Software comparisons and independent calculations

Consistent PLS: check the reliability correction

Consistent partial least squares (PLSc) adjusts ordinary PLS estimates using a reliability estimate. These tests check both valid corrections and cases where the correction cannot be calculated. They compare results with cSEM, an independent R package, and with independent formulas where the programs use different loading definitions.

Reviewed 4 November 2026
1.33e-15Largest difference in the first comparison

What the comparison shows

The tested calculations produced the expected estimates and rejected the cases with invalid reliability corrections. Some corrected loadings and rho_A reliability values were checked against independent formulas. Those results do not claim that QuickPLS corrected loadings equal cSEM’s unmodified common-factor loadings.

Check the correction as well as the PLS weights

PLSc needs a valid reliability estimate before it can adjust the ordinary PLS result. Reliability describes how consistently indicators measure their construct. A successful test must therefore check both the correction and what happens when that correction is invalid.

Ordinary PLS weights determine how indicators contribute to a construct score. Corrected loadings describe a different part of the result. Matching the weights alone would leave the corrected estimates untested.

Which software produced these results?

This comparison uses the version of the QuickPLS calculation code shown below, cSEM and independent loading formulas. The versions identify the actual test; later release labels do not describe a new run of it.

On a small screen, swipe the table sideways to see every column.

RecordQuickPLS build or result checkedReferenceComparison outcome
Consistent PLS correctionSaved calculation-engine 1.1.7 results; reviewed 28 October 2026cSEM 0.6.1, independent loading algebraPassed the comparisons documented for this record

Valid corrections and cases that should be rejected

The 24 cases include eight valid estimates and sixteen cases expected to be rejected because the reliability correction is invalid. Rejecting an invalid case is the expected result, rather than a missing successful fit.

The independent formulas also check the stated loading definition and rho_A, a reliability measure. This matters because two programs can use the word loading for differently defined results.

  • QuickPLS corrected loadings are not defined as cSEM common-factor loadings. Read the reference definition before treating a loading row as a direct package-to-package comparison.
What was checkedValue / scope
Benchmark scenarios24
Compared numbers and missing-value statuses448
Outside the stated tolerance0
Different scenario names in the comparison record24
Largest absolute difference1.33226762955019e-15

Read weights and corrected loadings separately

Each indicator has a weight row and a loading row. For x1, the loading is shown as 1 in both columns under the checked definition. That does not demonstrate perfect measurement in a research study. It shows agreement for this test input and loading calculation.

The display uses 15 significant digits. Differences were calculated before this display rounding, so they may be nonzero when the shown values look equal. Use the download to see the recorded precision and which reference formula applies to a row.

On a small screen, swipe the table sideways to see every column.

Quantity comparedQuickPLSReferenceAbsolute differenceComparison result
Outer weight: x1 in x
Result identifierouter_estimates/x/x1/weight
0.5024338231848250.5024338231848253.33066907387547e-16Pass
Outer loading: x1 in x
Result identifierouter_estimates/x/x1/loading
110Pass
Outer weight: x2 in x
Result identifierouter_estimates/x/x2/weight
0.5001239501177330.5001239501177332.22044604925031e-16Pass
Outer loading: x2 in x
Result identifierouter_estimates/x/x2/loading
0.9999933544294410.999993354429447.7715611723761e-16Pass
Outer weight: z1 in z
Result identifierouter_estimates/z/z1/weight
0.4952435237802620.4952435237802622.22044604925031e-16Pass
Outer loading: z1 in z
Result identifierouter_estimates/z/z1/loading
0.9997023053279470.9997023053279472.22044604925031e-16Pass

Allowed differences and expected missing results

The limits below apply after the reference has been expressed using the same result definition. Invalid corrections also have to be reported as unavailable where expected. A numerical match does not erase a difference between loading definitions.

abs(native - aligned_reference) <= absolute_tolerance + relative_tolerance * abs(aligned_reference)

In this formula, native is the QuickPLS value and aligned_reference is the reference value expressed using the same definition. abs means absolute value, so the sign does not affect the size of the difference. Absolute tolerance is a fixed allowance. Relative tolerance adds an allowance that grows with the size of the reference value. A separate check verifies whether a result is available or missing where expected; it does not estimate another effect.

On a small screen, swipe the table sideways to see every column.

Absolute toleranceRelative toleranceNumeric or missing-value checksWhere the comparison rule was specified
1e-081e-08448Comparison script

What these checks tell you

The evidence checks the correction where it is valid and the refusal to calculate it where it is invalid. It therefore covers more than agreement in the ordinary PLS weights.

Use the reference named for each result. A row checked by an independent formula supports that formula-based calculation; it should not be described as an unchanged cSEM output. Compared values and missing-result checks are not separate studies.

  • The error limits apply to the matched definitions. The loading-definition note remains part of the result even when every recorded check passes.

Inspect the comparison files

Use the CSV to inspect individual weights, corrected loadings and their differences. The JSON report explains the calculation versions, valid and rejected cases, and the reference definitions.

These are results from completed tests. Repeating them requires the original inputs, model, settings and software versions. The downloads do not include those datasets or the application source code.

Download the comparison results

See the values we checked, the differences we found and the software versions used. The download also identifies the original test files.

About these tests

These are QuickPLS development comparisons. Each record identifies the reference software or mathematical calculation, the tested build and the settings used.

How to understand these comparisons · Find checks for another method