What Is a Good Standard Deviation? 4 Checks for High or Low

There’s no number that counts as a good standard deviation on its own. A standard deviation of 2 can be tiny or huge, depending on what you measured, how big the values are and what you need the data to do. Instead of looking for a universal cutoff, check it against four things: the size of the mean, the range your data could possibly take, any limit or requirement you have to meet, and how much data you have.

This guide runs through each check with one example: eight bags of coffee from a filling machine that’s meant to put 250 g in every bag.

Standard Deviation Calculator

Separate values with commas, spaces or new lines. Paste a spreadsheet column straight in. For repeats, 7x3 means 7 three times.

Sample standard deviation–
Step by step

    The bags weigh 248, 252, 250, 251, 249, 250, 247 and 253 g. The mean is 250 g and the sample standard deviation is 2 g. Is that good?

    Check 1: Compare It With the Mean

    A spread of 2 g means one thing for 250 g bags and something very different for 5 g sachets. To compare spread with size, divide the standard deviation by the mean. This is the coefficient of variation (CV):

    2 ÷ 250 = 0.008, or 0.8%

    On this measure, the machine looks very consistent: a typical bag is within 1% of the average.

    You’ll often see rules of thumb such as “a CV above 1 is high” or “below 10% is good.” None of them is universal, and they contradict each other, because acceptable variation depends on the field. Some fields do publish their own limits. In US drug testing, for example, FDA reviewers expect the precision of a chromatographic lab method to have a CV of no more than 15%, or 20% at the lowest measurable concentration. Use a limit like that when your field has one, and otherwise treat CV as a way to compare similar datasets rather than a pass/fail test.

    The CV only works for data measured from a true zero, such as weights, times or prices. It means nothing for temperatures in Celsius or for scores that can go negative.

    Check 2: Compare It With the Possible Range

    When values can only fall between two limits, the standard deviation has a ceiling. The largest possible population standard deviation is half the range, which happens when half the values sit at each end.

    • On a 1–5 rating scale, the maximum is 2. An SD of 1.2 is 60% of the way to complete disagreement.
    • On a 0–100 test, the maximum is 50. An SD of 15 means scores are spread out, but nowhere near the limit.

    A rough check also works for unbounded data: for a moderate number of fairly symmetrical values, the standard deviation is often about a quarter of the range. The coffee bags range from 247 to 253 g, a spread of 6 g, which suggests about 1.5 g. The real value, 2 g, is in the same ballpark, so nothing unusual is going on. If the standard deviation were far larger than a quarter of the range, it would be worth checking your calculation. If it were far smaller, most of the range might come from one or two extreme values.

    Check 3: Compare It With What You Need

    This is often the check that decides the question. Suppose the coffee bags must weigh between 245 g and 255 g. Manufacturers compare that allowed width with six standard deviations, the spread that contains nearly all of a stable, bell-shaped process. The ratio is called Cp:

    Cp = (255 − 245) ÷ (6 × 2) = 10 ÷ 12 ≈ 0.83

    A common target in manufacturing is a Cp of at least 1.33, and some companies ask for 1.67 or 2. At 0.83, the machine’s natural spread is wider than the allowed range, so some bags will fall outside the limits even when the average is spot on.

    So the same 2 g standard deviation that looked excellent in Check 1 (a CV of 0.8%) isn’t good enough here. The machine would need a standard deviation of about 1.25 g to reach a Cp of 1.33. Relative spread tells you how consistent something is. Only a requirement tells you whether that’s consistent enough.

    The same idea applies outside factories. If a bus is scheduled every 10 minutes, a 1-minute standard deviation in arrival times is fine. If you have a 2-minute window to catch a connection, it isn’t.

    Check 4: How Much Data Is Behind It?

    A standard deviation from a handful of values is a rough estimate. If you took another eight coffee bags, you’d probably get a different answer. For roughly bell-shaped data, the true standard deviation could plausibly be this far either side of your sample value (95% confidence):

    Values in your sampleTrue SD likely between
    50.60 × and 2.87 × your sample SD
    100.69 × and 1.83 ×
    300.80 × and 1.34 ×
    1000.88 × and 1.16 ×

    With 8 bags and a sample SD of 2 g, the machine’s true standard deviation could easily be anywhere from about 1.3 g to 4 g. Before you decide a process is consistent, or compare two standard deviations that are close together, collect enough data. A few dozen values give a far more reliable picture than five.

    When a High Standard Deviation Is What You Want

    A low standard deviation is usually the goal when you want consistency: product weights, delivery times, lab measurements. But low isn’t always better.

    • Tests meant to rank students. If nearly everyone scores between 78% and 82%, the test can’t tell strong and weak students apart. A test designed to separate them needs a reasonable spread of scores.
    • Surveys. A small standard deviation on a rating question shows people agree. A large one shows opinions are split, which can be the most useful finding in the survey.
    • Investments. In finance, the standard deviation of returns is a common measure of risk. A higher figure means bigger swings both ways, which some investors accept in exchange for a chance of higher returns.

    So the question isn’t only “is it small?”, but “is the amount of variation the right amount for this purpose?”

    When the Standard Deviation Is the Wrong Tool

    The standard deviation describes spread well for roughly symmetrical data. It can mislead in two situations:

    • Skewed data. Incomes, house prices and waiting times often have a long tail of high values. A few large values push the standard deviation up, so it overstates the spread for most people. The interquartile range, the spread of the middle 50%, is often a better summary here.
    • Data with outliers. One mistaken entry can multiply the standard deviation. Plot your data before you trust the number. The calculator’s dot plot makes stray values easy to spot.

    For a step-by-step explanation of how the standard deviation itself is calculated, including sample vs population and the 68–95–99.7 rule, see the Standard Deviation Calculator page. To find the mean, median and interquartile range of the same data, use the Mean, Median, Mode Calculator.