# JASP Tutorial: Free Statistics Software for t-Tests, ANOVA and Bayesian Analysis JASP is a free, open-source statistics program with a point-and-click interface that runs t-tests, ANOVA, regression and Bayesian analyses without writing any code. You download it from the JASP project, open a data file, and the output updates live as you tick options in the analysis menu. That combination is unusual. Most statistics packages force a choice between an easy graphical interface with a subscription fee, or a free but command-driven environment that requires programming. JASP was built to close that gap, and it did so by putting both classical (frequentist) tests and Bayesian alternatives in the same menu for the same data [1]. This tutorial walks through the interface, the t-test and ANOVA workflows, the Bayesian machinery underneath them, and the reporting decisions that trip people up. ## What JASP Is and Where to Get It JASP is an open-source, cross-platform graphical statistics package. The name originally stood for Jeffreys's Amazing Statistics Program, a nod to the Bayesian pioneer Harold Jeffreys [2][3]. JASP runs under the Hood on the R statistical language, and several of its Bayesian routines are direct interfaces to established R packages. The Bayesian t-test, ANOVA and correlation procedures draw in part on the BayesFactor package for R [1], and the Bayesian regression module is built on the BAS package [4]. The **JASP download** is a single installer for Windows, macOS and Linux from the project website. There is no license key, no seat count and no subscription, which matters for students, teaching labs and researchers on small grants. Analyses are saved inside a single .jasp file that contains both the data and the output, so a colleague can reopen your file and see exactly which boxes you ticked. JASP also uses optional modules that add specialized analyses to the standard menus. The BFpack module, for example, adds Bayesian hypothesis tests for many kinds of correlation coefficients [5]. The meta-analysis module adds [Bayesian model](/blog/data-analysis/what-is-bayesian-model)-averaged meta-analysis [6]. Modules install from inside the program with a few clicks, and they keep the main interface uncluttered. ### Who JASP Is Built For The design targets researchers who need defensible inference but are not programmers: psychologists, nurses, pharmacists, clinicians, biologists and graduate students. A nursing education tutorial published in the *Journal of Advanced Nursing* uses JASP to teach a Bayesian independent t-test to nurse and midwifery researchers [7]. A guide in *BMC Psychiatry* introduces Bayesian alternatives to standard null hypothesis significance tests for exactly the same reason, citing the scarcity of accessible software [8]. A pharmacy education paper likewise used JASP to teach Bayesian reasoning to clinical pharmacists [2]. ### Analysis Overview | Analysis | Where it lives in JASP | Classical output | Bayesian output | Number to report | |--|--|--|--|--| | Independent t-test | T-Tests, Independent Samples | t, df, p, Cohen's d, 95% CI | Bayes factor, posterior median, 95% credible interval [7] | BF10 plus effect size with interval | | One-way ANOVA | ANOVA, One-Way ANOVA | F, p, effect size, post hoc tests | Bayes factor for model comparison | BF10 for the model against the null | | Correlation | Regression, Correlation | r, p | Bayes factor for null versus alternative [5][8] | BF10 and posterior estimate of r | | Linear regression | Regression, Linear Regression | Coefficients, standard errors | Inclusion probabilities, model-averaged coefficients [4][3] | Inclusion BF per predictor | | Reliability | Reliability analysis | [Cronbach's alpha](/blog/data-analysis/cronbachs-alpha-definition-formula) | Credible intervals for five coefficients [9] | Posterior probability that reliability exceeds .80 | | Meta-analysis | Meta-Analysis module | Pooled effect, I squared | Model-averaged Bayes factor [6][10] | BF10 for effect and heterogeneity | ## The Basic JASP Workflow JASP opens into three panes. The data spreadsheet sits on the left or top, the analysis ribbon runs along the top, and the results panel fills the rest of the window. 1. Open or paste your data. JASP reads CSV and SPSS-format files and accepts typed or pasted columns. 2. Check variable types. JASP infers scale, ordinal or nominal from the data, and you can override the guess in the column header. 3. Pick an analysis from the ribbon. The output appears immediately, empty until you assign variables. 4. Drag variables into the input boxes. The results table refreshes with every change, which makes exploration fast and visible. 5. Tick assumption checks, effect sizes and plots before you look at the p value. 6. Save the .jasp file so the analysis is reproducible. The live-refresh behavior is the main practical difference from menu-driven legacy packages, where you run an analysis and then inspect a static output window. In JASP, changing a factor or adding a covariate rewrites the table in front of you, which makes it easier to notice when a result is fragile. ## Running a JASP T-Test The t-test comparisons sit under T-Tests in the ribbon. Three variants cover most designs: one-sample, paired samples and independent samples. For an **independent samples t-test**, move the grouping variable (which must have exactly two levels) into the Grouping Variable box and the outcome into the Dependent Variable box. Under Options you can choose Student's t or Welch's t. Welch's version does not assume equal variances across groups, and it is the safer default when group sizes differ. You can also request the location parameter (mean difference or median difference), a confidence interval, descriptives and a rain-cloud or boxplot. Effect size output includes Cohen's d with a confidence interval. Report it. A p value tells you how surprising the data are under a null model, it does not tell you how large the difference is or how precisely you have estimated it. The Bayesian independent samples t-test sits in the same T-Tests menu under Bayesian Independent Samples. It returns two things [7]: - A **Bayes factor**, written BF10, which quantifies the relative evidence in the data for the [alternative hypothesis](/blog/data-analysis/alternative-hypothesis-definition-examples) (H1) over the null hypothesis (H0). BF01 is just the reciprocal and gives evidence for the null. - A **posterior distribution** of the effect size, summarized by a median and a 95% credible interval. A 95% credible interval means that, given the model, the prior and the data, there is a 95% posterior probability that the true effect falls inside that range [7]. A worked clinical-pharmacy example shows what these outputs look like in practice. Using a Cauchy prior with scale r = 0.707 at location 0, an analysis of emergency department visits produced BF = 4.082, meaning the observed data were about four times more likely under the alternative hypothesis that the intervention was effective. The median effect size was 0.337 with a 95% credible interval from 0.074 to 0.635 [2]. The authors then ran a robustness check, recomputing the Bayes factor across a range of prior widths to show the conclusion did not depend on one arbitrary choice [2]. The robustness check is not optional decoration. Bayes factors depend on the prior, so showing that the evidence is stable across reasonable priors is the Bayesian equivalent of a sensitivity analysis. ### Why the Bayesian Version Appeals to Researchers The usual frequentist p value can only reject a null hypothesis. It cannot support the null, and a non-significant result is routinely misread as "no effect" when it may simply mean "not enough data." A Bayes factor can quantify evidence in either direction, which makes it possible to argue that an effect is genuinely absent rather than merely undetected [11]. That evidence can also be expressed as an odds ratio, an easy format to communicate [8]. A second advantage is evidence monitoring. Bayesian hypothesis tests let researchers track evidence as data accumulate without needing to know the intention behind the data collection, which supports sequential designs and stopping rules [7][12]. A trialist can look at the Bayes factor after each block of participants and stop when the evidence is decisive, a practice that frequentist designs handle awkwardly. ## Running a JASP ANOVA The ANOVA family sits under ANOVA in the ribbon. One-Way ANOVA takes a single categorical factor and a continuous dependent variable. Factorial ANOVA takes several factors and reports main effects and interactions. Repeated measures ANOVA handles within-subject designs, and ANCOVA adds covariates [8]. The workflow mirrors the t-test: 1. Assign the dependent variable and drag factors into the Fixed Factors box. 2. Under Model, choose the terms you want, including interactions. 3. Tick Descriptive statistics and Bar plots for orientation. 4. Open Assumption Checks and request Levene's test for homogeneity of variance and a Q-Q plot for normality. 5. Under Post Hoc Tests, move the factor into the right-hand box and select a correction such as Tukey for all pairwise comparisons. 6. Under Additional Options or Effect Size, request eta squared, partial eta squared, or omega squared. Report the F statistic with its degrees of freedom, the p value, the effect size and the post hoc comparisons. Degrees of freedom matter because the F test has two of them, one for the effect and one for the error term. If Levene's test is significant, the equal-variance assumption is shaky. JASP offers a Welch correction for one-way designs, which adjusts the denominator degrees of freedom. Do not ignore the assumption check and then report the standard F test. The Bayesian ANOVA lives in the same menu. It compares a set of candidate models against each other, using the null model (no effects) as the reference. The output includes BF10 values for each model and, importantly, inclusion Bayes factors that pool evidence across matched models to answer a more focused question: how much evidence is there that a given effect is present, averaged over the models that could contain it [13][8]. For a two-group design, the Bayesian ANOVA and the Bayesian t-test give the same answer, which is a useful consistency check. Interpretation follows the usual convention. A Bayes factor near 1 is inconclusive, values around 3 to 10 indicate moderate evidence, and values above 10 indicate strong evidence, following Jeffreys's classic labeling scheme. Report which hypothesis the number favors, not just the magnitude. ## Bayesian Analysis in JASP Beyond the Basics JASP's Bayesian menu extends well past the t-test and ANOVA. The breadth is why the program became a standard teaching tool for applied Bayesian work [11][14]. ### Correlation and the BFpack Module Classical correlation testing asks whether r differs from zero, using a p value with well-known limitations. The BFpack module implements Bayesian tests for product-moment, polyserial and tetrachoric correlations, supports partial correlations that control for covariates, and can test whether two correlations are equal to each other as well as to zero, for both dependent and independent samples [5]. That last capability is rare in point-and-click software and it addresses a common applied question, such as whether a correlation is stronger in one subgroup than another. ### Reliability Analysis Most reported reliability analyses give a single Cronbach's alpha with no sense of sampling error. JASP implements Bayesian estimation routines for five popular single-test reliability coefficients, so you can attach a credible interval to the estimate and quantify its precision [9]. The posterior distribution also lets you ask a directly practical question, for example: what is the probability that the reliability of my test is larger than .80 [9]. That is a far more useful statement for a test developer than a bare point estimate. ### Regression With Bayesian Model Averaging Standard regression analysis has two stages. You select the "best" model, then you interpret its coefficients as if the model selection never happened. That second stage ignores model uncertainty and produces overconfident estimates that generalize poorly [4]. Bayesian model averaging keeps every model in play and weights each one by its posterior probability. JASP implements this through the BAS package [4]. The output includes posterior inclusion probabilities for each predictor, which answer the question "how much evidence is there that this variable belongs in the model" on a probability scale. A comparative study of body mass index predictors used JASP to show the practical difference: the frequentist model returns one coefficient per predictor, while the Bayesian model returns a full posterior distribution for each, plus posterior probabilities for ranking candidate models [3]. ### Meta-Analysis The JASP meta-analysis module handles Bayesian model-averaged meta-analysis, which combines models that assume the presence or absence of an effect with models that assume the presence or absence of between-study heterogeneity [6][10]. This design lets you quantify evidence for the absence of an effect, monitor evidence as studies accumulate, and draw inferences from multiple models at once [6]. Applied examples range from language development in children [6] to methodological work on publication bias, where JASP 0.18.3 was used alongside the metafor package in R to run a robust Bayesian model-averaging publication selection analysis [15]. ### Guidelines for Doing It Properly Van Doorn and colleagues published a set of JASP-specific reporting guidelines organized around four stages: planning the analysis, executing it, interpreting the results and reporting them [13]. Most guidance extends to Bayesian inference generally. The planning stage is the one people skip, and it is the one that determines whether a Bayes factor answers a question anyone asked. ## JASP vs SPSS The **JASP vs SPSS** comparison usually comes down to three things: cost, licensing and inference philosophy. SPSS is commercial software. Institutions pay for licenses, and students often depend on a campus arrangement that expires after graduation. JASP is free and open source for everyone, permanently [1]. SPSS is built around frequentist null hypothesis significance testing, and its Bayesian add-on is a separate licensed product. JASP ships frequentist and Bayesian versions of the same analysis side by side, so you can run a t-test and its Bayesian counterpart from the same menu on the same data, and see them in one results panel [8][1]. For a researcher who wants to report both, that saves considerable friction. SPSS has a deeper catalog of specialized procedures accumulated over decades, plus syntax files and a large legacy literature. JASP covers the core analyses most researchers actually run (t-tests, ANOVA, ANCOVA, correlation, regression, contingency tables, reliability, meta-analysis) and extends through modules. If a method exists as an R package but has no JASP module, you cannot reach it through the interface. The practical migration path is straightforward. JASP opens SPSS-format data files, so you can move a dataset across without re-entering anything. Collaborators who insist on SPSS output can still receive the raw data and a .jasp file that documents every setting. ## Common Mistakes and Limitations **Reading BF10 as the probability that the alternative is true.** A Bayes factor is a ratio of how likely the data are under two hypotheses. It is not a posterior probability unless you multiply by prior odds. Convert it with prior odds if you want a posterior probability, and state which you are reporting. **Reporting the wrong direction.** BF10 favors the alternative and BF01 favors the null. Mixed notation is the single most common error in Bayesian write-ups. Pick one, define it, and use it consistently. **Skipping the prior robustness check.** Bayes factors are prior-dependent, and JASP's default Cauchy prior for t-tests has a scale of 0.707 [2]. That default is a reasonable general-purpose choice, but reviewers will ask whether the conclusion survives other sensible priors. The robustness plot exists for this purpose. **Treating a non-significant p value as evidence for the null.** This is the classic error JASP's Bayesian tests were designed to fix [11]. If you need to claim absence of an effect, run the Bayesian version and report BF01. **Ignoring assumption checks because the interface makes it easy to click through.** JASP puts Levene's test and Q-Q plots one click away. Use them. Equal-variance and normality assumptions still matter for the classical tests, and the Welch correction exists precisely because they are often violated. **Confusing credible intervals with [confidence intervals](/blog/data-analysis/confidence-interval-formula-examples).** A 95% confidence interval describes repeated-sampling behavior of a procedure. A 95% credible interval describes belief about the parameter given the data and the prior [7]. They often look similar numerically and mean completely different things. **Forgetting that small samples stay imprecise in Bayesian analysis.** A Bayes factor computed from twelve participants is still a Bayes factor computed from twelve participants. Bayesian methods redistribute uncertainty, they do not remove it. **Treating inclusion Bayes factors as plain model Bayes factors.** The inclusion BF averages over matched models and answers a different question than the BF for a single model [13]. Check the footnote in the JASP output that explains which computation you are looking at. **Overwriting the .jasp file.** Because the file holds both data and analysis, a careless save can destroy the record of what you ran. Use version suffixes in the filename. **Assuming reproducibility without archiving the version.** JASP updates regularly, and module behavior can shift between versions. Record the version number in your methods section, the way the authors of a 2025 methodological meta-analysis recorded JASP 0.18.3 [15]. These limits are about interpretation, not about the software's accuracy. The computations are implemented from established R packages, and the tutorials that describe them are peer-reviewed [5][7][9][4][13][1][6]. Where things go wrong is almost always at the interface between the output and the sentence written in the manuscript. ## Frequently Asked Questions ### Is JASP really free? Yes. JASP is free and open source, with no license fee, subscription or seat limit [1]. You can install it on personal and lab machines without institutional approval. ### Can JASP open my SPSS data file? Yes. JASP reads SPSS-format data files as well as CSV files, so existing datasets transfer without conversion. ### What does a Bayes factor actually tell me? It is the ratio of how likely the observed data are under the alternative hypothesis versus the null. A BF10 of 4 means the data are four times more likely under the alternative [2]. ### Should I use Student's t or Welch's t in JASP? Use Welch's unless you have a strong reason to assume equal variances. It performs well when group sizes and variances differ. ### Does JASP handle repeated measures ANOVA? Yes. Repeated measures ANOVA is available from the ANOVA menu, alongside factorial ANOVA and ANCOVA [8]. ### What prior does JASP use by default for a Bayesian t-test? A Cauchy distribution centered at zero with a scale of 0.707 [2]. You can change the width, and you should run a robustness check across several widths. ### Do I need to know R to use JASP? No. Every analysis is point-and-click. Background packages such as BayesFactor and BAS are wired into the interface [4][1]. ### How do I report a Bayesian t-test in a paper? Report the test, the prior and its scale, BF10 (or BF01, clearly labeled), and the posterior median with its 95% credible interval. Add a robustness check if prior choice is contestable [13]. ## Sources 1. [Bayesian inference for psychology. Part II: Example applications with JASP.](https://pubmed.ncbi.nlm.nih.gov/28685272/) 2. [Introduction to Bayesian statistics: a practical framework for clinical pharmacists.](https://pubmed.ncbi.nlm.nih.gov/34697050/) 3. [Identifying the Influencing Factors for the BMI by Bayesian and Frequentist Multiple Linear Regression Models: A Comparative Study.](https://pubmed.ncbi.nlm.nih.gov/37970166/) 4. [A tutorial on Bayesian multi-model linear regression with BAS and JASP.](https://pubmed.ncbi.nlm.nih.gov/33835394/) 5. [A tutorial on Bayesian hypothesis testing of correlation coefficients using the BFpack-module in JASP.](https://pubmed.ncbi.nlm.nih.gov/41083837/) 6. [A tutorial on Bayesian model-averaged meta-analysis in JASP.](https://pubmed.ncbi.nlm.nih.gov/37099263/) 7. [A Tutorial on Conducting and Interpreting a Bayesian Independent T-Test Using Open-Source Software.](https://pubmed.ncbi.nlm.nih.gov/40755145/) 8. [Bayesian alternatives for common null-hypothesis significance tests in psychiatry: a non-technical guide using JASP.](https://pubmed.ncbi.nlm.nih.gov/29879931/) 9. [A tutorial on Bayesian single-test reliability analysis with JASP.](https://pubmed.ncbi.nlm.nih.gov/35581436/) 10. [Bayesian model-averaged meta-analysis in medicine.](https://pubmed.ncbi.nlm.nih.gov/34705280/) 11. [Bayesian alternatives to null hypothesis significance testing in biomedical research: a non-technical introduction to Bayesian inference with JASP.](https://pubmed.ncbi.nlm.nih.gov/32503439/) 12. [Bayesian inference for psychology. Part I: Theoretical advantages and practical ramifications.](https://pubmed.ncbi.nlm.nih.gov/28779455/) 13. [The JASP guidelines for conducting and reporting a Bayesian analysis.](https://pubmed.ncbi.nlm.nih.gov/33037582/) 14. [Applied Bayesian Approaches for Research in Motor Neuron Disease.](https://pubmed.ncbi.nlm.nih.gov/35401404/) 15. [Robust Bayesian Model Averaging Meta-Analysis of Menstrual Disorders in COVID-19 Survivors: A Methodological Meta-Analysis Study.](https://pubmed.ncbi.nlm.nih.gov/41394277/) ## Related Articles - [How to Use jamovi: A Free Statistics Guide for t-Tests, ANOVA and Regression](/blog/data-analysis/how-to-use-jamovi) - [GraphPad Prism Tutorial: t-Tests, ANOVA, Curve Fitting and Graphs for Biologists](/blog/data-analysis/graphpad-prism-tutorial) - [What Is a Bayesian Model? Definition and Examples](/blog/data-analysis/what-is-bayesian-model) - [One-Way vs Two-Way ANOVA: Differences, Assumptions and a Worked Example](/blog/research-skills/one-way-vs-two-way-anova) - [Two Sample t-Test: Formula, Calculation and Example](/blog/data-analysis/two-sample-t-test-formula-example)