Standard deviation tells you how spread out a set of values is from the average. A small standard deviation means most values cluster tightly around the mean, while a large one means they’re scattered widely. But knowing that definition is only the starting point. The real skill is looking at a standard deviation result and understanding what it means for your data, your decisions, or the report you’re reading.
What a Low or High Value Actually Tells You
A standard deviation close to zero means the data points are nearly identical to one another. As the number climbs, it signals more variability. If a class of students scores an average of 80 on a test with a standard deviation of 5, most students scored between 75 and 85. If the standard deviation were 20, scores would be all over the place, from the 60s to the high 90s, and the average of 80 wouldn’t represent any individual student very well.
This matters because standard deviation is really a measure of how trustworthy the mean is as a summary. When standard deviation is low relative to the mean, you can point to the average and say “that’s roughly what most values look like.” When it’s high, the average may not describe anyone in the group particularly well.
The 68-95-99.7 Rule
For data that follows a bell-shaped (normal) distribution, there’s a reliable pattern. About 68% of values fall within one standard deviation of the mean. About 95% fall within two standard deviations. And about 99.7% fall within three. This is called the empirical rule, and it’s the single most useful tool for interpreting standard deviation results.
Say the average adult body temperature in a study is 98.2°F with a standard deviation of 0.7°F. You’d expect roughly 68% of people to fall between 97.5°F and 98.9°F (one SD in each direction). About 95% would fall between 96.8°F and 99.6°F. A reading of 100.5°F would be more than three standard deviations above the mean, placing it well outside where 99.7% of values land. That’s statistically unusual.
This pattern only works cleanly when the data is roughly symmetrical and bell-shaped. For skewed data, like household income or hospital length-of-stay, the percentages won’t match these benchmarks, and the standard deviation becomes harder to interpret on its own.
Turning SD Into a Z-Score
A z-score converts any individual data point into a number that tells you how many standard deviations it sits from the mean. The calculation is straightforward: subtract the mean from the individual value, then divide by the standard deviation. A z-score of 1.5 means the value is one and a half standard deviations above average. A z-score of negative 2 means it’s two standard deviations below.
Z-scores are useful because they let you compare values from completely different scales. A student who scores 720 on one exam and 28 on another can’t compare those raw numbers. But if the 720 represents a z-score of 1.2 and the 28 represents a z-score of 1.8, the student performed relatively better on the second exam compared to their peers.
How Medical Reference Ranges Use SD
When a lab report flags a result as “normal” or “abnormal,” standard deviation is often working behind the scenes. Many clinical reference ranges are built by testing a large group of healthy people, calculating the mean and standard deviation, then defining the normal range as the mean plus or minus two standard deviations. This captures the central 95% of healthy results, leaving 2.5% excluded on each end.
This has a practical implication that surprises many people: even if you’re perfectly healthy, there’s a 5% chance any single lab value will fall outside the reference range. Run 20 different blood tests on a healthy person, and on average one will come back flagged. The “abnormal” result isn’t necessarily a sign of disease. It’s a statistical boundary, not a diagnostic one.
Comparing Spread Across Different Datasets
Standard deviation is expressed in the same units as your data. If you’re measuring height in centimeters, the SD is in centimeters. If you’re measuring salary in dollars, the SD is in dollars. That makes it impossible to directly compare the spread of two variables measured in different units. Saying a salary SD of $12,000 is “bigger” than a height SD of 4 cm doesn’t tell you anything meaningful.
The coefficient of variation (CV) solves this. It divides the standard deviation by the mean and expresses the result as a percentage. If one dataset has a CV of 8% and another has a CV of 22%, the second is relatively more spread out, regardless of units. This is commonly used in fields like biology and manufacturing to compare variability across measurements that are on entirely different scales.
SD in Quality Control and Manufacturing
The Six Sigma methodology, widely used in manufacturing and business, is built entirely on standard deviation. The goal is to shrink process variability so that the acceptable tolerance limits sit six standard deviations away from the mean. A process operating at six sigma produces only 3.4 defects per million opportunities.
To put that in perspective, here’s how defect rates drop as you tighten variability:
- One sigma: 691,462 defects per million (essentially a coin flip on quality)
- Two sigma: 308,538 defects per million
- Three sigma: 66,807 defects per million
- Four sigma: 6,210 defects per million
- Six sigma: 3.4 defects per million
If someone describes a process as “three sigma,” they’re saying it produces about 67,000 errors in every million attempts. That framing makes standard deviation a decision-making tool: it quantifies how reliable a process is.
Standard Deviation vs. Standard Error
These two get confused constantly, and they answer different questions. Standard deviation describes how spread out individual data points are. Standard error describes how precise your estimate of the average is. If you measure the blood pressure of 100 people, the standard deviation tells you how much individual readings vary from person to person. The standard error tells you how confident you can be that your sample average is close to the true population average.
Standard error shrinks as your sample size grows, because larger samples give more reliable estimates of the mean. Standard deviation doesn’t necessarily shrink with more data, because adding more people to a study doesn’t make individuals less variable. When reading a research paper or report, check which one is being presented. A graph showing “mean ± SD” is illustrating the range of individual variation. A graph showing “mean ± SE” is illustrating the precision of the average, and the error bars will look deceptively small.
When Standard Deviation Misleads
Standard deviation assumes the data is reasonably symmetrical. When it’s not, the number can paint a misleading picture. Income data is a classic example: a small number of very high earners pulls the mean upward, and the standard deviation becomes inflated by those extreme values. In a dataset like that, neither the mean nor the standard deviation describes the typical person’s experience well.
Outliers create the same problem on a smaller scale. A single extreme value can inflate the standard deviation significantly, especially in small samples. If you’re working with data that has obvious extreme points or a lopsided shape, the median and interquartile range are often more informative measures of center and spread. Standard deviation works best when the data is roughly bell-shaped and free of extreme anomalies.
Population vs. Sample SD
You’ll sometimes encounter two slightly different versions of the standard deviation formula. The difference comes down to what you divide by. If your data includes every member of the group you care about (the entire population), you divide by the total number of data points. If your data is a sample drawn from a larger group, you divide by one fewer than the number of data points.
That “minus one” adjustment, called Bessel’s correction, compensates for the fact that a sample tends to slightly underestimate the true variability of the full population. In practice, for large samples the difference is negligible. For small samples (under 30 or so), it matters more. Unless you’ve measured literally 100% of the population you’re interested in, the sample version with the minus-one adjustment is the appropriate choice. Most spreadsheet software and statistical tools default to this version.
Reading SD in Published Reports
In scientific papers and professional reports, standard deviation is typically abbreviated as SD and reported alongside the mean. You’ll see formats like “M = 7.7, SD = 1.3” or “mean (SD) = 42.5 (6.8).” When you see this, you can immediately apply the empirical rule: roughly two-thirds of individuals in the study fell within one SD of that mean, and about 95% fell within two.
Pay attention to the ratio between the SD and the mean. If a study reports a mean response time of 450 milliseconds with an SD of 30, the data is tightly clustered and the average is a solid summary. If the SD is 200, nearly half the size of the mean, individual responses varied enormously, and the average alone doesn’t capture the full picture. That ratio, whether you calculate it formally as a coefficient of variation or just eyeball it, is often more informative than the raw SD number on its own.