How this instrument works
Skewness measures the asymmetry of a distribution around its mean. A distribution with skewness of 0 is perfectly symmetric — its left and right sides mirror each other, like a classic bell curve. Positive skewness means the distribution has a longer or fatter tail stretching to the right, with most values clustered on the lower end and a few large values pulling the tail out (common in income data, for instance). Negative skewness means the opposite: a longer tail stretching to the left, with most values clustered high and a few small values pulling the tail down.
This calculator uses the Fisher-Pearson coefficient of skewness, g1 = m3 / m2^1.5, where m2 and m3 are the second and third sample moments about the mean (the average squared and cubed deviation from the mean, respectively). Cubing the deviations — rather than squaring, as variance does — is what makes the formula sensitive to direction: symmetric deviations above and below the mean cancel out in the sum, while an asymmetric distribution leaves a nonzero residual whose sign points toward the longer tail.
Skewness matters practically because many common statistical methods (like standard confidence intervals and t-tests) assume roughly symmetric, normally-shaped data — strongly skewed data can make those methods less reliable or push you toward the mean's less-representative sibling. Because it's a single unitless number, skewness also gives a much faster gut-check than eyeballing a histogram, especially when you're screening many variables at once.
- Enter your dataset into Data set — numbers separated by commas, spaces or new lines — at least 3 numbers are required.
- Read Skewness (Fisher-Pearson coefficient, g1) — zero means symmetric, positive means a longer right tail, negative means a longer left tail.
- Judge the size, not just the sign — values roughly between -0.5 and 0.5 are often considered fairly symmetric in practice, while larger magnitudes indicate more pronounced skew, though there's no single universal cutoff.
Worked example — the dataset 2, 4, 4, 4, 5, 5, 7, 9
Enter 2, 4, 4, 4, 5, 5, 7, 9 into Data set — numbers separated by commas, spaces or new lines. This dataset has a mean of 5, with most values clustered between 2 and 5 but a couple of higher values (7 and 9) stretching out to the right.
Skewness (Fisher-Pearson coefficient, g1) reads 0.656250 — a positive value, confirming that visual impression: the distribution has a mild right-leaning tail. The result is cross-checked against scipy.stats.skew(bias=True), the standard Python reference implementation, and matches exactly for this dataset.
Questions
What does a skewness of 0 mean?
It means the distribution is perfectly symmetric around its mean — the left and right sides mirror each other exactly. This happens for genuinely symmetric distributions like a normal (bell-shaped) curve, and also, by construction, for any dataset that's symmetric around its own mean, like 1, 2, 3, 4, 5, where the cubed deviations above and below the mean cancel out exactly.
How do I know if skewness is positive or negative just by looking at data?
Positive skewness means the tail stretches to the right — most values are bunched on the lower end, with a few unusually large values pulling the average up and creating that long right tail. Negative skewness is the mirror image: most values bunched high, with a few unusually small values dragging a tail out to the left. A quick way to check: if the mean is noticeably larger than the median, skewness is usually positive; if the mean is noticeably smaller than the median, skewness is usually negative.
How large does skewness need to be before it's a concern?
There's no single universal threshold, but a common rule of thumb treats values roughly within ±0.5 as fairly symmetric, values between ±0.5 and ±1 as moderately skewed, and values beyond ±1 as highly skewed. What counts as 'too skewed' really depends on what you're using the data for — methods that assume near-normal data (like many t-tests or confidence intervals) become less reliable as skewness grows larger.
Why does the formula cube the deviations instead of squaring them?
Squaring (as in variance) always produces a positive number regardless of whether a value is above or below the mean, which is exactly why variance can't detect asymmetry — it only measures spread. Cubing preserves the sign: deviations above the mean stay positive when cubed, deviations below stay negative. In a symmetric distribution those positive and negative cubed deviations cancel out to zero; in an asymmetric one they don't, and the sign of what's left over reveals which direction the longer tail points.
Do I need at least 3 data points to compute skewness?
Yes — you need at least 3 points, and in practice quite a few more than that to get a stable, meaningful estimate. Skewness computed from a very small sample can be noisy and swing wildly with just one or two additional data points, so treat skewness estimates from small samples (under roughly 20-30 points) with some caution.