STEYX in WPS Spreadsheet: Regression Standard Error with Complete Residuals

• Introduced: documented by 2019-06-17 in the official WPS lesson; the first release is not established.
• Availability: the lesson identifies WPS Office for Windows Personal, without a numeric release range.
• Scope: exact current supported builds and other WPS clients remain unverified. Historical naming does not establish retirement.
• Checked: October 14, 2026. Documentation and independent calculations; no native WPS test was performed.

=STEYX(known_y, known_x)

The required ranges contain dependent Y values and independent X values. The dated lesson documents #N/A for unequal data counts and #DIV/0! for empty inputs or fewer than three data points.

Referenced text, logical values and blank cells are described as ignored, while numeric zero is included. The source does not fully define differently positioned missing-value pairing, so all primary examples use complete numeric pairs. Direct coercion and error-input behavior are not inferred from the range controls.

Examples use English names and comma separators. Keep the Y argument first even when the table is arranged with X in the left column.

Enter X values 1, 2, 3 and 4 in A2:A5 and corresponding Y values 1, 3, 4 and 8 in B2:B5:

=STEYX(B2:B5,A2:A5)

The independent least-squares line with an intercept is Y = -1.5 + 2.2X. Its fitted and residual values are:

Squared residuals total 1.8. There are four observations, so the documented denominator is 4-2 = 2. Expected STEYX is SQRT(0.9), approximately 0.948683298.

In C2 enter =-1.5+2.2*A2, in D2 enter =B2-C2, and in E2 enter =D2^2. Fill all three formulas through row 5:

=SQRT(SUM(E2:E5)/(4-2))

Expected result: approximately 0.948683298, matching the independently modeled STEYX calculation. The residuals also sum to zero in this fitted intercept model, which is a useful arithmetic check.

The source’s equivalent centered-sums equation uses X-square total 5, Y-square total 26, and centered cross-product total 11. The residual square total is 26-11^2/5 = 1.8.

Dividing the same squared residual total by n instead of n-2 gives:

=SQRT(SUM(E2:E5)/4)

Expected result: approximately 0.670820393. This is the root mean squared residual for these four observations, not the documented STEYX result.

The two values differ because their denominators answer different statistical questions. Do not label every square-root-of-average-error expression “standard error” without specifying its model and degrees of freedom.

Multiply every original Y by ten. The fitted values and residuals also multiply by ten, so the squared residual total becomes 180. The expected STEYX becomes approximately 9.486832981, ten times the original.

Adding 100 to every original Y instead shifts the fitted intercept without changing residuals. Expected STEYX stays approximately 0.948683298.

Changing X units consistently leaves this model’s fitted Y values unchanged after the line coefficients are adjusted. Changing Y units changes the error’s scale. Report the units beside the result rather than treating it as a dimensionless quality score.

Repeat the complete four-pair dataset twice. The independently fitted line remains the same, and the residual squared total doubles to 3.6. With eight observations, n-2 becomes 6.

The resulting expected STEYX is SQRT(0.6), approximately 0.774596669. Duplicating records does not preserve this statistic and does not create new independent evidence for the regression.

For a separate exact-line control, X=1,2,3 and Y=3,5,7 have zero residuals and an expected STEYX of 0. At least three observations are used, consistent with the source’s explicit minimum-count rule.

STEYX is not a complete prediction interval. A target-specific interval needs additional statistical assumptions and calculations; this one scalar does not by itself describe future uncertainty at every X.

RSQ measures squared correlation, while STEYX is expressed in Y units. SQRT and SUM support the transparent residual check.

A constant-X dataset makes the centered equation’s X denominator zero; the source does not establish its exact native error here. Mixed-position pairing, overflow and floating-point cancellation remain unverified boundaries.

• WPS Academy: STEYX, 2019-06-17 (Chinese).
• Editorial testing standards.

Worked outputs are independently calculated expectations, not captured native results. Function-specific source gaps and product/client boundaries remain explicitly scoped.