How this instrument works
Root mean square works in the order its name gives you, read right to left in practice: square every value, find the mean of the three squares, and finally pull a square root out of that mean. Squaring first is the whole trick. It turns every input positive before anything gets averaged, so a value of −3 contributes exactly as much as a value of 3, and the closing root brings the answer back down to the original units instead of leaving you with squared ones.
This construction places RMS among the four classical Pythagorean means — harmonic, geometric, arithmetic, and quadratic (RMS is the quadratic one) — and it is provably the largest of the four whenever the inputs are not all identical, with equality only when every value already matches. Feed the formula three equal numbers and it hands back that same number unchanged; feed it three unequal ones and it always reports something above their plain arithmetic average, never below.
The reason RMS gets reached for instead of a plain mean, whenever error or deviation is being scored, is that squaring exaggerates whichever entry sits farthest from typical, and rooting afterward restores the original units. Forecasters run this identical construction over the gap between a prediction and reality and call it root-mean-square error, RMSE: a model with two tiny misses and one huge one earns a worse RMSE than a model with three medium misses totalling the same absolute error, because that one oversized term dominates a sum of squares far more than it would dominate a plain sum.
- Type your first reading into Value 1 — a plus sign, a minus sign, or zero all work fine.
- Type the second reading into Value 2.
- Type the third reading into Value 3; order never changes the result, since squaring erases sign and addition doesn't care about sequence.
- Read Root mean square for the squared-averaged-and-rooted answer, always at or above the plain average's absolute size.
Worked example — the RMS of 3, 4, and 5
Put 3 in Value 1, 4 in Value 2, and 5 in Value 3. Squaring gives 9, 16, and 25; summing gives 50; dividing by 3 gives 16.666667; and its root, 4.08248290463863, is the figure this sheet returns to full precision. Notice it sits above the plain arithmetic mean of 3, 4, and 5, which is exactly 4, because the larger value, 5, pulls harder on a sum of squares than it would on a sum of plain values.
That gap between 4 and 4.082 is small here because 3, 4, and 5 are fairly close together, but it grows with spread. Replace the 3 with a −3 and the plain average collapses toward 2, while the RMS barely moves, because squaring never lets the negative sign shrink the total the way it shrinks a plain sum.
Questions
How is root mean square different from a plain average?
A plain average sums values and divides by the count; RMS squares each value first, divides by the count, then takes a square root. Squaring removes sign before anything is combined, so negative and positive values reinforce the result instead of cancelling. For 3, 4, and 5 the two methods land close — 4 versus 4.082 — but for values that include cancelling negatives, such as −3, 3, and 0, the plain average is 0 while the RMS is about 2.449.
Why square the values instead of just taking absolute values first?
Squaring is differentiable everywhere, which is why calculus-based methods like least squares are built on sums of squares rather than sums of absolute values — Gauss and Legendre chose squared error specifically because it is easy to minimize with algebra. Absolute value gives a related but different quantity, the mean absolute deviation, which behaves differently under further averaging and is harder to work with analytically.
Can the root mean square ever be smaller than the plain average?
No, not for real numbers. RMS sits at or above a simple average taken from those same three magnitudes, meeting it only once every input already shares one common size — three copies of 5 return a simple average and an RMS that are both exactly 5. Any spread among the three entries forces RMS above that simple average, never below it.
Does this give the same number as a standard deviation?
Related but not identical. Standard deviation is the RMS of each value's deviation from the group's own mean, not the RMS of the raw values themselves. This sheet computes RMS directly on whatever three numbers you enter; subtract their average from each one first, then run the same three numbers through this formula, and you get the standard deviation instead.
What happens if all three values are zero?
The result is zero, the one case where RMS matches the plain average exactly regardless of sign considerations, since 0 squared is still 0. It is also the only input where RMS itself equals zero — for any set containing a nonzero value, squaring guarantees a positive sum inside the root, so the result is strictly positive.
Where does RMS actually get used outside a statistics class?
Household AC voltage ratings — 120 V or 230 V — are RMS figures, chosen because RMS voltage tells you the equivalent steady voltage that would deliver the same heating power. Audio equipment rates amplifier output in RMS watts for the same reason, and error metrics in forecasting and signal processing, such as RMSE, apply the identical squaring-averaging-rooting pattern to prediction errors instead of raw measurements.