ONLYOFFICE FISHER: Correlation Transform, Domain and Worked Checks
Use ONLYOFFICE FISHER with its open input domain, logarithmic checks, near-endpoint examples and clear limits on statistical interpretation.
Verification status: See the evidence and test scope described in this guide. How formula examples are checked
FISHER transforms a number strictly between −1 and 1 onto the Fisher scale. It is often used with correlation coefficients, but the function itself does not calculate a correlation, choose a statistical model or return a p-value.
Availability and version history
- Introduced: The first introduction release and date are not established by the checked sources.
- Availability: Currently documented in ONLYOFFICE’s rolling function help; no final supported-version endpoint is established.
- Scope: Shared Spreadsheet Editor documentation, with formulas corroborated by a pinned official source-code revision. Individual client/build parity is unverified.
- Checked: October 14, 2026, against the official FISHER reference, source and vendor screenshot. Example outputs were independently calculated; these cases were not tested in the native application.
Syntax and the open interval
=FISHER(x)
The official reference requires one numeric input with −1 < x < 1. The endpoints −1 and 1 are outside that documented domain. There is no optional base, sample-size or confidence-level parameter.
The mathematical transformation is:
=0.5*LN((1+x)/(1-x))
ONLYOFFICE’s published implementation at a pinned revision uses this expression. That source corroborates the calculation; it does not identify the implementation in every installed build. LN is the natural logarithm, not base 10.
Example: compare FISHER with its logarithmic formula
Create FisherScale. Put Input x, Fisher value and Logarithmic check in A1:C1. Enter these inputs in A2:A7:
| Row | A: x | Expected B and C, approximately |
|---|---|---|
| 2 | -0.75 | -0.9729550745 |
| 3 | -0.5 | -0.5493061443 |
| 4 | 0 | 0 |
| 5 | 0.5 | 0.5493061443 |
| 6 | 0.75 | 0.9729550745 |
| 7 | 0.9 | 1.4722194896 |
In B2 enter =FISHER(A2). In C2 enter =0.5*LN((1+A2)/(1-A2)). Fill B2:C2 down through row 7.
The outputs keep the sign of the inputs and are not restricted to the original −1…1 interval. For example, x = 0.9 becomes approximately 1.4722. A transformed value greater than 1 is not a probability error: this is a different scale.
Put Round trip in D1, enter =FISHERINV(B2) in D2 and fill to D7. The mathematical targets are the original values −0.75,−0.5,0,0.5,0.75,0.9. Small displayed rounding differences should be distinguished from a different formula.
Why values near an endpoint spread apart
Keep the first table intact. Put x and Fisher value in F1:G1. Enter 0.9, 0.99 and 0.999 in F2:F4. Enter =FISHER(F2) in G2 and fill through G4.
| Input | Expected transformed value, approximately |
|---|---|
| 0.9 | 1.4722194896 |
| 0.99 | 2.6466524124 |
| 0.999 | 3.8002011673 |
The inputs get closer to 1 while the transformed values continue to increase. In the formula, 1-x approaches zero. Do not round a valid near-endpoint input to exactly 1 before evaluating it; that changes the input into an excluded boundary value.
Averaging before and after transformation gives different answers
Create TwoCorrelations, with 0.2 in A2 and 0.8 in A3. Put Correlation and Fisher value in A1:B1. Enter =FISHER(A2) in B2 and fill to B3.
- D2:
=AVERAGE(A2:A3)gives 0.5. - D3:
=FISHERINV(AVERAGE(B2:B3))gives approximately 0.5721224617.
This is an algebraic demonstration of a nonlinear transformation, not a recommendation to pool two studies with equal weight. A statistical aggregation needs assumptions about the observations, sample sizes and dependence. FISHER alone supplies none of those inputs.
Reproduce the vendor illustration
On VendorFisher, enter 0.56, −0.89 and 0.12 in A1:A3. Enter =FISHER(A1) in C1 and fill through C3.
| Input | Independent value, approximately | Vendor display |
|---|---|---|
| 0.56 | 0.6328331867 | 0.6328332 |
| -0.89 | -1.4219258711 | -1.4219259 |
| 0.12 | 0.1205810284 | 0.120581 |
These reproduce the numerical contents of the official screenshot.
Troubleshooting and limits
Check the input scale: enter a correlation of 0.8, not the number 80. Percentage formatting may show 80% while the stored numeric value is 0.8.
The help states the open input domain but does not enumerate every native error token, text-coercion case or extreme numerical boundary. The pinned implementation is evidence about one source revision, not a guarantee of identical rounding or array behavior in every product. These examples use ordinary scalar numbers and preserve full intermediate precision.