A statistical measure is a number that summarizes or describes a characteristic of a dataset. It takes raw data, which on its own can be overwhelming, and distills it into a single value you can interpret and compare. Household income in a city, the spread of test scores in a classroom, the strength of a link between exercise and heart health: each of these can be captured by a specific statistical measure designed for the job.
Statistical measures fall into several functional categories, each answering a different question about your data: Where is the center? How spread out are the values? What shape does the distribution take? How strong is the relationship between two variables? Understanding these categories gives you a practical framework for choosing the right tool.
Measures of Central Tendency
Central tendency measures try to describe an entire dataset with one value that represents the middle. The three main ones are the mean, the median, and the mode.
The mean is the arithmetic average: add all the values and divide by how many there are. It’s the most widely used central measure because every data point contributes to the calculation. For data that follows a bell-shaped (normal) distribution, the mean is generally the best summary of the center. Its weakness is sensitivity to outliers. A single extreme value, like one salary of $10 million in a company where most people earn $60,000, can drag the mean far from what’s typical.
The median is the middle value when you line up all observations from smallest to largest. It is far less affected by outliers and skewed data, which is why you’ll see it used for things like home prices and income figures. When a dataset is perfectly symmetrical, the mean and median are the same number. As data becomes more lopsided, the median does a better job representing the center.
The mode is simply the most frequently occurring value. It’s most useful for categorical data, where averaging doesn’t make sense. If you survey people’s favorite ice cream flavor, the mode tells you the most popular choice. For numerical data, the mode is less informative because the most common value can sometimes sit far from the bulk of the data.
Measures of Spread
Knowing the center of your data is only half the picture. You also need to know how much the values vary. Two classrooms can have the same average test score, but in one class everyone scored between 78 and 82, while in the other scores ranged from 40 to 100. Measures of spread capture that difference.
The range is the simplest: it’s the gap between the largest and smallest values. It’s easy to calculate but heavily influenced by a single extreme observation.
The interquartile range (IQR) is more robust. It measures the spread of the middle 50% of your data by subtracting the value at the 25th percentile from the value at the 75th percentile. Because it ignores the tails, it handles outliers well, and it pairs naturally with the median when your data is skewed.
The standard deviation is the most commonly used measure of spread. It tells you, on average, how far individual data points fall from the mean. A small standard deviation means the values cluster tightly around the average; a large one means they’re scattered. In a normal distribution, about 68% of data falls within one standard deviation of the mean, about 95% within two, and about 99.7% within three. This pattern, known as the empirical rule, makes the standard deviation especially powerful for understanding how unusual a particular value is.
The variance is simply the standard deviation squared. It’s used heavily in statistical formulas behind the scenes, but because its units are squared (dollars squared, for instance), the standard deviation is usually easier to interpret in everyday terms.
Pairing the Right Measures Together
When data is normally distributed, the standard pairing is the mean with the standard deviation. When data is skewed or contains significant outliers, researchers switch to the median with the interquartile range. For categorical data, percentages are the go-to. Choosing the wrong combination can give a misleading summary, which is why recognizing the shape of your data matters.
Measures of Position
Sometimes you don’t just want to describe the dataset as a whole. You want to know where a specific value sits relative to everything else. That’s what measures of position do.
Percentiles divide a dataset into 100 equal groups. If your child is at the 90th percentile for height, that means they’re taller than about 90% of children in the reference group. Quartiles are a simplified version, splitting data into four groups of roughly 25% each. The first quartile (Q1) is the 25th percentile, the second quartile (Q2) is the median, and the third quartile (Q3) is the 75th percentile.
A z-score takes a different approach. It tells you how many standard deviations a particular observation sits above or below the mean. A z-score of 0 means the value equals the mean. A z-score of +2 means it’s two standard deviations above, putting it higher than roughly 95% of normally distributed data. Z-scores are especially useful for comparing values from different datasets that use different scales, like comparing a score on one exam to a score on a completely different exam.
Measures of Shape
Shape measures describe what a distribution looks like rather than where its center is or how spread out it is. The two main ones are skewness and kurtosis.
Skewness measures how symmetrical a distribution is. A skewness of zero means the data is evenly balanced around the center. Positive skewness means the right tail is stretched out (a few unusually high values), and negative skewness means the left tail is stretched out (a few unusually low values). Income data, for example, tends to be positively skewed because a small number of very high earners pull the right tail outward.
Kurtosis measures how heavy or light the tails of a distribution are compared to a normal distribution. High kurtosis means the dataset has more extreme values (heavy tails), while low kurtosis means extreme values are rare (light tails). This matters in fields like finance, where heavy tails signal a higher chance of extreme events.
Measures of Association
All the measures above describe a single variable. Measures of association quantify the relationship between two variables.
The correlation coefficient is the most common. It’s a single number between -1 and +1 that captures the strength and direction of a linear relationship. A value of +1 means a perfect positive relationship: as one variable goes up, the other goes up by a perfectly proportional amount. A value of -1 means a perfect inverse relationship. A value near 0 means no linear pattern exists.
Two main types cover most situations. The Pearson correlation works with continuous numerical data and assumes a roughly linear relationship. The Spearman correlation works with ranked data or when the relationship isn’t strictly linear, making it more flexible for real-world datasets that don’t follow tidy patterns. In both cases, the closer the coefficient is to +1 or -1, the stronger the association.
Measures Used in Inference
Descriptive measures summarize data you already have. Inferential measures help you draw conclusions about a larger population based on a sample. This distinction matters: a number calculated from a sample is called a statistic, while the corresponding true value for the entire population is called a parameter. The goal of most quantitative research is to estimate population parameters using sample statistics.
A p-value is one of the most widely cited inferential measures. It quantifies how compatible your observed data is with a specific hypothesis, typically the hypothesis that there’s no real effect or difference. A p-value of 0.03, for example, means that if there truly were no effect, you’d see results at least this extreme only about 3% of the time. The conventional threshold is 0.05, below which results are labeled “statistically significant,” though this cutoff is a convention, not a universal truth. A p-value does not tell you how large or important an effect is, only how surprising the data would be under the assumption of no effect.
A confidence interval gives you a range of plausible values for the true population parameter. A 95% confidence interval means that if you repeated the study many times using the same method, about 95% of the intervals you’d calculate would contain the true value. Unlike a p-value, a confidence interval communicates both the estimated size of an effect and the uncertainty around it, making it a more informative summary in most practical situations.
Choosing the Right Measure
The best statistical measure depends on two things: the type of data you have and the question you’re trying to answer. Categorical data (like favorite color or yes/no responses) calls for modes and percentages. Continuous data that follows a bell curve is well served by the mean and standard deviation. Continuous data that’s skewed or contains outliers is better described by the median and interquartile range.
If you want to understand one variable, central tendency and spread measures are your starting point. If you want to understand the relationship between two variables, correlation coefficients are the tool. And if you want to generalize findings from a sample to a broader population, inferential measures like p-values and confidence intervals become essential. Each type of statistical measure answers a specific question, and picking the right one is often more important than the calculation itself.