FISHER in WPS Spreadsheet: Transform a Value with Fisher’s Log Ratio

=FISHER(x)

The required x is numeric and must lie strictly between −1 and 1. The function transforms one value; it does not calculate a correlation from paired observations. If your input is a correlation coefficient, calculate and justify that coefficient separately.

The official equation is z = 0.5 × ln((1+x)/(1−x)). This is also the inverse hyperbolic tangent. Its output is not restricted to −1 through 1 and is not a probability.

In a fresh worksheet, put Input in A1 and Fisher transform in B1. Enter:

Enter B2 and copy down. To check row 3 by hand, (1+0.2)/(1−0.2) = 1.5; half of ln(1.5) is approximately 0.2027325541. Reversing the input sign reverses the transformed sign, which explains the paired ±0.2 rows.

The last row demonstrates why treating the output as a correlation or probability is wrong: it exceeds 1 while the input remains inside the valid domain.

Consider inputs 0.2 and 0.8. Their raw difference is 0.6. Their transformed difference is approximately 0.8958797346, from 1.0986122887 − 0.2027325541. These are changes on different scales; neither should be labelled as the other.

For a mathematical reverse check, if z is the transformed value, (exp(2z)−1)/(exp(2z)+1) recovers x. At z = half ln(1.5), exp(2z) = 1.5, giving 0.5/2.5 = 0.2. This identity explains the transform without asserting an additional WPS function’s availability.

As x approaches 1 or −1, the transformed magnitude grows. Rounded source data that become exactly an endpoint therefore change a valid finite calculation into an invalid input. Retain the original numeric coefficient rather than retyping a heavily rounded display.

The lesson explicitly documents #NUM! when x ≤ −1 or x ≥ 1, including both endpoints. Accordingly =FISHER(1) and =FISHER(-1) are documented error expectations. It documents #VALUE! for nonnumeric input, such as =FISHER("bad").

The source uses broad language suggesting that the transformation generates a normal distribution. A deterministic transformation of one number does not establish normality, independence, sample size adequacy or the validity of a hypothesis test. The usual statistical use with correlations requires additional model assumptions; those assumptions and a complete test are not supplied by this one-argument function.

No p-value, confidence level or decision threshold is calculated here. Text coercion, missing data and extreme near-endpoint numerical accuracy are not established by these bounded examples.

The official WPS Academy FISHER lesson, published 2019-06-06, was rechecked on October 19, 2026. The lesson explicitly labels WPS Office for Windows Personal. The publication date is evidence that the function was documented by that date, not its introduction date. Exact numeric builds, first and last supported versions remain unknown. The lesson does not establish retirement or deprecation, or parity across platforms and editions.

These are original documentation-based examples with independently calculated expected values. They were not executed in WPS. Vendor demonstrations are historical documentation, not our application tests. Display rounding, client precision and workbook interchange remain unverified; the extra digits in this guide are mathematical reference values. English function names and comma separators follow the lesson.

• LN: evaluate a natural logarithm.
• EXP: evaluate the exponential used in the mathematical reverse identity.
• CORREL: calculate a correlation from paired data under its own contract.