SVEND vs Minitab

Most tool comparisons in this category come down to which methods each one implements, and on that axis the answer is dull: both compute the same indices from the same standards, and where they differ the differences are listed below and are mostly about coverage rather than correctness.

The interesting difference is what survives the analysis. A statistical package produces a number and a chart, and what happens to them afterwards is somebody’s filing problem. SVEND is built so the run leaves a record: which instrument produced the readings, whether it was in calibration on the day, what estimator produced the dispersion, and a chain that shows the recorded acts have not been altered since. A quality function is judged on the evidence it can produce a year later, and that is the axis this page is really about.

If your organization requires Taguchi methods, mixture designs, or Monte Carlo simulation natively. If your customer specification mandates Minitab output format. If your team has 20 years of Minitab macros and the switching cost exceeds the license savings. These are legitimate reasons.

If the work does not stop at the number. An instrument coming back from calibration needs a sticker, and SVEND computes what that sticker says from the record rather than asking whoever is printing it — there is no way to get a CALIBRATED label for an instrument that is not. The square on it is either a QR a phone can scan to reach the certificate, or a GS1 DataMatrix carrying the instrument’s Global Individual Asset Identifier under your organization’s own GS1 prefix, which your scanner and your asset register already understand. An organization without a prefix gets no mark rather than an invented one.

If you need value stream mapping with discrete-event simulation beside the analysis that justified it. If you need web access without IT installing desktop software. If your budget is per-seat-per-month, not capital expenditure. If you want Bayesian capability analysis. If you want free tools your customers and suppliers can use without a license. And if you want to see the reference behind the number — results are calibrated against NIST Statistical Reference Datasets, with control charts and capability cross-checked against R’s qcc.

The honest account of a migration, because the cost of one is usually underestimated by whoever is proposing it.

Worksheets move. A .MTW opens on the bench and a study leaves as one. Column names longer than eight characters are truncated by the format itself rather than by us, and a missing value travels as the sentinel Minitab uses for a blank rather than as a zero, because a zero in somebody’s capability study is a measurement that never happened.

Macros do not move. There is no translation layer for a session macro, and pretending otherwise would waste an evaluation. A team with a large macro library has a real switching cost, and it belongs in the decision.

Installation stops being a step. The bench runs in a browser, so a new analyst is working the same afternoon and a plant with a locked-down desktop image needs no software review to add one.

An auditor asking why a number should be believed is asking two separate questions, and they have different answers.

Is the arithmetic right? The statistics are calibrated against the NIST Statistical Reference Datasets, whose values are certified to fifteen significant digits, and control charts and capability are cross-checked against R’s qcc. A result that misses a certified value fails the build rather than reaching a customer. The current calibration result →

Was it the right arithmetic for this process? This is the question a capability number usually cannot answer. A dispersion estimated from subgroup ranges is admissible under one of the eleven time-dependent distribution models ISO 22514-2 describes, and tool wear and lot-to-lot variation are different models. SVEND states which estimators produced an index and the restriction they carry, so the answer to the auditor is on the record rather than reconstructed from memory.

The comparison above is real, and it is also not the only way to read the question. Most of the people weighing this already have years of work in Minitab worksheets and no intention of throwing that away, which is sensible. SVEND reads and writes .MTW worksheets in both directions — open one on the bench, run the study, take a worksheet back. A Gage R&R exports with RunOrder, StdOrder, Part and Operator, which is the shape the analysis expects on the other side.

A browser bench for the everyday work and a desktop package for what it has done well for decades is a perfectly reasonable way to run a quality function. How worksheets move between them →