Square Root and Inverse Transform Calculator

StatisticsLast updated: August 22, 2026

Tukey's ladder of powers arranges the transformations used to pull a skewed variable toward symmetry into a single ordered list: square (p = 2), raw data (p = 1), square root (p = 1/2), cube root (p = 1/3), logarithm (p = 0), inverse square root (p = −1/2), and inverse transformation (p = −1). The further down the ladder you go, the more sharply large values are pulled in; the stronger the right skew, the further down you have to move.

This calculator applies all seven rungs of the ladder to your data set, computes the skewness and kurtosis of each one, and recommends the rung with the smallest absolute skewness as the best transformation. The comparison table shows at a glance how much each transformation improves matters, so choosing a transformation becomes a decision based on numbers rather than trial and error.

All values must be positive (the logarithm and the inverse transformation are undefined at zero). Separate them with commas, spaces, or line breaks.
The comparison table always shows every rung; this choice decides which transformation the histogram and the step-by-step solution refer to.

Tukey's Ladder of Powers

p = 2      y = x²           ↑ corrects left skewness
p = 1      y = x            ─ raw data (no transformation)
p = 1/2    y = √x           ↓ mild right skewness
p = 1/3    y = ∛x           ↓ moderate right skewness
p = 0      y = ln(x)        ↓ strong right skewness
p = −1/2   y = −1/√x        ↓ very strong right skewness
p = −1     y = −1/x         ↓ strongest rung of the ladder
General form:  y = (xᵖ − 1) / p   (p ≠ 0),   y = ln(x)  (p = 0)
Criterion:     g₁ = [n / ((n−1)(n−2))] · Σ((yᵢ − ȳ) / s_y)³ → smallest |g₁|

The minus sign placed in front of the negative-exponent transformations preserves the order of the observations; without it, the largest value would be moved to the smallest position. The general form (xᵖ − 1)/p is the Box-Cox transformation restricted to the rungs of the ladder.

How to Calculate

  1. Paste a data set of positive values into the box; at least 5 observations are required, because skewness is not reliable in smaller samples.
  2. Look at the raw skewness in the results card: if it is positive you need to move down the ladder, if negative, up.
  3. Read the skewness and kurtosis of all seven rungs side by side in the comparison table; the ★ marks the rung with the smallest absolute skewness.
  4. If you wish, pick a specific transformation from the 'Transformation to examine' list and inspect its histogram and improvement percentage separately.
  5. Watch for the region of the absolute-skewness chart where the rungs are close together: when two rungs are nearly tied, choose the one that is easier to interpret (usually the square root or the logarithm).
  6. State the transformation you chose in your report together with its formula, and use the inverse transformation when converting results back to the raw unit.

Worked Examples

Right-skewed count data

The raw skewness of the 14 observations is 0.973; the best rung of the ladder is the square root (√x), which brings skewness down to 0.090 for an improvement of 90.8%. One rung further down, the cube root gives -0.281 and the logarithm -1.142, overcorrecting; kurtosis also falls from 0.697 to -0.226.

Sample size (n): 14 · Raw skewness (g₁): 0.973 · Recommended transformation: Square root — y = √x

Examining the square root transformation by hand

Raw skewness is 1.164, and the manually selected square root transformation lowers it to 0.572, an improvement of 50.9%. On the same data the best rung of the ladder is the logarithm: it moves skewness to -0.155 for an improvement of 86.7%, while the cube root stops at 0.345.

Sample size (n): 12 · Raw skewness (g₁): 1.164 · Recommended transformation: Logarithm — y = ln(x)

Symmetric data that needs no transformation

In this perfectly symmetric series the raw skewness is 0.000, and no rung of the ladder can improve on it — the square root gives -0.072 and the logarithm -0.144, breaking the symmetry. The calculator therefore recommends the 'No transformation (y = x)' rung, and the warning line advises staying on the raw scale.

Sample size (n): 10 · Raw skewness (g₁): 0.000 · Recommended transformation: No transformation — y = x

Frequently Asked Questions

What is Tukey's ladder of powers?
Tukey's ladder of powers is an ordering of transformations of the form x^p by their exponent p: p = 1 is the raw data, p = 1/2 the square root, p = 0 the logarithm, and p = −1 the inverse transformation. The smaller p becomes, the more strongly large values are compressed. The practical rule is simple: move down the ladder for right-skewed data and up for left-skewed data, continuing until the skewness reaches an acceptable level.
When is the square root transformation preferred?
It is the first choice for count data and for Poisson-type variables whose variance grows with the mean; the √x transformation makes the variance approximately independent of the mean. It is also gentler than the logarithm, so it suffices for mild and moderate skewness. For counts that include zeros, corrected versions such as √(x + 0.5) or √x + √(x+1) are used.
Why is −1/x used instead of 1/x in the inverse transformation?
The 1/x transformation reverses the ordering: the largest observation is moved to the smallest value, which flips the sign of the skewness and the whole interpretation with it. Putting a minus sign in front (−1/x) preserves the ordering, so the transformed variable increases in the same direction as the raw one. The magnitudes are identical; only the direction is corrected.
Is choosing the transformation with the smallest skewness always right?
No. Skewness is only one criterion; kurtosis, the behavior of outliers, and the distribution of the residuals should all be assessed too. Interpretability matters as well: on a logarithmic scale coefficients read as percentage change, whereas an inverse transformation yields something like 'amount per unit', which is unfamiliar in most fields. When two rungs have similar skewness, prefer the one that is easier to interpret.
How does this differ from the Box-Cox transformation?
Box-Cox uses the same family of powers but estimates the exponent p (λ) as a continuous value that maximizes the likelihood, rather than picking it from fixed rungs; it can land on intermediate values such as λ = 0.27. Tukey's ladder works with interpretable round rungs instead. In practice the λ found by Box-Cox is usually rounded to the nearest rung of the ladder for reporting.
What should I do with data that contains zeros or negative values?
The square root is defined at zero, but the logarithm and the inverse transformation are not; with negative values even the square root is undefined. The standard solution is to add a constant to the whole series: x + 1 or x + |min(x)| + 1. The cube root is defined for negative values as well and can be used without adding a constant; the Yeo-Johnson transformation solves the same problem as an extension of Box-Cox.
How are these transformations applied in R and SPSS?
In R you write sqrt(x), x^(1/3), log(x), and -1/x directly; the boxcox() function of the MASS package estimates the best exponent, and powerTransform() from the car package does the same with Yeo-Johnson support. In SPSS, use the expressions SQRT(x), x**(1/3), and -1/x in the Transform > Compute Variable dialog. In both programs you should recompute and compare the skewness after the transformation.