LOGEST in ONLYOFFICE Spreadsheet Editor

• Introduced: Original introduction release is not established by the checked sources.
• Available: Listed in current unversioned ONLYOFFICE help, checked October 18, 2026.
• Implementation inspected: Official SDKJS v10.0.0.169, commit 3679e0375a758d7dc59d31337a45461be3525ed8. The snapshot does not prove identical behavior in every installed editor.
• Retired or deprecated: These sources do not establish retirement or formal deprecation.
• Testing: Numerical examples are independent mathematical expectations, and implementation discrepancies are labeled source predictions. No ONLYOFFICE application or calculation-engine execution was performed.

The examples use English function names and commas. Use your locale’s argument separator if different. Retain full precision in formulas; round only the displayed result. Nested output grids describe source-derived array contents. Native spill placement, obstruction errors, array-entry shortcuts and recalculation behavior have not been tested in the application. Reserve a separate output area and follow the array-formula entry method supported by your installed editor; do not assume a particular keyboard shortcut or automatic spill behavior.

=LOGEST(known_y, [known_x], [const], [stats])

One to four arguments are registered. LOGEST fits LN(y)=LN(b)+x*LN(m) for one predictor, then exponentiates the coefficient row to return m and b for y=b*m^x. It does not fit an ordinary linear model directly to the original response values.

Omitted known_x uses 1,2,…,N. TRUE or omitted const estimates b; FALSE fixes b=1, corresponding to zero intercept in log space. FALSE or omitted stats returns one coefficient row. TRUE requests a five-row statistics array. With K predictors, the first row is mK,…,m1,b: the multiplier order is reversed relative to the known-x predictor columns or rows.

For multiple predictors, the original-scale equation is y=b*m1^x1*m2^x2*.... The returned values are multiplicative bases, not additive slopes or percentages. A multiplier 2 means doubling per x unit; a multiplier 1.02 means a 2% multiplicative increase per x unit.

Put x=0,1,2 in A1:A3 and y=3,6,12 in B1:B3. Evaluate =LOGEST(B1:B3,A1:A3,TRUE,FALSE) in a separate result area. The independently expected row is 2,3.

The logarithms form the exact line LN(y)=LN(3)+x*LN(2). Exponentiating its slope gives 2 and exponentiating its intercept gives 3. Reconstruct a fitted value as =3*2^2, which is 12, or a prediction at x=3 as =3*2^3, which is 24. GROWTH performs such prediction directly without returning the coefficients.

If known_x is omitted, the implied x values are 1,2,3. =LOGEST(B1:B3) then has independent expectation 2,1.5, because 1.5*2^1=3. The fitted training responses can match while the meaning of b changes with the x origin.

With const FALSE on the explicit x=0,1,2 data, the independent multiplier is 2*3^(3/5) ≈ 3.866364089864 and b is 1. Thus =LOGEST(B1:B3,A1:A3,FALSE,FALSE) predicts 3.866364089864,1. The multiplier is refitted to the constraint; the unconstrained multiplier 2 is not simply retained.

For a nonperfect, independently checkable example, use x=1,2,3,4 in A1:A4 and enter =EXP(3), =EXP(5), =EXP(6), =EXP(9) in B1:B4. The log responses are 3,5,6,9. Evaluate =LOGEST(B1:B4,A1:A4,TRUE,TRUE).

The inspected source exponentiates the coefficient row, not the entire statistics array. Row 2 is therefore not an original-scale standard error of m or b. Adding 0.264575 directly to the returned multiplier does not form a justified confidence limit.

The arithmetic check uses centered x-square sum 5, centered cross-product 9.5 and log-residual sum of squares 0.7. With df=2, residual mean square is 0.35. Standard errors are sqrt(0.07) and sqrt(0.525), just as for the corresponding linear regression on log y. The function does not itself return confidence intervals or an original-scale prediction interval.

On a fresh sheet enter predictor pairs (0,0), (1,0), (0,1), (1,1) in A1:B4 and responses 2,6,10,30 in C1:C4. They follow y=2*3^x1*5^x2. =LOGEST(C1:C4,A1:B4,TRUE,FALSE) has independent expectation 5,3,2, reversing the two predictor multipliers before b.

For responses arranged as a column, observations run down rows and predictors occupy columns. For a response row, predictors occupy rows. In stats mode the output has five rows and K+1 columns; unused positions beyond the first two columns in rows 3–5 are source-predicted #N/A placeholders. Do not read them as missing estimated coefficients.

Zero or negative known_y values predict internal not_numeric, conventionally #NUM!. Shape mismatches predict bad_reference (#REF!) before element-type checks; text or blank elements in compatible matrices predict wrong_value_type (#VALUE!). Use paired numeric observations and positive responses rather than guessing that invalid rows will be skipped.

The shared simple-regression helper substitutes 1 for a zero x-square denominator. Identical x values with a constant can therefore produce unit multiplier and geometric-mean b, despite an unidentified growth rate. Its statistics branch also treats zero regression sum of squares as exact fit and can overwrite residual sum of squares with 0 and R² with 1. Those degenerate statistics must not be taken as valid goodness-of-fit evidence.

Exact-fit and zero-degree-of-freedom branches can produce zero standard errors and an unavailable F statistic. Multiple-predictor singularity may lead to #VALUE! or numerical instability; reliable rank reduction is not guaranteed. Inspect the design rather than treating successful numeric output as proof that parameters are identifiable.

The fit minimizes squared log residuals. Exponentiating a fitted log response is not generally an unbiased estimate of the arithmetic mean y, and no general bias correction is applied. Check the model, residual behavior and extrapolation range before interpreting a growth multiplier as a stable future relationship.

Recalculate representative formulas when moving a workbook between editors. Matching names do not guarantee identical argument coercion, array behavior, numerical precision or error handling. These source-derived examples do not establish automatic spill behavior or a general import/export guarantee.

• LN
• EXP
• GEOMEAN
• RSQ

• ONLYOFFICE LOGEST help, checked October 18, 2026.
• Official SDKJS implementation, 2617–3169, tag v10.0.0.169 and exact commit above.