Standard Deviation and Measurement Units

Why Take the Square Root of Variance?

Variance measures how far the possible values of a random variable lie from its mean, but it does so using squared distances. That squaring is useful in mathematical calculations, yet it can make the result difficult to interpret: if a variable measures time in minutes, its variance is in square minutes. Taking the square root returns a measure of spread in minutes, the same units as the original variable.

Standard deviation is therefore a companion to variance. It preserves the information about spread while expressing its scale in more familiar units. The square root does not turn standard deviation into an exact distance that every outcome lies from the mean. Instead, it is a root-mean-square summary of deviations, whose meaning and limitations matter when interpreting it.

Definition and Units

Let $X$ be a discrete random variable with possible values $x$ and probability mass function $P(X=x)$. Assume its second moment is finite, so its variance is finite. Write its mean as $\mu=E(X)$. The variance is the probability-weighted average of squared deviations from the mean. Standard deviation is defined from that variance.

Definition: The standard deviation of $X$, denoted $\sigma_X$ (or $\sigma$ when the variable is clear), is the nonnegative square root of its variance: $$ \sigma_X=\sqrt{\operatorname{Var}(X)} =\sqrt{E[(X-\mu)^2]} =\sqrt{\sum_x (x-\mu)^2P(X=x)}. $$

Every term $(x-\mu)^2P(X=x)$ in the variance sum is nonnegative: a squared deviation cannot be negative, and probabilities are nonnegative. Thus $\operatorname{Var}(X)\geq 0$, and its nonnegative square root is defined. Standard deviation is also nonnegative. If the variable has units $u$, deviations have units $u$, squared deviations and variance have units $u^2$, and standard deviation has units $u$.

For example, if $X$ is measured in minutes, $\operatorname{Var}(X)$ is in minutes squared and $\sigma_X$ is in minutes. The unit is part of the interpretation: a standard deviation of 3.2 minutes and a variance of 3.2 minutes squared are not interchangeable statements.

Worked Examples

Worked Example: Standard Deviation of a Bus Count

Suppose a hypothetical probability model describes the number $X$ of buses that arrive late during a randomly selected morning. It assigns probabilities $P(X=0)=0.50$, $P(X=1)=0.30$, and $P(X=2)=0.20$. Find the mean, variance, and standard deviation, and interpret the standard deviation.

First, the probabilities are valid because $0.50+0.30+0.20=1.00$. The mean number of late buses is $$ \mu=E(X)=0(0.50)+1(0.30)+2(0.20)=0+0.30+0.40=0.70. $$

The variance is the expected squared deviation from $0.70$ late buses: $$ \begin{aligned} \operatorname{Var}(X) &=(0-0.70)^2(0.50)+(1-0.70)^2(0.30)+(2-0.70)^2(0.20)\\ &=0.49(0.50)+0.09(0.30)+1.69(0.20)\\ &=0.245+0.027+0.338\\ &=0.610\text{ buses}^2. \end{aligned} $$

Take the nonnegative square root: $$ \sigma_X=\sqrt{0.610\text{ buses}^2}=0.7810\text{ buses}. $$

In this model, the number of late buses typically varies around its mean of $0.70$ by a root-mean-square amount of about $0.781$ buses. The phrase “root-mean-square amount” is important: it does not claim that each morning’s count is exactly $0.781$ buses away from the mean, or that $0.781$ is the average absolute deviation.

Worked Example: Finding Standard Deviation From the Possible Scores

A hypothetical game awards a score $X$, in points, of 2, 4, or 8, with probabilities 0.25, 0.50, and 0.25, respectively. Calculate the standard deviation and explain its units.

The probabilities sum to $1$. First compute the mean: $$ \mu=2(0.25)+4(0.50)+8(0.25)=0.50+2.00+2.00=4.50\text{ points}. $$

Using the expected squared deviation, $$ \begin{aligned} \operatorname{Var}(X) &=(2-4.50)^2(0.25)+(4-4.50)^2(0.50)+(8-4.50)^2(0.25)\\ &=6.25(0.25)+0.25(0.50)+12.25(0.25)\\ &=1.5625+0.1250+3.0625\\ &=4.750\text{ points}^2. \end{aligned} $$

Therefore, $$ \sigma_X=\sqrt{4.750\text{ points}^2}=2.179\text{ points}. $$

The variance is $4.750$ points squared, while the standard deviation is about $2.179$ points. In context, $2.179$ points describes the scale of the scores’ deviations around their mean of $4.50$ points. It does not mean that every score is $2.179$ points from the mean; the possible scores are 2, 4, and 8.

Worked Example: Comparing Variance Units With Standard-Deviation Units

A hypothetical machine produces rods whose length $X$ has this discrete model: 10 cm with probability 0.25, 12 cm with probability 0.50, and 14 cm with probability 0.25. Find the variance and standard deviation, then state which quantity is more directly expressed in the measurement unit.

The probability total is $0.25+0.50+0.25=1$. The mean length is $$ \mu=10(0.25)+12(0.50)+14(0.25)=2.5+6+3.5=12\text{ cm}. $$

The deviations from the mean are $-2$, $0$, and $2$ cm. Squaring and weighting them gives $$ \begin{aligned} \operatorname{Var}(X) &=(10-12)^2(0.25)+(12-12)^2(0.50)+(14-12)^2(0.25)\\ &=4(0.25)+0(0.50)+4(0.25)\\ &=1+0+1=2\text{ cm}^2. \end{aligned} $$

Taking the square root, $$ \sigma_X=\sqrt{2\text{ cm}^2}=1.414\text{ cm}. $$

Both results describe spread, but $1.414$ cm is directly on the length scale used to measure the rods. The variance, $2\text{ cm}^2$, is a squared-unit quantity used in calculations. Reporting the variance as $2$ cm would be a units error.

Why Standard Deviation Is a Root-Mean-Square Deviation

The definition also gives a useful interpretation. Imagine selecting one value of $X$ according to its probability distribution and considering its distance from the mean. The signed deviation is $X-\mu$, but positive and negative deviations would cancel if simply averaged. Variance squares deviations first, so they cannot cancel. Standard deviation then takes the square root of that probability-weighted average of squares.

Measure deviations.
For each possible value $x$, calculate $x-\mu$, the signed distance from the mean.

Square and average.
Square each deviation and weight it by its probability. Their sum is $E[(X-\mu)^2]=\operatorname{Var}(X)$.

Return to the original scale.
Take the nonnegative square root of the variance. The result, $\sigma_X$, has the same units as $X$.

This reasoning explains both the usefulness and the limitation of standard deviation. Squaring gives more weight to larger deviations than smaller ones, while the final square root returns the summary to the original scale. However, $\sigma_X$ is not generally $E[|X-\mu|]$, the probability-weighted average absolute deviation. Those are different summaries because squaring and taking absolute values are different operations.

Result: Standard deviation is the square root of the expected squared deviation from the mean. It has the same units as the random variable, whereas variance has the square of those units. A standard deviation describes a scale of spread, not a guaranteed distance for each outcome.

Interpreting and Checking the Result

A standard deviation should be described together with the variable, its units, and its mean when the mean is relevant. For instance, “the standard deviation is $1.414$ cm” is more informative than “the spread is $1.414$,” because the former identifies the scale and unit. The interpretation should not imply that all observations fall within one standard deviation of the mean; the standard deviation alone does not establish such a bound for every distribution.

Several checks help catch errors:

Nonnegativity: variance and standard deviation cannot be negative.
Square-root check: squaring the reported standard deviation should recover the variance, apart from rounding.
Unit check: standard deviation has the variable’s units; variance has squared units.
Scale check: the answer should make sense in context. A standard deviation of 0.8 buses is a spread in a bus count, not in buses squared.
Probability check: for a discrete model, the probabilities used in the weighted sum must add to 1.

A zero standard deviation has a precise interpretation. Because each squared deviation is nonnegative, a variance of zero can occur only if every value with positive probability equals the mean. In that case, the random variable has no variation under the model. Conversely, if every possible value with positive probability is the same, all deviations from the mean are zero, so the variance and standard deviation are both zero.

Common Mistakes

Reporting the variance as if it were standard deviation. If variance is $4.750\text{ points}^2$, standard deviation is $\sqrt{4.750}=2.179$ points, not $4.750$ points.
Forgetting the square root. The variance calculation ends with a weighted sum of squared deviations. Standard deviation requires one additional step: take the nonnegative square root.
Calling standard deviation the average absolute distance. It is a root-mean-square deviation. The average absolute distance uses absolute deviations instead of squared deviations and is generally a different quantity.
Assuming every outcome is one standard deviation from the mean. Standard deviation summarizes overall spread; individual outcomes can be closer to or farther from the mean.

A useful correction to the first mistake is to write the units beside each quantity as the calculation proceeds. If the variable is measured in minutes, the variance line should end in minutes squared, and taking its square root should give minutes. The units expose whether the calculation has stopped at variance or reached standard deviation.

Check Your Understanding

Answer each question using the definitions and unit reasoning from this tutorial.

A discrete random variable has variance $9\text{ kg}^2$. What is its standard deviation, including units?
Why can the standard deviation never be negative?
A variable records waiting time in seconds. State the units of its variance and standard deviation.
Explain why a standard deviation of 2.4 units does not mean every possible value is exactly 2.4 units from the mean.
Can the variance be zero when two different values both have positive probability? Explain.