Every analysis here runs on the R packages your field already cites. We supply the interface, the figures and the write-up — not a second implementation of the maths that could quietly disagree with the published record.
No account needed. Your data stays in your browser — only the numbers for one model are sent to be computed, and nothing is stored.
Pool results across studies and find out why they disagree
Combine effect sizes from several studies into one estimate, quantify how much the studies really differ, test whether subgroups or study characteristics explain it, and check whether the missing studies would have changed the answer.
Test whether your measurement model fits
Specify which items measure which construct and check whether the data agree — fit indices, standardised loadings, reliability, and the modification indices that show exactly where the model breaks down. Then test whether the instrument works the same way in every group.
Fit a path model with latent variables
Draw the paths between your variables — measured or latent — and fit them all at once. Direct, indirect and total effects with bootstrapped confidence intervals, explained variance for every outcome, model fit, and a path diagram you can put in a paper.
Each of these will slot into the same workbench — same data loading, same project files, same automatic reporting.
Find the structure underlying a set of items
Parallel analysis and scree plots to decide how many factors, then extraction and rotation with a pattern matrix you can actually read.
Check whether a scale holds together
Cronbach's alpha, McDonald's omega, item-total correlations and what happens to reliability if each item is dropped.
Meta-analysis runs on metafor; confirmatory factor analysis and structural equation modelling run on lavaan, and so will exploratory factor analysis. These are the packages cited in thousands of published papers. We do not reimplement their statistics, so there is no second implementation to disagree with the literature.
Each module ships a validation suite that checks every path — effect-size construction, transformations, weights, diagnostics — against direct calls to the underlying package on the same data. If a number would differ, the build fails.
Read the guide →Worked example · BCG vaccine trials
The classic demonstration that a pooled estimate can hide a moderator: BCG works far better in trials run further from the equator.
Reproduce this in the workbench →Worked example · Holzinger & Swineford 1939
The dataset every factor-analysis tutorial starts from. Three factors, nine tests, and a model that fits well enough to publish and badly enough to argue about.
Reproduce this in the workbench →Worked example · Industrialisation and democracy
Bollen's 75-country model, with three latent variables and the indirect route that carries most of the effect. The decomposition is read off the arrows — you never write it down.
Reproduce this in the workbench →