Two ways to check the same kind of model
A matrix summarises the relationships between data columns without listing each person's answers. Matrix inputs let us compare estimates under different measurement, weighting and scaling settings. The raw-data test starts with individual observations, so it can also check each person's construct scores and residuals, the differences left unexplained by the fitted model.
A path estimate describes a relationship between constructs. A loading describes an indicator’s relationship with its construct score. R2 describes the variation explained in an outcome. Checking one of these does not automatically check the others.
Which software produced these results?
The first row identifies the QuickPLS calculation code used directly for the matrix-input test. The second identifies the installed application used for the Org example. These are the tested versions; the table does not say that each later product release was rerun.
On a small screen, swipe the table sideways to see every column.
| Record | QuickPLS build or result checked | Reference | Comparison outcome |
|---|---|---|---|
| PLS moment inputs versus cSEM | Saved calculation-engine 1.1.7 results; reviewed 28 October 2026 | cSEM 0.6.1, R 4.6.1 | Passed the comparisons documented for this record |
| Installed PLS/PLSc/HTMT/blindfolding | 1.01.19 | cSEM 0.6.1, SEMinR 2.5.0, independent transformations | Passed the comparisons documented for this record |
Compare estimates across the matrix-input settings
The comparison uses 72 test configurations covering Mode A/B/mixed measurement blocks, Path/Factor weighting, scaling, and covariance/correlation conventions. A measurement block is the group of indicators assigned to a construct. Mode A and Mode B use different rules to calculate indicator weights; mixed models contain both.
Path and Factor are different rules for updating scores using the relationships between constructs. Covariance keeps information about the original variable scales; correlation expresses relationships on a standardised scale. Matching these choices lets the comparison check weights, loadings, paths, R2 and effects on the same basis.
- These saved results were calculated by running QuickPLS code directly, outside the Microsoft Store app.
- Matrix inputs contain no individual data rows, so this test cannot compare individual scores. It also does not test confidence intervals or other statistical inference.
| What was checked | Value / scope |
|---|---|
| Benchmark scenarios | 72 |
| Compared numbers and missing-value statuses | 2160 |
| Outside the stated tolerance | 0 |
| Different scenario names in the comparison record | 72 |
| Largest absolute difference | 4.34097202628436e-14 |
Inspect a weight and its corresponding loading
The first two rows concern indicator x1 in construct x. Its weight determines its contribution to the score; its loading describes how closely it follows that score. Read the output label before comparing the numbers, because a weight and a loading answer different questions.
Values are displayed to 15 significant digits. The difference uses the numbers before this display rounding, so equal-looking entries can have a small nonzero difference. The download preserves the recorded precision and explains any changes made to align the definitions.
On a small screen, swipe the table sideways to see every column.
| Quantity compared | QuickPLS | Reference | Absolute difference | Comparison result |
|---|---|---|---|---|
Outer weight: x1 in xResult identifierweight/x/x1 | 0.502433823184825 | 0.502433823184825 | 1.11022302462516e-16 | Pass |
Outer loading: x1 in xResult identifierloading/x/x1 | 0.997460496235385 | 0.997460496235385 | 1.11022302462516e-16 | Pass |
Outer weight: x2 in xResult identifierweight/x/x2 | 0.500123950117733 | 0.500123950117733 | 0 | Pass |
Outer loading: x2 in xResult identifierloading/x/x2 | 0.997436953945498 | 0.997436953945498 | 1.11022302462516e-16 | Pass |
Outer weight: z1 in zResult identifierweight/z/z1 | 0.495243523780262 | 0.495243523780262 | 0 | Pass |
Outer loading: z1 in zResult identifierloading/z/z1 | 0.996798203047095 | 0.996798203047095 | 1.11022302462516e-16 | Pass |
Check scores and diagnostics from the installed app
The Org dataset has 305 cases, four reflective constructs and 21 indicators. Reflective indicators are treated as measures of an underlying construct. The largest difference in unrounded PLS results was 6.25e-9.
This example also checks consistent PLS (PLSc), which applies a reliability correction; HTMT, which helps assess whether reflective constructs are distinct; and blindfolding, which leaves selected data values out for a prediction-related calculation. The reference included 21 cSEM fits with observations omitted for blindfolding, plus one fit of the original data.
- This is one empirical dataset. Its large check count includes data, scores, residuals and repeated outputs from the same analysis.
- Where the app exported 4/6-decimal values, the reference was compared at that same precision. Equal rounded values do not show that the underlying full-precision values are identical.
- Individual unrounded blindfolding results were unavailable, so that part of the comparison is limited to the saved outputs.
- The example does not cover every setting, statistical inference, null-effect case or missing-data pattern.
| What was checked | Value / scope |
|---|---|
| Compared numbers and missing-value statuses | 19122 |
| Outside the stated tolerance | 0 |
| Different scenario names in the comparison record | 3 |
| Largest absolute difference | 6.25279295007886e-09 |
Inspect direct effects and indicator loadings
The first rows show direct effects, followed by indicator loadings. The first direct effect is positive and the second is negative in both calculations. Their small numerical differences concern software agreement; they do not tell you whether either relationship is statistically significant.
The identifiers keep the original result-row order. Values are displayed to 15 significant digits, while the absolute differences use the numbers before this display rounding. The download retains the recorded precision and the reference adjustments.
On a small screen, swipe the table sideways to see every column.
| Quantity compared | QuickPLS | Reference | Absolute difference | Comparison result |
|---|---|---|---|---|
Organizational Identification → Affective Commitment (Joy)Result identifierdirect_effects/0/direct_effect | 0.613918191541941 | 0.613918191541606 | 3.34843264226947e-13 | Pass |
Organizational Identification → Affective Commitment (Love)Result identifierdirect_effects/1/direct_effect | -0.409190265204391 | -0.409190265228607 | 2.42166287023338e-11 | Pass |
Organizational Prestige → Organizational IdentificationResult identifierdirect_effects/2/direct_effect | 0.362253394380517 | 0.362253394384034 | 3.5170200085588e-12 | Pass |
Outer loading: Ac joy1 in Affective Commitment (Joy)Result identifierouter_loadings/0/loading | 0.7428151741878 | 0.742815174589978 | 4.02178512715068e-10 | Pass |
Outer loading: Ac joy2 in Affective Commitment (Joy)Result identifierouter_loadings/1/loading | 0.788433671909429 | 0.788433671925671 | 1.62416746718463e-11 | Pass |
Outer loading: Ac joy3 in Affective Commitment (Joy)Result identifierouter_loadings/2/loading | 0.822259930785207 | 0.822259930969232 | 1.84025017446743e-10 | Pass |
Allowed differences for the matrix-input test
The table gives the numerical error limits for the matrix comparison. The reported maximum difference falls within these limits. Numeric values and the presence or absence of results are checked separately.
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 tolerance | Relative tolerance | Numeric or missing-value checks | Where the comparison rule was specified |
|---|---|---|---|
| 1e-08 | 1e-08 | 2160 | Comparison script |
Allowed differences for the installed-app test
These limits apply after matching result definitions and the precision of exported values. They should not be read as a full-precision claim for outputs available only in rounded form.
abs(native - aligned_reference) <= absolute_tolerance + relative_tolerance * abs(aligned_reference)
Use the same fixed-plus-proportional rule explained above, with the limits in this table. Missing-result checks still ask whether the output is unavailable in the expected places.
On a small screen, swipe the table sideways to see every column.
| Absolute tolerance | Relative tolerance | Numeric or missing-value checks | Where the comparison rule was specified |
|---|---|---|---|
| 1e-08 | 1e-08 | 19122 | Specified separately for each compared value |
What these checks tell you
The matrix test supports agreement across its stated calculation settings. The installed-app test supports agreement for the saved Org example, including the outputs described above. Neither result means that a research model is valid simply because the software calculations agree.
Count each compared value or missing-result check once. Several checks come from the same dataset and model, so the totals are not counts of independent studies.
- For matrix inputs, use the error limits beside the matrix test.
- For the installed-app example, also check whether the result was compared before or after export rounding.
Inspect the comparison files
Open the CSV in a spreadsheet to inspect the named values and differences. The JSON report records the versions, settings and findings behind the matrix-input and installed-app comparisons.
These downloads report completed tests. To repeat one, you also need its original input, model, settings and software versions. The original datasets and application source code are not included.
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