PlanMaker STEYX: Residual Standard Error and Worked Regression Checks

• Introduced: Documented by the PlanMaker 2010 manual; the first introduction release is not established.
• Availability: Documented in PlanMaker 2026; this is not a final working-version endpoint.
• Scope: The reference belongs to the 2026/NX manual family. Individual FreeOffice and mobile-build availability remains unverified.
• Checked: October 14, 2026, against the official function reference. Examples use vendor results and independent calculations; no native PlanMaker test was performed.

=STEYX(y_values, x_values)

Both arguments are required, normally as cell ranges. The PlanMaker reference puts dependent y values first and independent x values second, and identifies least squares as the fitting method. Use complete, aligned numeric pairs for the examples.

For a straight line with a fitted intercept and slope, the standard residual-error calculation is the square root of the residual sum of squares divided by n−2. The two fitted coefficients account for the two subtracted degrees of freedom. NIST’s least-squares treatment gives the general n−p form. This mathematical check reproduces SoftMaker’s published example; the short PlanMaker entry does not print its complete implementation algorithm.

Formulas use English names and commas. Expected decimal results below are rounded for display.

Create ResidualSpread with Input x and Response y in A1:B1. Enter:

Enter these summaries in H:

H5 is the observation count for this complete numeric fixture. It should not be used to silently resolve missing or differently filtered pairs.

Put Fitted y, Residual and Squared residual in C1:E1. Enter =$H$3+$H$2*A2 in C2, =B2-C2 in D2 and =D2^2 in E2. Fill C2:E2 through row 5.

In H7 enter =SUM(E2:E5); expect 7.2. In H8 enter =SQRT(H7/(H5-2)); expect approximately 1.897366596, checking H4 through the residual calculation.

In H9 enter =SQRT(H7/H5); expect approximately 1.341640786. That is the root mean squared residual using a denominator of four. It is a different summary from the fitted-line residual standard error, which uses two here. State the denominator rather than treating the labels as interchangeable.

On ResidualSpread, put Response times ten in J1 and Response plus 100 in K1. Enter =10*B2 in J2 and =B2+100 in K2, filling both through row 5.

Scaling the response by ten scales every residual and this spread measure by ten. Adding 100 shifts the fitted intercept but leaves residuals unchanged. STEYX is therefore expressed in response units, not a unitless percentage.

Create PerfectLine with x values 1, 2, 3, 4 in A2:A5 and y values 3, 5, 7, 9 in B2:B5; put x and y headers in row 1. Enter =STEYX(B2:B5,A2:A5) in D2. The mathematical target is 0 because the line y = 1 + 2*x has no residuals for these four points.

Zero on this fixture does not guarantee perfect predictions elsewhere. With two distinct points, a line can fit exactly, but the n−2 residual-error denominator is zero. The cited entry does not specify the exact native result for that boundary, so the control deliberately uses four observations.

Create ResistorError with Temperature C and Resistance ohms in A1:B1. Enter temperatures 8, 20, 25, 28 in A2:A5 and resistances 261, 508, 608, 680 in B2:B5.

In D2 enter =STEYX(B2:B5,A2:A5). The independently reconstructed residual sum of squares is 46058/931, approximately 49.471535983. Dividing by 4−2 and taking the square root gives approximately 4.973506609, which rounds to the documented 4.97351.

To reproduce that check in cells, put =SLOPE(B2:B5,A2:A5) in F2 and =INTERCEPT(B2:B5,A2:A5) in F3. Enter =(B2-($F$3+$F$2*A2))^2 in C2 and fill to C5. Put =SUM(C2:C5) in F5 and =SQRT(F5/2) in F6. F5 and F6 give the squared-error total and standard-error target above.

Keep y first, preserve row correspondence and inspect individual residuals before relying on one summary. A curved pattern or a single unusual observation can make a straight-line summary inadequate.

The short source does not fully specify missing-value handling, exact native errors, extreme precision limits or every small-sample boundary. Constant x values also prevent an ordinary unique slope fit. Do not assign error tokens by analogy with another product.

A smaller STEYX in the same response units indicates less residual spread on the supplied observations. Comparing different units, different observations or different model assumptions needs more care; the function does not establish causal reliability or future prediction coverage.

• SLOPE for the fitted rate of change.
• INTERCEPT for the fitted response at zero.
• FORECAST for a fitted point estimate.
• RSQ for a unitless squared-correlation summary.