Grubbs' Test: Detecting a Single Outlier in Your Data

What Is an Outlier and Why Does It Matter?

An outlier is a data point that is unusually far from the rest of your observations. Maybe you are measuring the reaction times of 30 participants in a psychology experiment, and 29 of them respond in 200 to 400 milliseconds, but one participant takes 3,000 milliseconds. That extreme value is an outlier. Outliers matter because they can dramatically distort your results. A single extreme value can pull the mean far from where it would otherwise be, inflate the standard deviation, and make statistical tests unreliable. Before you analyze your data, it is important to check whether any values are so extreme that they might not belong with the rest.

What Is Grubbs' Test?

Grubbs' test, also known as the maximum normed residual test, is a formal statistical procedure for detecting a single outlier in a dataset. Rather than relying on a subjective judgment ("that number looks too big"), Grubbs' test gives you an objective, principled way to decide whether the most extreme value in your data is statistically unusual enough to be considered an outlier.

How the Test Works

The idea behind Grubbs' test is straightforward. First, you identify the data point that is furthest from the mean. Then you calculate the test statistic, which is simply the distance of that extreme point from the mean, divided by the standard deviation of the entire dataset. In other words, the test statistic tells you how many standard deviations the most extreme point lies from the average.

This test statistic is then compared to a critical value that depends on your sample size and your chosen significance level (commonly 0.05). If the test statistic exceeds the critical value, you conclude that the extreme point is a statistically significant outlier. If it does not exceed the critical value, you do not have sufficient evidence to call it an outlier — it could just be natural variation in your data.

A Concrete Example

Imagine a researcher studying the weekly study hours of university students. She surveys 20 students and gets values mostly between 5 and 25 hours, but one student reports 65 hours. To apply Grubbs' test, she calculates the mean of all 20 values, then computes how far 65 is from that mean in terms of standard deviations. Suppose the mean is 15 hours and the standard deviation is 8 hours. The test statistic would be (65 − 15) / 8 = 6.25. She then looks up the critical value for a sample of 20 at the 0.05 significance level. If the critical value is, say, 2.71, then her test statistic of 6.25 far exceeds it, and she can conclude that 65 hours is a statistically significant outlier.

Key Assumptions

Grubbs' test comes with an important assumption: it requires that the underlying data (without the suspected outlier) follow an approximately normal distribution. The normal distribution is the familiar bell-shaped curve where most values cluster around the middle and extreme values become increasingly rare. If your data are heavily skewed or follow a very different distribution, Grubbs' test may produce misleading results. It is a good idea to check a histogram or use a normality test before applying Grubbs' test.

The One-Outlier-at-a-Time Limitation

A critical limitation of Grubbs' test is that it is designed to detect only one outlier at a time. If your dataset contains two or more outliers, they can actually mask each other. Here is why: when multiple extreme values are present, they pull the mean and standard deviation in their direction, making each individual outlier appear less extreme than it really is. This phenomenon is called the masking effect. As a result, Grubbs' test might fail to flag any of the outliers, even though several are present.

If you suspect your data might contain more than one outlier, you should consider using a test that is specifically designed for that situation, such as the Generalized Extreme Studentized Deviate (ESD) test.

What to Do When You Find an Outlier

Finding a statistically significant outlier does not automatically mean you should delete it. The right course of action depends on why the outlier exists. If you can trace it to a clear error — a data entry mistake, a malfunctioning instrument, or a participant who misunderstood the instructions — then removing it is usually justified. But if the outlier is a genuine observation that simply happens to be extreme, removing it could bias your results. In that case, a better approach might be to run your analysis both with and without the outlier and report both sets of results. Many researchers also use robust statistical methods that are less sensitive to extreme values, which can sidestep the problem entirely.

Whatever you decide, always document your reasoning. Transparency about how you handled outliers is an important part of good research practice.