarta runs every calculation on your computer, and your data rows stay local unless you type them into the chat.