Foundations

Standard deviation: measuring how much values vary

Understand the concept of variability, differentiate between sample and population measures, and maintain a distinction between standard deviations and standard errors.

Spread in the original units

An SD measures how far apart each observation lies from the mean. If the SD is small, the observations will tend to lie closer to the mean. If the SD is large, there is more variation among the observations. If you have measurements in terms of minutes (waiting time), it follows that the SD will be expressed in minutes.

When calculating the SD, we square the differences (deviations) which gives larger weights to the larger deviations. Therefore, a single extreme value can significantly affect the result. Look at the individual values or look at a graph in conjunction with the number of values.

Population or sample?

For a complete population, average the squared deviations from its mean using N as the denominator, then take the square root. For a sample, divide the squared-deviation total by n minus 1, then take the square root. Here n is the sample size.

Using n minus 1 gives an unbiased estimate of population variance under independent, identically distributed sampling with finite variance. The sample SD itself is not exactly unbiased. At least two observations are needed for the usual sample SD.

Different from standard error

SD describes variation among observations. Standard error describes sampling uncertainty in an estimate. For an independent sample mean, its estimated standard error is s/√n. More observations may improve precision without reducing individual variability.

The rules that state approximately 68% of the observations fall within one SD apply to normally distributed populations only. These rules do not guarantee anything for skewed or heavily tailed distributions. Use medians and/or IQRs to provide alternative summaries.

Worked example: population or sample?

For values 4, 6 and 8, the average is 6. The square of the deviation from the average for all three values is 4, 0 and 4.

Population SD = √(8 / 3) ≈ 1.633. Sample SD = √(8 / 2) = 2.

Different denominators will give you different answers to different statistical questions. Choose which equation fits what type of question; i.e., if these are your whole population (entire), or just a sample.

SD describes how much individual observations vary. Standard error describes the sampling uncertainty of an estimate, such as a mean.

Sources for further reading