How this instrument works
Chebyshev's theorem states that for any data set — no matter its shape, skewed, bimodal, or anything else — at least 1 - 1/k² of the values must fall within k standard deviations of the mean, for any k greater than 1. That 'any distribution' clause is what makes it different from the empirical rule (the 68-95-99.7 rule), which gives tighter, more useful percentages but only works when the data is at least roughly normal (bell-shaped).
The trade-off is precision for generality. At k = 2, Chebyshev's theorem only guarantees at least 75% of values fall within 2 standard deviations, whereas the empirical rule promises about 95% — a much stronger claim, but one that requires the normal-distribution assumption to actually hold. Chebyshev's 75% is a worst-case floor: many real distributions comfortably beat it, but none can ever fall below it, which is what makes the bound provably true for every possible distribution with a finite variance.
This makes Chebyshev's theorem the right tool specifically when you don't know, or can't assume, the shape of your data — you still know something guaranteed, even in total ignorance of the distribution's shape. It shows up in quality control and risk analysis for exactly this reason: a manufacturer who can't be sure a process output is normally distributed can still state a defensible, guaranteed minimum for how much output falls within a given tolerance band.
- Enter how many standard deviations out you want to check into k — number of standard deviations. It must be greater than 1.
- Read Minimum % of data within k standard deviations (any distribution) for the guaranteed floor — the true percentage for your actual data will be at or above this figure.
- k doesn't have to be a whole number — try a value like 1.5 to see the bound at a fractional distance from the mean.
- If you know (or are willing to assume) your data is roughly normal, this site's empirical-rule instrument gives the tighter, more informative 68/95/99.7 percentages instead.
Worked example — the guaranteed floor at k = 2
Enter 2 into k — number of standard deviations. The instrument computes 1 - 1/2² = 1 - 0.25 = 0.75, so Minimum % of data within k standard deviations (any distribution) reads 75.0%.
That means: whatever your data actually looks like — symmetric, lopsided, with two humps, anything — at least 75% of its values are guaranteed to fall within 2 standard deviations of the mean. Contrast that with the empirical rule's roughly 95% at the same distance: the empirical rule's number is bigger and more useful, but only trustworthy once you've confirmed (or are willing to assume) the data is close to normally shaped. Chebyshev's 75% needs no such assumption.
Questions
How is Chebyshev's theorem different from the empirical rule?
The empirical rule (68-95-99.7) gives sharper, more precise percentages but only applies to roughly normal, bell-shaped data. Chebyshev's theorem gives a looser, weaker guarantee — at least 75% within 2 standard deviations, versus the empirical rule's ~95% — but that guarantee holds for absolutely any distribution, skewed or not, because it doesn't assume any particular shape at all. Use the empirical rule when you know your data is roughly normal, and Chebyshev's theorem when you don't or can't assume that.
Why must k be greater than 1?
At k = 1, the formula gives 1 - 1/1² = 0, a guarantee of 'at least 0%,' which is true but says nothing useful. For k less than 1, the bound would go negative, which is meaningless as a percentage. Chebyshev's theorem only produces a genuinely informative floor once k exceeds 1, which is why the instrument requires it.
Does 'at least' mean the real percentage could be higher?
Yes, almost always. Chebyshev's theorem gives a worst-case guaranteed minimum, not the exact figure — most real data sets, especially anything close to symmetric, comfortably exceed the Chebyshev floor. A normally distributed data set, for instance, actually has about 95% of values within 2 standard deviations, well above Chebyshev's guaranteed 75% minimum for that same k. The bound only becomes tight (nearly exact) for specially constructed, worst-case distributions.
Can k be a non-integer value like 1.5?
Yes — the formula works for any real number k greater than 1. At k = 1.5, the guaranteed minimum works out to 1 - 1/1.5² = 1 - 0.4444 ≈ 55.6% of values falling within 1.5 standard deviations of the mean, for any distribution.
Why does this bound work for absolutely any distribution?
It follows directly from how the standard deviation is defined, without needing any assumption about the data's shape. Roughly: if too much of the data sat far from the mean, the average squared distance from the mean (the variance) would necessarily be larger than it actually is — so a large fraction of the data being more than k standard deviations away would contradict the very definition of the standard deviation being computed from that data. That self-consistency argument holds regardless of shape, which is why the guarantee is universal.
What's a practical use for this outside a statistics class?
It's useful whenever you need a defensible, worst-case guarantee about spread but can't verify or assume normality — for example, quality control on a manufacturing process with an unknown or unverified output distribution, or setting a conservative risk tolerance band when the shape of a data set is genuinely unknown. Chebyshev's theorem lets you state 'at least X% will fall in this range' with certainty, even in total ignorance of the distribution's shape.