Congruence Test
The bootstrapped congruence-coefficient test of Franke, Sarstedt & Danks (2021) — are two constructs empirically distinguishable in their nomological networks, or are they redundant? Paste your item data and your SEMinR model; no R and no install.
Your data never leaves your computer
The entire analysis — estimating the model, every bootstrap resample, every confidence interval — runs inside your browser, on your own machine. There is no upload, no server, no account, and no logging. Nothing you paste is ever transmitted anywhere. Once this page has loaded, the only request it can make is fetching the bundled demo dataset from this site, and only if you press “Load demo values”.
You do not have to take our word for it: disconnect from the internet and run your analysis anyway. It works offline. That check takes ten seconds and proves the point better than any badge.
The engine is the open-source TypeScript port of SEMinR and seminrExtras, and it reproduces the R implementation exactly — including R's random number generator, so a given seed resamples the same rows R would.
Your model is estimated here in your browser, and the bootstrap
re-estimates it on every resample — the same procedure as
seminrExtras::congruence_test(). HTMT is computed for you and shown
alongside the tested coefficients.
1. Paste your indicator data
One column per item with a header row of item names, one row per
case. Comma- or tab-separated, so pasting straight from Excel or SmartPLS
works. Extra columns are ignored. Missing values coded -99 are
mean-replaced, as in the textbook.
2. Paste your SEMinR model
The same code you would run in R. constructs() with
composite() / reflective(),
multi_items() / single_item(), optional
weights = mode_B, plus relationships() with
paths(). Assignments and comments are fine.
Options
Results
How to read this. A pair is significant when the whole bootstrap confidence interval falls below the threshold — the two constructs are empirically distinguishable. Non-significant pairs are the ones to worry about: redundancy cannot be ruled out. Diff is the distance from the threshold; the verdict comes from the confidence interval, not from Diff or T Stat.
The HTMT-based column on the right is a point estimate only. It is congruence computed over the disattenuated HTMT matrix rather than the construct correlation matrix, so comparing it with Original Est. shows whether measurement error is inflating the apparent congruence. It is not bootstrapped — no confidence interval, no significance test — so treat it as context, never as a second verdict.