LOGEST Function (OpenOffice Calc)

• Introduced: First release not established; documented by OpenOffice.org 3.3.
• Supported in: Apache OpenOffice 4.1.16; this is a verified release, not a final supported version.
• Scope: Desktop Calc, English function names and semicolon-separated formulas.
• Checked: October 14, 2026. Eight ordinary scalar or INDEX-selected reference formulas were recalculated from generated ODF in 4.1.16 on Linux x86-64, Build 9816. Source and independent arithmetic were checked.

LOGEST fits a linear least-squares model to LN(y) and returns exponential-model coefficients. For one predictor, the first result row is m then b in y=b*m^x. Optional statistics describe the regression in log space, not the original-unit squared errors. Release implementation.

LOGEST(KnownY[;KnownX[;Constant[;Statistics]]])

• KnownY: required strictly positive responses. KnownX: optional predictors; omitted values default to a simple 1,2,… sequence.
• Constant: optional TRUE; FALSE fixes the log intercept to zero, corresponding to b=1.
• Statistics: optional FALSE; TRUE requests five rows. The examples explicitly enable the intercept and statistics.

Enter the following numbers in their stated worksheet rows. K14 is unused; the prediction range is K10:K13.

GROWTH uses the positive values in L10:L14. The other regression examples use J10:J14. Keep each observed x-y pair aligned.

The native column contains saved 4.1.16 results; independent references use high-precision arithmetic and the stated source rules. Values are shown to twelve decimal places where needed. The range examples use the numeric setup above.

Using x=1,2,3,4,5 and positive y=5,8,12,14,18, the first row gives approximately m=1.366357374223 and b=4.072259457279. The fitted curve is b*m^x. Neither coefficient is the ordinary straight-line slope or intercept of y itself.

The native row-3 components give log-space R²≈0.960011906986 and residual standard error≈0.116312783848. Row 4 gives F≈72.022332246511 and degrees of freedom=3; row 5 gives explained and residual sums of squares approximately 0.974365910879 and 0.040585991059.

Let a and c be the intercept and slope from regressing LN(y) on x. The returned coefficients are b=EXP(a) and m=EXP(c). The source exponentiates that first coefficient row, but its statistics continue to use the transformed log response.

Thus residual standard error is sqrt(log-space residual SS/3), and R² is the explained log-space sum divided by the total log-space sum. Coefficient standard errors in the source’s second row are also for the underlying log regression; do not read them as unchanged standard errors of m and b. The native table selects the other eight reported components.

All known y values must be positive for the log transformation, and the shared matrix checker requires compatible numeric data. The native examples have a full-rank single predictor and nonzero log residuals. Invalid-domain, missing-value, text, rank-deficient and exact-fit cases were not executed.

The INDEX wrappers retrieve selected components; no GUI whole-matrix placement is claimed. A log fit and an original-scale fit optimize different errors. Exponentiating the fitted log value does not automatically resolve retransformation bias or create a prediction interval.

GROWTH, LINEST, ln.

• Function-specific release implementation
• CheckMatrix regression helper
• CalulateRGPRKP regression helper

The generated ODF produced the informational load warning “Not all attributes could be read.” The sheet was hard-recalculated and saved separately. All eight ordinary formula identities, inputs and selected results were reconciled; a stale SUM cache changed from 999999 to 3 and an uncached SUM returned 7. This establishes the stated examples, not every workbook style, invalid input or GUI whole-array entry. INDEX wrappers select individual returned components. Other explanatory identities are independent calculations. The authoring-tool XLSX was not executed in OpenOffice, and its preview values are not the native results reported here.