Variability vs Standard Deviation

Variability is how much values differ, and the word covers two things that are measured with two different numbers. Inside one data set it is the spread of the observations, reported by the standard deviation $s$ or the interquartile range. Across repeated samples from the same population it is sampling variability, the bounce in a statistic such as $\bar{x}$ (x-bar) from one sample to the next, reported by a standard error.

Two classes both average 75 on a test. Class A scored 71, 73, 75, 77, 79 and Class B scored 55, 65, 75, 85, 95. The means are identical, $375/5 = 75$ in each. The sums of squared deviations are 40 and 1000, so $s = \sqrt{40/4} = 3.16$ against $s = \sqrt{1000/4} = 15.81$. A center on its own describes neither class, which is why a description always names variability alongside it.

The second kind shrinks with sample size. If individual measurements have $\sigma = 12$, the sample mean from $n = 4$ has standard deviation $12/\sqrt{4} = 6$, and from $n = 36$ it is $12/\sqrt{36} = 2$. Nine times the data, one third of the bounce.

That is where "a bigger sample has less variability" goes wrong. A larger sample does not make the individuals more alike, and $\sigma$ is a property of the population that no sample size touches. What shrinks is the variability of the statistic computed from the sample. Say which of the two you mean, because the first claim is false and the second is the foundation under every confidence interval in the course.

Low variability is also not accuracy. An estimator can land in nearly the same place every time and still land in the wrong place every time, which is bias rather than variability, and a bigger sample does not fix it. Variability is the width of the pattern on the target; bias is where the pattern is centered. Both have to be small before an estimate is worth anything.

Standard deviation reports spread as a typical distance between a value and the mean, carried in the units of the data. For a sample it is $s = \sqrt{\frac{\sum (x_i - \bar{x})^2}{n-1}}$, where $x_i$ are the observations, $\bar{x}$ (x-bar) is the sample mean, and $n$ is how many values there are. For a whole population the symbol becomes $\sigma$ (sigma), the mean becomes $\mu$ (mu), and the divisor is $n$ instead of $n - 1$. The word typical is loose on purpose: $s$ is the square root of an average squared distance, not the average of the distances.

Take the five values 4, 8, 11, 13, 14. The mean is $50/5 = 10$, so the deviations are -6, -2, 1, 3, and 4. Squared they are 36, 4, 1, 9, and 16, which sum to 66. Divide by $n - 1 = 4$ for the variance $s^2 = 16.5$, then take the square root: $s = 4.06$. Treating the same five numbers as a whole population instead gives $\sigma = \sqrt{66/5} = 3.63$, so settle that question before starting.

The sentence to unlearn is "the standard deviation is the average distance from the mean." For those five values the average distance really is $(6 + 2 + 1 + 3 + 4)/5 = 3.2$, which is the mean absolute deviation, and it is not 4.06. Squaring before averaging gives far-out values more weight, so unless every value sits the same distance out, $s$ lands above the plain average distance.

A standard deviation is never negative, and it equals 0 in exactly one case: every value in the set is identical, so every deviation is 0. It is also not resistant. Change that 14 to a 44 and the median stays at 11 while $s$ goes from 4.06 to 16.02, because one distance of 28 becomes 784 inside the sum and swamps the other four.

Report it with the variable and the units attached, never as a bare number. In the Fall 2026 course this sits in Unit 1, whose topic 1.7 is titled Summary Statistics for One Quantitative Variable.

Finance uses this exact statistic as its measure of risk. The standard deviation of an investment's period returns is computed by the same steps used here and is the number quoted as an asset's risk: standard deviation of returns.